question
stringlengths
101
2.69k
chosen
stringlengths
43
2.27k
rejected
stringlengths
42
2.18k
Table 3 shows our knowledge analysis experiments' results. Llama-2-7b-chat performs similarly to Llama-2-7b on Hellaswag and MULU, while showing a slightly better performance on ARC and C-eval-tw. In the ARC and Hellaswag benchmarks, almost all continually pre-trained models outperform Llama-2-7b-chat. In the MULU benchmark, most continual pre-trained models tend to outperform Llama-2-7b-chat. However, in the case of the C-eval-tw benchmark, there is no clear pattern when comparing the efficacy of models utilizing simple methods for continual pre-training against Llama-2-7b-chat. It is worth noting that the observed differences mentioned above are subtle.
Table 3์€ ์šฐ๋ฆฌ์˜ ์ง€์‹ ๋ถ„์„ ์‹คํ—˜ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. Llama-2-7b-chat์€ Hellaswag๊ณผ MULU์—์„œ Llama-2-7b์™€ ์œ ์‚ฌํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด๋ฉฐ, ARC์™€ C-eval-tw์—์„œ๋Š” ์•ฝ๊ฐ„ ๋” ๋‚˜์€ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค. ARC์™€ Hellaswag ๋ฒค์น˜๋งˆํฌ์—์„œ๋Š” ๊ฑฐ์˜ ๋ชจ๋“  ์ง€์†์ ์œผ๋กœ ์‚ฌ์ „ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ๋“ค์ด Llama-2-7b-chat๋ณด๋‹ค ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค. MULU ๋ฒค์น˜๋งˆํฌ์—์„œ๋„ ๋Œ€๋ถ€๋ถ„์˜ ์ง€์†์ ์œผ๋กœ ์‚ฌ์ „ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ๋“ค์ด Llama-2-7b-chat๋ณด๋‹ค ์šฐ์ˆ˜ํ•œ ๊ฒฝํ–ฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ C-eval-tw ๋ฒค์น˜๋งˆํฌ์˜ ๊ฒฝ์šฐ, ์ง€์†์ ์ธ ์‚ฌ์ „ ํ›ˆ๋ จ์„ ์œ„ํ•œ ๊ฐ„๋‹จํ•œ ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜๋Š” ๋ชจ๋ธ๋“ค์˜ ํšจ์œจ์„ฑ์„ Llama-2-7b-chat๊ณผ ๋น„๊ตํ•  ๋•Œ ๋ช…ํ™•ํ•œ ํŒจํ„ด์ด ์—†์Šต๋‹ˆ๋‹ค. ์œ„์—์„œ ์–ธ๊ธ‰ํ•œ ์ฐจ์ด์ ๋“ค์€ ๋ฏธ๋ฌ˜ํ•˜๋‹ค๋Š” ์ ์„ ์ฃผ๋ชฉํ•  ๊ฐ€์น˜๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
ํ‘œ 3์€ ์ง€์‹ ๋ถ„์„ ์‹คํ—˜ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. Llama-2-7b-chat์€ ํ—ฌ๋ผ์Šค์›ฉ๊ณผ MULU์—์„œ Llama-2-7b์™€ ์œ ์‚ฌํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด๋ฉฐ, ARC์™€ C-eval-tw์—์„œ ์•ฝ๊ฐ„ ๋” ๋‚˜์€ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค. ARC ๋ฐ Hellaswag ๋ฒค์น˜๋งˆํฌ์—์„œ๋Š” ๊ฑฐ์˜ ๋ชจ๋“  ์—ฐ์† ์‚ฌ์ „ ํ•™์Šต ๋ชจ๋ธ์ด Llama-2-7b-chat๋ณด๋‹ค ์„ฑ๋Šฅ์ด ์šฐ์ˆ˜ํ•ฉ๋‹ˆ๋‹ค. MULU ๋ฒค์น˜๋งˆํฌ์—์„œ๋Š” ๋Œ€๋ถ€๋ถ„์˜ ์—ฐ์† ์‚ฌ์ „ ํ•™์Šต ๋ชจ๋ธ์ด Llama-2-7b-chat๋ณด๋‹ค ์„ฑ๋Šฅ์ด ์šฐ์ˆ˜ํ•œ ๊ฒฝํ–ฅ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ C-eval-tw ๋ฒค์น˜๋งˆํฌ์˜ ๊ฒฝ์šฐ, ์—ฐ์† ์‚ฌ์ „ ํ•™์Šต์„ ์œ„ํ•œ ๊ฐ„๋‹จํ•œ ๋ฐฉ๋ฒ•์„ ํ™œ์šฉํ•œ ๋ชจ๋ธ์˜ ํšจ๊ณผ๋ฅผ Llama-2-7b-chat์™€ ๋น„๊ตํ•  ๋•Œ ๋ช…ํ™•ํ•œ ํŒจํ„ด์ด ๋‚˜ํƒ€๋‚˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์œ„์—์„œ ์–ธ๊ธ‰ํ•œ ๊ด€์ฐฐ๋œ ์ฐจ์ด์ ์€ ๋ฏธ๋ฏธํ•˜๋‹ค๋Š” ์ ์— ์œ ์˜ํ•  ํ•„์š”๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
Downstream EvaluationFollowing much previous work (Brown et al., 2020; Black et al., 2022; Touvron et al., 2023, _inter alia_), we report zero-shot performance on a set of downstream tasks. Our evaluation suite consists of 8 core tasks corresponding closely to the commonsense reasoning task set reported by Touvron et al. (2023) and Touvron et al. (2023) (see Table 6 for a list of tasks). Given the scale of the models being evaluated, such tasks were selected at the beginning of model development due to their naturalness (e.g., all can formulated as text completion scoring tasks) and ability to provide meaningful signals throughout training (see Figure 1).
ํ›„์† ํ‰๊ฐ€: ๋งŽ์€ ์ด์ „ ์—ฐ๊ตฌ(Brown et al., 2020; Black et al., 2022; Touvron et al., 2023, _inter alia_)๋ฅผ ๋”ฐ๋ฅด๋ฉฐ, ์šฐ๋ฆฌ๋Š” ์ผ๋ จ์˜ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์ž‘์—…์— ๋Œ€ํ•œ ์ œ๋กœ์ƒท ์„ฑ๋Šฅ์„ ๋ณด๊ณ ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ํ‰๊ฐ€ ๋ชจ์Œ์€ Touvron et al. (2023)๊ณผ Touvron et al. (2023)์ด ๋ณด๊ณ ํ•œ ์ƒ์‹ ์ถ”๋ก  ์ž‘์—… ์„ธํŠธ์™€ ๋ฐ€์ ‘ํ•˜๊ฒŒ ๊ด€๋ จ๋œ 8๊ฐœ์˜ ํ•ต์‹ฌ ์ž‘์—…์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค(์ž‘์—… ๋ชฉ๋ก์€ ํ‘œ 6 ์ฐธ์กฐ). ํ‰๊ฐ€๋˜๋Š” ๋ชจ๋ธ์˜ ๊ทœ๋ชจ๋ฅผ ๊ฐ์•ˆํ•  ๋•Œ, ์ด๋Ÿฌํ•œ ์ž‘์—…์€ ์ž์—ฐ์Šค๋Ÿฌ์›€(์˜ˆ: ๋ชจ๋‘ ํ…์ŠคํŠธ ์™„์„ฑ ์ ์ˆ˜ ์ž‘์—…์œผ๋กœ ๊ตฌ์„ฑ๋  ์ˆ˜ ์žˆ์Œ)๊ณผ ํ›ˆ๋ จ ์ „๋ฐ˜์— ๊ฑธ์ณ ์˜๋ฏธ ์žˆ๋Š” ์‹ ํ˜ธ๋ฅผ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ ๋•Œ๋ฌธ์— ๋ชจ๋ธ ๊ฐœ๋ฐœ ์ดˆ๊ธฐ์— ์„ ํƒ๋˜์—ˆ์Šต๋‹ˆ๋‹ค(๊ทธ๋ฆผ 1 ์ฐธ์กฐ).
๋‹ค์šด์ŠคํŠธ๋ฆผ ํ‰๊ฐ€๋งŽ์€ ์„ ํ–‰ ์—ฐ๊ตฌ(Brown et al., 2020; Black et al., 2022; Touvron et al., 2023, _inter alia_)์— ๋”ฐ๋ผ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์ž‘์—… ์„ธํŠธ์— ๋Œ€ํ•œ ๋ฌด์„  ์ƒท ์„ฑ๋Šฅ์„ ๋ณด๊ณ ํ•ฉ๋‹ˆ๋‹ค. ํ‰๊ฐ€ ์ œํ’ˆ๊ตฐ์€ Touvron et al. (2023) ๋ฐ Touvron et al. (2023)์— ๋ณด๊ณ ๋œ ์ƒ์‹ ์ถ”๋ก  ์ž‘์—… ์„ธํŠธ์— ๋ฐ€์ ‘ํ•˜๊ฒŒ ํ•ด๋‹นํ•˜๋Š” 8๊ฐœ์˜ ํ•ต์‹ฌ ์ž‘์—…์œผ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค(์ž‘์—… ๋ชฉ๋ก์€ ํ‘œ 6 ์ฐธ์กฐ). ํ‰๊ฐ€ ๋Œ€์ƒ ๋ชจ๋ธ์˜ ๊ทœ๋ชจ๋ฅผ ๊ณ ๋ คํ•˜์—ฌ ๋ชจ๋ธ ๊ฐœ๋ฐœ ์ดˆ๊ธฐ์— ์ž์—ฐ์Šค๋Ÿฌ์›€(์˜ˆ: ๋ชจ๋‘ ํ…์ŠคํŠธ ์™„์„ฑ ์ ์ˆ˜ ์ž‘์—…์œผ๋กœ ๊ณต์‹ํ™”ํ•  ์ˆ˜ ์žˆ์Œ)๊ณผ ํ›ˆ๋ จ ์ „๋ฐ˜์— ๊ฑธ์ณ ์˜๋ฏธ ์žˆ๋Š” ์‹ ํ˜ธ๋ฅผ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ๋Š” ์ž‘์—…์„ ์„ ์ •ํ–ˆ์Šต๋‹ˆ๋‹ค(๊ทธ๋ฆผ 1 ์ฐธ์กฐ).
* User's questions can refer to his/her previous (or multiple turns before) question. * User's questions can also refer to the previous (or multiple turns before) answer from the agent. * Try to use pronouns or common nouns to replace the entities mentioned before. * Try to make the user's questions diverse. For the same type of questions, try to use different ways to express it in different dialogue annotations. * Other user question types we need * Given the answer from agent, ask for extra information (e.g., What else...; any other...; is there more...). * Switch the topic and starting a new thread in conversation. * Ask two questions at the same time.
* ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์€ ์ด์ „(๋˜๋Š” ์—ฌ๋Ÿฌ ์ฐจ๋ก€ ์ „)์˜ ์งˆ๋ฌธ์„ ์ฐธ์กฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. * ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์€ ์—์ด์ „ํŠธ์˜ ์ด์ „(๋˜๋Š” ์—ฌ๋Ÿฌ ์ฐจ๋ก€ ์ „)์˜ ๋‹ต๋ณ€์„ ์ฐธ์กฐํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. * ์ด์ „์— ์–ธ๊ธ‰๋œ ์—”ํ‹ฐํ‹ฐ๋ฅผ ๋Œ€๋ช…์‚ฌ๋‚˜ ์ผ๋ฐ˜ ๋ช…์‚ฌ๋กœ ๋Œ€์ฒดํ•˜๋ ค๊ณ  ๋…ธ๋ ฅํ•˜์„ธ์š”. * ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์„ ๋‹ค์–‘ํ•˜๊ฒŒ ๋งŒ๋“œ์„ธ์š”. ๋™์ผํ•œ ์œ ํ˜•์˜ ์งˆ๋ฌธ์— ๋Œ€ํ•ด, ๋‹ค๋ฅธ ๋Œ€ํ™” ์ฃผ์„์—์„œ ๋‹ค๋ฅธ ๋ฐฉ์‹์œผ๋กœ ํ‘œํ˜„ํ•˜๋ ค๊ณ  ๋…ธ๋ ฅํ•˜์„ธ์š”. * ์šฐ๋ฆฌ๊ฐ€ ํ•„์š”๋กœ ํ•˜๋Š” ๋‹ค๋ฅธ ์‚ฌ์šฉ์ž ์งˆ๋ฌธ ์œ ํ˜•: * ์—์ด์ „ํŠธ์˜ ๋‹ต๋ณ€์„ ๋ฐ”ํƒ•์œผ๋กœ ์ถ”๊ฐ€ ์ •๋ณด๋ฅผ ์š”์ฒญํ•˜์„ธ์š” (์˜ˆ: "๋‹ค๋ฅธ ๊ฒƒ์€..."; "๋‹ค๋ฅธ ๊ฒƒ์ด ์žˆ๋‚˜์š”..."; "๋” ์žˆ๋‚˜์š”..."). * ์ฃผ์ œ๋ฅผ ์ „ํ™˜ํ•˜๊ณ  ๋Œ€ํ™”์—์„œ ์ƒˆ๋กœ์šด ์ฃผ์ œ๋ฅผ ์‹œ์ž‘ํ•˜์„ธ์š”. * ๋™์‹œ์— ๋‘ ๊ฐ€์ง€ ์งˆ๋ฌธ์„ ํ•˜์„ธ์š”.
* ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์€ ์ด์ „ ์งˆ๋ฌธ(๋˜๋Š” ๋ช‡ ๋ฒˆ ์ „)์„ ์–ธ๊ธ‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. * ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์€ ์ƒ๋‹ด์›์˜ ์ด์ „(๋˜๋Š” ๋ช‡ ๋ฒˆ ์ „) ๋‹ต๋ณ€์„ ์–ธ๊ธ‰ํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. * ์•ž์„œ ์–ธ๊ธ‰ํ•œ ๊ฐœ์ฒด๋ฅผ ๋Œ€์‹ ํ•˜๊ธฐ ์œ„ํ•ด ๋Œ€๋ช…์‚ฌ ๋˜๋Š” ์ผ๋ฐ˜ ๋ช…์‚ฌ๋ฅผ ์‚ฌ์šฉํ•˜์„ธ์š”. * ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์„ ๋‹ค์–‘ํ•˜๊ฒŒ ํ•˜์„ธ์š”. ๊ฐ™์€ ์œ ํ˜•์˜ ์งˆ๋ฌธ์ธ ๊ฒฝ์šฐ, ๋‹ค๋ฅธ ๋Œ€ํ™” ์ฃผ์„์—์„œ ๋‹ค๋ฅธ ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ ํ‘œํ˜„ํ•˜์„ธ์š”. * ์šฐ๋ฆฌ๊ฐ€ ํ•„์š”๋กœ ํ•˜๋Š” ๋‹ค๋ฅธ ์‚ฌ์šฉ์ž ์งˆ๋ฌธ ์œ ํ˜•* ์ƒ๋‹ด์›์˜ ๋‹ต๋ณ€์ด ์ฃผ์–ด์ง€๋ฉด ์ถ”๊ฐ€ ์ •๋ณด๋ฅผ ์š”์ฒญํ•ฉ๋‹ˆ๋‹ค(์˜ˆ: ๋‹ค๋ฅธ ๊ฒƒ...; ๋‹ค๋ฅธ ๊ฒƒ...; ๋” ์žˆ๋Š”์ง€...). * ๋Œ€ํ™”์—์„œ ์ฃผ์ œ๋ฅผ ์ „ํ™˜ํ•˜๊ณ  ์ƒˆ ์Šค๋ ˆ๋“œ๋ฅผ ์‹œ์ž‘ํ•ฉ๋‹ˆ๋‹ค. * ๋™์‹œ์— ๋‘ ๊ฐœ์˜ ์งˆ๋ฌธ์„ ํ•ฉ๋‹ˆ๋‹ค.
In our exploration of model merging in the parameter space, we experimented with diverse configurations, such as varying the assignment of merging parameters across different layer groups. However, due to a constrained dataset, we did not witness notable improvements in performance correlating with increase in configuration complexity. Consequently, we focused our reporting on a PS merged model (Model 4 in Table 1) that adopts the simplest setting: considering each source model as a singular layer and allocating two DARE-TIES associated parameters to each for evolutionary merging. Figure 3 illustrates the evolved parameter configuration post PS merging.
๋ชจ๋ธ ๋ณ‘ํ•ฉ์— ๋Œ€ํ•œ ํŒŒ๋ผ๋ฏธํ„ฐ ๊ณต๊ฐ„ ํƒ์ƒ‰์—์„œ, ์šฐ๋ฆฌ๋Š” ๋‹ค์–‘ํ•œ ๋ ˆ์ด์–ด ๊ทธ๋ฃน ๊ฐ„์˜ ๋ณ‘ํ•ฉ ํŒŒ๋ผ๋ฏธํ„ฐ ํ• ๋‹น์„ ๋ณ€ํ™”์‹œํ‚ค๋Š” ๋“ฑ ๋‹ค์–‘ํ•œ ๊ตฌ์„ฑ์œผ๋กœ ์‹คํ—˜์„ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ œํ•œ๋œ ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ์ธํ•ด ๊ตฌ์„ฑ ๋ณต์žก์„ฑ ์ฆ๊ฐ€์™€ ์„ฑ๋Šฅ ํ–ฅ์ƒ ๊ฐ„์˜ ๋šœ๋ ทํ•œ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๊ด€์ฐฐํ•˜์ง€ ๋ชปํ–ˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์šฐ๋ฆฌ๋Š” ๊ฐ€์žฅ ๋‹จ์ˆœํ•œ ์„ค์ •์„ ์ฑ„ํƒํ•œ PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ(Table 1์˜ Model 4)์— ๋Œ€ํ•œ ๋ณด๊ณ ์— ์ง‘์ค‘ํ–ˆ์Šต๋‹ˆ๋‹ค: ๊ฐ ์†Œ์Šค ๋ชจ๋ธ์„ ๋‹จ์ผ ๋ ˆ์ด์–ด๋กœ ๊ฐ„์ฃผํ•˜๊ณ , ์ง„ํ™”์  ๋ณ‘ํ•ฉ์„ ์œ„ํ•ด ๊ฐ ๋ชจ๋ธ์— ๋‘ ๊ฐœ์˜ DARE-TIES ๊ด€๋ จ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ํ• ๋‹นํ–ˆ์Šต๋‹ˆ๋‹ค. Figure 3์€ PS ๋ณ‘ํ•ฉ ํ›„ ์ง„ํ™”๋œ ํŒŒ๋ผ๋ฏธํ„ฐ ๊ตฌ์„ฑ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
๋งค๊ฐœ๋ณ€์ˆ˜ ๊ณต๊ฐ„์—์„œ ๋ชจ๋ธ ๋ณ‘ํ•ฉ์„ ํƒ์ƒ‰ํ•˜๋ฉด์„œ ๋‹ค์–‘ํ•œ ๋ ˆ์ด์–ด ๊ทธ๋ฃน์— ๋ณ‘ํ•ฉ ๋งค๊ฐœ๋ณ€์ˆ˜ ํ• ๋‹น์„ ๋ณ€๊ฒฝํ•˜๋Š” ๋“ฑ ๋‹ค์–‘ํ•œ ๊ตฌ์„ฑ์„ ์‹คํ—˜ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ œํ•œ๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋กœ ์ธํ•ด ๊ตฌ์„ฑ ๋ณต์žก์„ฑ์˜ ์ฆ๊ฐ€์™€ ๊ด€๋ จ๋œ ์„ฑ๋Šฅ ํ–ฅ์ƒ์ด ๋ˆˆ์— ๋„์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๊ฐ€์žฅ ๊ฐ„๋‹จํ•œ ์„ค์ •์„ ์ฑ„ํƒํ•˜๋Š” PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ(ํ‘œ 1์˜ ๋ชจ๋ธ 4)์— ์ง‘์ค‘ํ•˜์—ฌ ๋ณด๊ณ ํ–ˆ์Šต๋‹ˆ๋‹ค: ๊ฐ ์†Œ์Šค ๋ชจ๋ธ์„ ๋‹จ์ผ ๋ ˆ์ด์–ด๋กœ ๊ฐ„์ฃผํ•˜๊ณ  ์ง„ํ™” ๋ณ‘ํ•ฉ์„ ์œ„ํ•ด ๊ฐ ๋ชจ๋ธ์— ๋‘ ๊ฐœ์˜ DARE-TIES ๊ด€๋ จ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ํ• ๋‹นํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ฆผ 3์€ PS ๋ณ‘ํ•ฉ ํ›„ ์ง„ํ™”๋œ ํŒŒ๋ผ๋ฏธํ„ฐ ๊ตฌ์„ฑ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
Moreover, we introduce MiniCPM family, including MiniCPM-DPO, MiniCPM-128K, and MiniCPM-MoE. We conduct evaluations of MiniCPM family against established benchmarks and illuminate their impressive capabilities as SLMs: (1) The foundation models surpasses Mistral-7B, LLama-13B. (2) The dpo model surpasses zephyr-7B (Tunstall et al., 2023) on MTBench (Zheng et al., 2024) (3) The 2.4B MiniCPM-128K model demonstrates performance either surpassing or matching that of models like Yarn-Mistral-7B-128K (Peng et al., 2023) and ChatGLM3-6B-128K (Du et al., 2021). (4) The MiniCPM-MoE, with 4B activated parameters, is on par with Llama2-34B (Touvron et al., 2023).
๊ฒŒ๋‹ค๊ฐ€, ์šฐ๋ฆฌ๋Š” MiniCPM-DPO, MiniCPM-128K, MiniCPM-MoE๋ฅผ ํฌํ•จํ•œ MiniCPM ํŒจ๋ฐ€๋ฆฌ๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ํ™•๋ฆฝ๋œ ๋ฒค์น˜๋งˆํฌ์— ๋Œ€ํ•ด MiniCPM ํŒจ๋ฐ€๋ฆฌ๋ฅผ ํ‰๊ฐ€ํ•˜๊ณ , ์ด๋“ค์˜ SLM์œผ๋กœ์„œ์˜ ์ธ์ƒ์ ์ธ ๋Šฅ๋ ฅ์„ ์กฐ๋ช…ํ•ฉ๋‹ˆ๋‹ค: (1) ๊ธฐ๋ณธ ๋ชจ๋ธ์€ Mistral-7B, LLama-13B๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. (2) dpo ๋ชจ๋ธ์€ MTBench (Zheng et al., 2024)์—์„œ zephyr-7B (Tunstall et al., 2023)๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. (3) 2.4B MiniCPM-128K ๋ชจ๋ธ์€ Yarn-Mistral-7B-128K (Peng et al., 2023) ๋ฐ ChatGLM3-6B-128K (Du et al., 2021)์™€ ๊ฐ™์€ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ๋Šฅ๊ฐ€ํ•˜๊ฑฐ๋‚˜ ์ผ์น˜ํ•˜๋Š” ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. (4) 4B ํ™œ์„ฑํ™” ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๊ฐ€์ง„ MiniCPM-MoE๋Š” Llama2-34B (Touvron et al., 2023)์™€ ๋™๋“ฑํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค.
๋˜ํ•œ MiniCPM-DPO, MiniCPM-128K, MiniCPM-MoE๋ฅผ ํฌํ•จํ•œ MiniCPM ์ œํ’ˆ๊ตฐ์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ์กด ๋ฒค์น˜๋งˆํฌ์— ๋Œ€ํ•œ MiniCPM ์ œํ’ˆ๊ตฐ์˜ ํ‰๊ฐ€๋ฅผ ์ˆ˜ํ–‰ํ•˜๊ณ , ์ด ์ œํ’ˆ๊ตฐ์ด SLM์œผ๋กœ์„œ์˜ ์ธ์ƒ์ ์ธ ๊ธฐ๋Šฅ์„ ์กฐ๋ช…ํ•ฉ๋‹ˆ๋‹ค: (1) ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์€ Mistral-7B, LLama-13B๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. (2) dpo ๋ชจ๋ธ์€ MTBench(Zheng et al., 2024)์—์„œ zephyr-7B(Tunstall et al., 2023)๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. (3) 2.4B MiniCPM-128K ๋ชจ๋ธ์€ Yarn-Mistral-7B-128K(Peng et al., 2023) ๋ฐ ChatGLM3-6B-128K(Du et al., 2021)์™€ ๊ฐ™์€ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ๋Šฅ๊ฐ€ํ•˜๊ฑฐ๋‚˜ ์ด์— ํ•„์ ํ•˜๋Š” ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. (4) ํ™œ์„ฑํ™”๋œ ํŒŒ๋ผ๋ฏธํ„ฐ๊ฐ€ 4B์ธ MiniCPM-MoE๋Š” Llama2-34B(Touvron et al., 2023)์™€ ๋™๋“ฑํ•œ ์ˆ˜์ค€์ž…๋‹ˆ๋‹ค.
Figure 6: Comparative performance trends of different language models over various training checkpoints, as fitted by the Eq.(1). (a) We present the CEval Score Trends, showing a consistent increase in performance with larger training data sizes. (b) We depict the CMMLU Score Trends, with a similar positive correlation between score and training size. (c) We illustrate the MMLU Score Trends, with all models demonstrating performance enhancements correlating with the scale of training data. Each graph plots individual data points for different versions of the models and overlays a curve fit to represent the scaling lawโ€™s predictive performance scaling.
Figure 6: ๋‹ค์–‘ํ•œ ํ›ˆ๋ จ ์ฒดํฌํฌ์ธํŠธ์—์„œ ์„œ๋กœ ๋‹ค๋ฅธ ์–ธ์–ด ๋ชจ๋ธ์˜ ๋น„๊ต ์„ฑ๋Šฅ ์ถ”์„ธ๋ฅผ Eq.(1)์— ๋งž์ถฐ ๋‚˜ํƒ€๋‚ธ ๊ฒƒ์ž…๋‹ˆ๋‹ค. (a) CEval ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ์ œ์‹œํ•˜๋ฉฐ, ๋” ํฐ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ํฌ๊ธฐ์™€ ํ•จ๊ป˜ ์„ฑ๋Šฅ์ด ์ผ๊ด€๋˜๊ฒŒ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. (b) CMMLU ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ๋ฌ˜์‚ฌํ•˜๋ฉฐ, ์ ์ˆ˜์™€ ํ›ˆ๋ จ ํฌ๊ธฐ ์‚ฌ์ด์˜ ์œ ์‚ฌํ•œ ๊ธ์ •์  ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. (c) MMLU ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ์„ค๋ช…ํ•˜๋ฉฐ, ๋ชจ๋“  ๋ชจ๋ธ์ด ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์˜ ๊ทœ๋ชจ์™€ ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ์žˆ๋Š” ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ฐ ๊ทธ๋ž˜ํ”„๋Š” ๋ชจ๋ธ์˜ ๋‹ค์–‘ํ•œ ๋ฒ„์ „์— ๋Œ€ํ•œ ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋ฅผ ํ”Œ๋กœํŒ…ํ•˜๊ณ , ์Šค์ผ€์ผ๋ง ๋ฒ•์น™์˜ ์˜ˆ์ธก ์„ฑ๋Šฅ ์Šค์ผ€์ผ๋ง์„ ๋‚˜ํƒ€๋‚ด๊ธฐ ์œ„ํ•ด ๊ณก์„  ํ”ผํŒ…์„ ์˜ค๋ฒ„๋ ˆ์ดํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 6: ๋‹ค์–‘ํ•œ ํ›ˆ๋ จ ์ฒดํฌํฌ์ธํŠธ์— ๋Œ€ํ•œ ๋‹ค์–‘ํ•œ ์–ธ์–ด ๋ชจ๋ธ์˜ ๋น„๊ต ์„ฑ๋Šฅ ์ถ”์„ธ๋ฅผ Eq.(1)์— ๋งž์ถฐ ํ‘œ์‹œํ•ฉ๋‹ˆ๋‹ค. (a) ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ํฌ๊ธฐ๊ฐ€ ํด์ˆ˜๋ก ์„ฑ๋Šฅ์ด ์ง€์†์ ์œผ๋กœ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ฃผ๋Š” CEval ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. (b) ์ ์ˆ˜์™€ ํ›ˆ๋ จ ๊ทœ๋ชจ ์‚ฌ์ด์˜ ์œ ์‚ฌํ•œ ์–‘์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋ณด์—ฌ์ฃผ๋Š” CMMLU ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. (c) ๋ชจ๋“  ๋ชจ๋ธ์ด ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์˜ ๊ทœ๋ชจ์™€ ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ์žˆ๋Š” ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์—ฌ์ฃผ๋Š” MMLU ์ ์ˆ˜ ์ถ”์„ธ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ฐ ๊ทธ๋ž˜ํ”„๋Š” ๋ชจ๋ธ์˜ ์—ฌ๋Ÿฌ ๋ฒ„์ „์— ๋Œ€ํ•œ ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋ฅผ ํ”Œ๋กฏํ•˜๊ณ  ์Šค์ผ€์ผ๋ง ๋ฒ•์น™์˜ ์˜ˆ์ธก ์„ฑ๋Šฅ ์Šค์ผ€์ผ๋ง์„ ๋‚˜ํƒ€๋‚ด๋Š” ๊ณก์„  ํ”ผํŠธ๋ฅผ ์˜ค๋ฒ„๋ ˆ์ดํ•ฉ๋‹ˆ๋‹ค.
We mitigate challenges associated with document complexity by hiring workers who have already read a book published in 2023 or 2024 (to avoid data contamination) for enjoyment prior to beginning the annotation task. We produce summaries for these books via fiveconfigurations of the hierarchical summarization methodology described in Chang et al. (2023b), each of which varies the base LLM and chunk size. Following prior work on faithfulness and factuality evaluation, such as LongEval (Krishna et al., 2023) and FactScore (Min et al., 2023), we decompose each summary into a list of claims which are then individually verified against the input document.
๋ฌธ์„œ์˜ ๋ณต์žก์„ฑ๊ณผ ๊ด€๋ จ๋œ ๋ฌธ์ œ๋ฅผ ์™„ํ™”ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ์ฃผ์„ ์ž‘์—…์„ ์‹œ์ž‘ํ•˜๊ธฐ ์ „์— 2023๋…„ ๋˜๋Š” 2024๋…„์— ์ถœํŒ๋œ ์ฑ…์„ ์ด๋ฏธ ์ฝ์€ ์ž‘์—…์ž๋“ค์„ ๊ณ ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ ์˜ค์—ผ์„ ํ”ผํ•˜๊ณ ์ž ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ฑ…๋“ค์— ๋Œ€ํ•ด Chang et al. (2023b)์—์„œ ์„ค๋ช…ํ•œ ๊ณ„์ธต์  ์š”์•ฝ ๋ฐฉ๋ฒ•๋ก ์˜ ๋‹ค์„ฏ ๊ฐ€์ง€ ๊ตฌ์„ฑ ๋ฐฉ์‹์„ ํ†ตํ•ด ์š”์•ฝ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ๊ตฌ์„ฑ ๋ฐฉ์‹์€ ๊ธฐ๋ณธ LLM๊ณผ ์ฒญํฌ ํฌ๊ธฐ๋ฅผ ๋‹ค๋ฅด๊ฒŒ ์„ค์ •ํ•ฉ๋‹ˆ๋‹ค. LongEval (Krishna et al., 2023) ๋ฐ FactScore (Min et al., 2023)์™€ ๊ฐ™์€ ์‹ ๋ขฐ์„ฑ๊ณผ ์‚ฌ์‹ค์„ฑ ํ‰๊ฐ€์— ๊ด€ํ•œ ์ด์ „ ์—ฐ๊ตฌ๋ฅผ ๋”ฐ๋ฅด๋ฉฐ, ๊ฐ ์š”์•ฝ์„ ์ฃผ์žฅ ๋ชฉ๋ก์œผ๋กœ ๋ถ„ํ•ดํ•œ ํ›„ ์ž…๋ ฅ ๋ฌธ์„œ์™€ ๊ฐœ๋ณ„์ ์œผ๋กœ ๋Œ€์กฐํ•˜์—ฌ ๊ฒ€์ฆํ•ฉ๋‹ˆ๋‹ค.
๋ฌธ์„œ์˜ ๋ณต์žก์„ฑ๊ณผ ๊ด€๋ จ๋œ ๋ฌธ์ œ๋ฅผ ์™„ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ์ฃผ์„ ์ž‘์—…์„ ์‹œ์ž‘ํ•˜๊ธฐ ์ „์— ์ด๋ฏธ 2023๋…„ ๋˜๋Š” 2024๋…„์— ์ถœํŒ๋œ ์ฑ…์„ ์žฌ๋ฏธ๋กœ ์ฝ์€ ์ ์ด ์žˆ๋Š” ์ž‘์—…์ž๋ฅผ ๊ณ ์šฉํ•ฉ๋‹ˆ๋‹ค(๋ฐ์ดํ„ฐ ์˜ค์—ผ์„ ํ”ผํ•˜๊ธฐ ์œ„ํ•ด). ์ด๋Ÿฌํ•œ ์ฑ…์— ๋Œ€ํ•œ ์š”์•ฝ์€ Chang et al. (2023b)์— ์„ค๋ช…๋œ ๊ณ„์ธต์  ์š”์•ฝ ๋ฐฉ๋ฒ•๋ก ์˜ ๋‹ค์„ฏ ๊ฐ€์ง€ ๊ตฌ์„ฑ์„ ํ†ตํ•ด ์ƒ์„ฑํ•˜๋ฉฐ, ๊ฐ ๊ตฌ์„ฑ์€ ๊ธฐ๋ณธ LLM๊ณผ ์ฒญํฌ ํฌ๊ธฐ๋ฅผ ๋ณ€๊ฒฝํ•ฉ๋‹ˆ๋‹ค. ์ถฉ์‹ค๋„ ๋ฐ ์ง„์‹ค์„ฑ ํ‰๊ฐ€์— ๊ด€ํ•œ ์„ ํ–‰ ์—ฐ๊ตฌ(LongEval(Krishna et al., 2023) ๋ฐ FactScore(Min et al., 2023))์— ๋”ฐ๋ผ ๊ฐ ์š”์•ฝ์„ ํด๋ ˆ์ž„ ๋ชฉ๋ก์œผ๋กœ ๋ถ„ํ•ดํ•œ ๋‹ค์Œ ์ž…๋ ฅ ๋ฌธ์„œ์™€ ๋น„๊ตํ•˜์—ฌ ๊ฐœ๋ณ„์ ์œผ๋กœ ๊ฒ€์ฆํ•ฉ๋‹ˆ๋‹ค.
\(\bullet\)_Decomposing into easy, detailed sub-tasks._ To solve complex tasks, it is important to decompose the difficult task into several more easier, detailed sub-tasks for helping LLMs accomplish the goal step by step, which is closely related to the planning technique in Section 6.4. For example, following the suggestion [454], we can explicitly list the sub-tasks in the form of multiple numbered items (_e.g._, _"Braid a coherent narrative by performing the following tasks: 1...., 2...., 3...."_). By decomposing a target task into sub-tasks, LLMs can focus on solving easier sub-tasks and finally achieve more accurate results for complex tasks.
โ€ข_์‰ฌ์šด ์„ธ๋ถ€ ํ•˜์œ„ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ดํ•˜๊ธฐ._ ๋ณต์žกํ•œ ์ž‘์—…์„ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์–ด๋ ค์šด ์ž‘์—…์„ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋” ์‰ฌ์šด ์„ธ๋ถ€ ํ•˜์œ„ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ดํ•˜์—ฌ LLM์ด ๋ชฉํ‘œ๋ฅผ ๋‹จ๊ณ„๋ณ„๋กœ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ๋•๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์„น์…˜ 6.4์˜ ๊ณ„ํš ๊ธฐ๋ฒ•๊ณผ ๋ฐ€์ ‘ํ•œ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ œ์•ˆ [454]์— ๋”ฐ๋ผ, ์—ฌ๋Ÿฌ ๋ฒˆํ˜ธ๊ฐ€ ๋งค๊ฒจ์ง„ ํ•ญ๋ชฉ์˜ ํ˜•ํƒœ๋กœ ํ•˜์œ„ ์ž‘์—…์„ ๋ช…์‹œ์ ์œผ๋กœ ๋‚˜์—ดํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค (_์˜ˆ: _"๋‹ค์Œ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜์—ฌ ์ผ๊ด€๋œ ์„œ์‚ฌ๋ฅผ ์—ฎ์œผ์„ธ์š”: 1...., 2...., 3...."_). ๋ชฉํ‘œ ์ž‘์—…์„ ํ•˜์œ„ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ดํ•จ์œผ๋กœ์จ, LLM์€ ๋” ์‰ฌ์šด ํ•˜์œ„ ์ž‘์—…์„ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ๊ฒฐ๊ตญ ๋ณต์žกํ•œ ์ž‘์—…์— ๋Œ€ํ•ด ๋” ์ •ํ™•ํ•œ ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
\(\bullet\)_์‰ฌ์šด ์„ธ๋ถ€ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ด._ ๋ณต์žกํ•œ ์ž‘์—…์„ ํ•ด๊ฒฐํ•˜๋ ค๋ฉด ์–ด๋ ค์šด ์ž‘์—…์„ ๋ช‡ ๊ฐ€์ง€ ๋” ์‰ฌ์šด ์„ธ๋ถ€ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ดํ•˜์—ฌ LLM์ด ๋‹จ๊ณ„๋ณ„๋กœ ๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•˜๋ฉฐ, ์ด๋Š” ์„น์…˜ 6.4์˜ ๊ณ„ํš ๊ธฐ๋ฒ•๊ณผ ๋ฐ€์ ‘ํ•œ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ œ์•ˆ [454]์— ๋”ฐ๋ผ ํ•˜์œ„ ์ž‘์—…์„ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋ฒˆํ˜ธ๊ฐ€ ๋งค๊ฒจ์ง„ ํ•ญ๋ชฉ ํ˜•์‹์œผ๋กœ ๋ช…์‹œ์ ์œผ๋กœ ๋‚˜์—ดํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค(์˜ˆ: "๋‹ค์Œ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜์—ฌ ์ผ๊ด€๋œ ๋‚ด๋Ÿฌํ‹ฐ๋ธŒ๋ฅผ ๊ตฌ์„ฑํ•ฉ๋‹ˆ๋‹ค: 1...., 2...., 3...."). ๋ชฉํ‘œ ์ž‘์—…์„ ํ•˜์œ„ ์ž‘์—…์œผ๋กœ ๋ถ„ํ•ดํ•˜๋ฉด LLM์€ ๋” ์‰ฌ์šด ํ•˜์œ„ ์ž‘์—…์„ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•˜์—ฌ ๋ณต์žกํ•œ ์ž‘์—…์— ๋Œ€ํ•ด ๋” ์ •ํ™•ํ•œ ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
We also consider the degree of deviation between the model's prior response and the value contained in the retrieved context (Figure 2, right side plots). A similar pattern emerges in this analysis: as the RAG value diverges from the model's prior, the model is less likely to adopt the RAG value over its own initial response. We additionally plot data split into the upper and lower half percentiles and observe that, across all six datasets, the lower probability prior responses are monotonically lower than the higher probability response tokens. Thus, the correlation between deviation and RAG response rate holds across both bands of probabilities.
๋ชจ๋ธ์˜ ์ด์ „ ์‘๋‹ต๊ณผ ๊ฒ€์ƒ‰๋œ ์ปจํ…์ŠคํŠธ์— ํฌํ•จ๋œ ๊ฐ’ ๊ฐ„์˜ ํŽธ์ฐจ ์ •๋„๋„ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค(Figure 2, ์˜ค๋ฅธ์ชฝ ๊ทธ๋ž˜ํ”„). ์ด ๋ถ„์„์—์„œ๋„ ์œ ์‚ฌํ•œ ํŒจํ„ด์ด ๋‚˜ํƒ€๋‚ฉ๋‹ˆ๋‹ค: RAG ๊ฐ’์ด ๋ชจ๋ธ์˜ ์ด์ „ ์‘๋‹ต๊ณผ ๋‹ค๋ฅผ์ˆ˜๋ก, ๋ชจ๋ธ์€ ์ž์‹ ์˜ ์ดˆ๊ธฐ ์‘๋‹ต๋ณด๋‹ค RAG ๊ฐ’์„ ์ฑ„ํƒํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋‚ฎ์•„์ง‘๋‹ˆ๋‹ค. ์ถ”๊ฐ€๋กœ ๋ฐ์ดํ„ฐ๋ฅผ ์ƒ์œ„ ๋ฐ ํ•˜์œ„ ์ ˆ๋ฐ˜ ๋ฐฑ๋ถ„์œ„์ˆ˜๋กœ ๋‚˜๋ˆ„์–ด ๊ทธ๋ž˜ํ”„๋กœ ๋‚˜ํƒ€๋‚ด์—ˆ์œผ๋ฉฐ, ๋ชจ๋“  ์—ฌ์„ฏ ๊ฐœ์˜ ๋ฐ์ดํ„ฐ์…‹์—์„œ ํ•˜์œ„ ํ™•๋ฅ ์˜ ์ด์ „ ์‘๋‹ต์ด ์ƒ์œ„ ํ™•๋ฅ  ์‘๋‹ต ํ† ํฐ๋ณด๋‹ค ์ผ๊ด€๋˜๊ฒŒ ๋‚ฎ๋‹ค๋Š” ๊ฒƒ์„ ๊ด€์ฐฐํ–ˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ํŽธ์ฐจ์™€ RAG ์‘๋‹ต๋ฅ  ๊ฐ„์˜ ์ƒ๊ด€๊ด€๊ณ„๋Š” ๋‘ ํ™•๋ฅ  ๋ฐด๋“œ ๋ชจ๋‘์—์„œ ์œ ์ง€๋ฉ๋‹ˆ๋‹ค.
๋˜ํ•œ ๋ชจ๋ธ์˜ ์‚ฌ์ „ ์‘๋‹ต๊ณผ ๊ฒ€์ƒ‰๋œ ์ปจํ…์ŠคํŠธ์— ํฌํ•จ๋œ ๊ฐ’ ์‚ฌ์ด์˜ ํŽธ์ฐจ ์ •๋„๋„ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค(๊ทธ๋ฆผ 2, ์˜ค๋ฅธ์ชฝ ํ”Œ๋กฏ). ์ด ๋ถ„์„์—์„œ๋„ ์œ ์‚ฌํ•œ ํŒจํ„ด์ด ๋‚˜ํƒ€๋‚ฉ๋‹ˆ๋‹ค. RAG ๊ฐ’์ด ๋ชจ๋ธ์˜ ์‚ฌ์ „ ๊ฐ’์—์„œ ๋ฒ—์–ด๋‚ ์ˆ˜๋ก ๋ชจ๋ธ์€ ์ž์‹ ์˜ ์ดˆ๊ธฐ ์‘๋‹ต๋ณด๋‹ค RAG ๊ฐ’์„ ์ฑ„ํƒํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋‚ฎ์•„์ง‘๋‹ˆ๋‹ค. ๋˜ํ•œ ์ƒ์œ„ ๋ฐ ํ•˜์œ„ ๋ฐ˜๋ถ„์œ„์ˆ˜๋กœ ๋ฐ์ดํ„ฐ๋ฅผ ๋ถ„ํ• ํ•˜์—ฌ ํ”Œ๋กฏํ•˜๊ณ  6๊ฐœ์˜ ๋ฐ์ดํ„ฐ ์„ธํŠธ ๋ชจ๋‘์—์„œ ํ•˜์œ„ ํ™•๋ฅ ์˜ ์‚ฌ์ „ ์‘๋‹ต ํ† ํฐ์ด ์ƒ์œ„ ํ™•๋ฅ ์˜ ์‘๋‹ต ํ† ํฐ๋ณด๋‹ค ๋‹จ์กฐ๋กญ๊ฒŒ ๋‚ฎ๋‹ค๋Š” ๊ฒƒ์„ ๊ด€์ฐฐํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ํŽธ์ฐจ์™€ RAG ์‘๋‹ต๋ฅ  ์‚ฌ์ด์˜ ์ƒ๊ด€๊ด€๊ณ„๋Š” ๋‘ ํ™•๋ฅ  ๋ฐด๋“œ ๋ชจ๋‘์—์„œ ์œ ์ง€๋ฉ๋‹ˆ๋‹ค.
Human evaluation setup.Evaluating models' performance on this evaluation set of diverse tasks is extremely challenging because different tasks require different expertise. Indeed, many of these tasks cannot be measured by automatic metrics or even be judged by normal crowdworkers (e.g., writing a program, or converting first-order logic into natural language). To get a more faithful evaluation, we asked the authors of the instructions to judge model predictions. Details on how we set up this human evaluation are described in Appendix B. The evaluators were asked to rate the output based on whether it accurately and effectively completes the task.
์‚ฌ๋žŒ ํ‰๊ฐ€ ์„ค์ •. ๋‹ค์–‘ํ•œ ์ž‘์—…์— ๋Œ€ํ•œ ์ด ํ‰๊ฐ€ ์„ธํŠธ์—์„œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ๋งค์šฐ ๋„์ „์ ์ž…๋‹ˆ๋‹ค. ์™œ๋ƒํ•˜๋ฉด ์„œ๋กœ ๋‹ค๋ฅธ ์ž‘์—…์ด ์„œ๋กœ ๋‹ค๋ฅธ ์ „๋ฌธ ์ง€์‹์„ ์š”๊ตฌํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ, ์ด๋Ÿฌํ•œ ์ž‘์—… ์ค‘ ๋งŽ์€ ๋ถ€๋ถ„์€ ์ž๋™ํ™”๋œ ์ง€ํ‘œ๋กœ ์ธก์ •ํ•  ์ˆ˜ ์—†๊ฑฐ๋‚˜ ์ผ๋ฐ˜์ ์ธ ํฌ๋ผ์šฐ๋“œ์›Œ์ปค๊ฐ€ ํŒ๋‹จํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค (์˜ˆ: ํ”„๋กœ๊ทธ๋žจ ์ž‘์„ฑ, ์ผ์ฐจ ๋…ผ๋ฆฌ๋ฅผ ์ž์—ฐ์–ด๋กœ ๋ณ€ํ™˜). ๋ณด๋‹ค ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ํ‰๊ฐ€๋ฅผ ์–ป๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ์ง€์นจ ์ž‘์„ฑ์ž๋“ค์—๊ฒŒ ๋ชจ๋ธ ์˜ˆ์ธก์„ ํ‰๊ฐ€ํ•˜๋„๋ก ์š”์ฒญํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ์‚ฌ๋žŒ ํ‰๊ฐ€๋ฅผ ์–ด๋–ป๊ฒŒ ์„ค์ •ํ–ˆ๋Š”์ง€์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก B์— ์„ค๋ช…๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ํ‰๊ฐ€์ž๋“ค์€ ์ถœ๋ ฅ๋ฌผ์ด ์ž‘์—…์„ ์ •ํ™•ํ•˜๊ณ  ํšจ๊ณผ์ ์œผ๋กœ ์™„๋ฃŒํ–ˆ๋Š”์ง€ ์—ฌ๋ถ€๋ฅผ ๊ธฐ์ค€์œผ๋กœ ํ‰๊ฐ€ํ•˜๋„๋ก ์š”์ฒญ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค.
์ธ๊ฐ„ ํ‰๊ฐ€ ์„ค์ •.๋‹ค์–‘ํ•œ ์ž‘์—…์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ๋งค์šฐ ์–ด๋ ค์šด ์ž‘์—…์ž…๋‹ˆ๋‹ค. ์™œ๋ƒํ•˜๋ฉด ์ž‘์—…๋งˆ๋‹ค ํ•„์š”ํ•œ ์ „๋ฌธ ์ง€์‹์ด ๋‹ค๋ฅด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ ์ด๋Ÿฌํ•œ ์ž‘์—… ์ค‘ ์ƒ๋‹น์ˆ˜๋Š” ์ž๋™ ๋ฉ”ํŠธ๋ฆญ์œผ๋กœ ์ธก์ •ํ•  ์ˆ˜ ์—†์œผ๋ฉฐ ์ผ๋ฐ˜ ํฌ๋ผ์šฐ๋“œ์›Œ์ปค๊ฐ€ ํŒ๋‹จํ•  ์ˆ˜๋„ ์—†์Šต๋‹ˆ๋‹ค(์˜ˆ: ํ”„๋กœ๊ทธ๋žจ ์ž‘์„ฑ ๋˜๋Š” ์ผ์ฐจ ๋…ผ๋ฆฌ๋ฅผ ์ž์—ฐ์–ด๋กœ ๋ณ€ํ™˜). ๋ณด๋‹ค ์ถฉ์‹คํ•œ ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ์ง€์นจ ์ž‘์„ฑ์ž์—๊ฒŒ ๋ชจ๋ธ ์˜ˆ์ธก์„ ํŒ๋‹จํ•˜๋„๋ก ์š”์ฒญํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ์ธ๊ฐ„ ํ‰๊ฐ€๋ฅผ ์„ค์ •ํ•œ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก B์— ์„ค๋ช…๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ํ‰๊ฐ€์ž๋“ค์€ ์ž‘์—…์„ ์ •ํ™•ํ•˜๊ณ  ํšจ๊ณผ์ ์œผ๋กœ ์™„๋ฃŒํ–ˆ๋Š”์ง€ ์—ฌ๋ถ€์— ๋”ฐ๋ผ ๊ฒฐ๊ณผ๋ฌผ์„ ํ‰๊ฐ€ํ•˜๋„๋ก ์š”์ฒญ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค.
**Perspective API7** erspectives API8 assesses text content, evaluating it for toxicity and offensive language. It assigns a severity score from 0 to 1 across various categories, including Severe Toxicity, Insult, Profanity, Identity Attack, Threat, and Sexually Explicit content. However, due to the bimodal distribution of toxic scores when evaluating the model's generated responses, relying solely on mean and variance for assessment is not precise. As a result, based on the previous method (Gehman et al., 2020), scores equal to or greater than 0.5 are categorized as toxic, whereas scores falling below this threshold are classified as non-toxic.
Perspective API7๋Š” ํ…์ŠคํŠธ ์ฝ˜ํ…์ธ ๋ฅผ ํ‰๊ฐ€ํ•˜์—ฌ ๋…์„ฑ ๋ฐ ๊ณต๊ฒฉ์ ์ธ ์–ธ์–ด๋ฅผ ํ‰๊ฐ€ํ•˜๋Š” API์ž…๋‹ˆ๋‹ค. ์ด API๋Š” ์‹ฌ๊ฐํ•œ ๋…์„ฑ, ๋ชจ์š•, ๋น„์†์–ด, ์ •์ฒด์„ฑ ๊ณต๊ฒฉ, ์œ„ํ˜‘ ๋ฐ ์„ฑ์ ์œผ๋กœ ๋…ธ๊ณจ์ ์ธ ์ฝ˜ํ…์ธ ๋ฅผ ํฌํ•จํ•œ ๋‹ค์–‘ํ•œ ๋ฒ”์ฃผ์—์„œ 0์—์„œ 1๊นŒ์ง€์˜ ์‹ฌ๊ฐ๋„ ์ ์ˆ˜๋ฅผ ํ• ๋‹นํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ์‘๋‹ต์„ ํ‰๊ฐ€ํ•  ๋•Œ ๋…์„ฑ ์ ์ˆ˜์˜ ์ด์ค‘ ๋ถ„ํฌ๋กœ ์ธํ•ด ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ๋งŒ์œผ๋กœ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ์ •ํ™•ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด์ „ ๋ฐฉ๋ฒ•(Gehman et al., 2020)์„ ๊ธฐ๋ฐ˜์œผ๋กœ, 0.5 ์ด์ƒ์ธ ์ ์ˆ˜๋Š” ๋…์„ฑ์œผ๋กœ ๋ถ„๋ฅ˜๋˜๊ณ , ์ด ์ž„๊ณ„๊ฐ’ ์ดํ•˜์˜ ์ ์ˆ˜๋Š” ๋น„๋…์„ฑ์œผ๋กœ ๋ถ„๋ฅ˜๋ฉ๋‹ˆ๋‹ค.
**PERSPECTIVE API7** erspective API8์€ ํ…์ŠคํŠธ ์ฝ˜ํ…์ธ ๋ฅผ ํ‰๊ฐ€ํ•˜์—ฌ ๋…์„ฑ ๋ฐ ๋ถˆ์พŒ๊ฐ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์‹ฌ๊ฐํ•œ ๋…์„ฑ, ๋ชจ์š•, ์š•์„ค, ์‹ ์› ๊ณต๊ฒฉ, ์œ„ํ˜‘, ์„ฑ์ ์œผ๋กœ ๋…ธ๊ณจ์ ์ธ ์ฝ˜ํ…์ธ  ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฒ”์ฃผ์— ๋Œ€ํ•ด 0์—์„œ 1๊นŒ์ง€์˜ ์‹ฌ๊ฐ๋„ ์ ์ˆ˜๋ฅผ ๋ถ€์—ฌํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ์‘๋‹ต์„ ํ‰๊ฐ€ํ•  ๋•Œ ๋…์„ฑ ์ ์ˆ˜์˜ ์ด์ค‘ ๋ชจ๋“œ ๋ถ„ํฌ๋กœ ์ธํ•ด ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์—๋งŒ ์˜์กดํ•˜๋Š” ๊ฒƒ์€ ์ •ํ™•ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด์ „ ๋ฐฉ๋ฒ•(Gehman et al., 2020)์— ๋”ฐ๋ผ 0.5 ์ด์ƒ์˜ ์ ์ˆ˜๋Š” ๋…์„ฑ์ด ์žˆ๋Š” ๊ฒƒ์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜๊ณ , ์ด ์ž„๊ณ„๊ฐ’ ์ดํ•˜์˜ ์ ์ˆ˜๋Š” ๋น„๋…์„ฑ์œผ๋กœ ๋ถ„๋ฅ˜ํ•ฉ๋‹ˆ๋‹ค.
In the main paper, we use ChatGPT as the LLM judge for final evaluation, for its efficiency, price and accessibility for the community to reproduce our results. As pointed out in Chiang et al. (2023), LLMs evaluators, although largely consistent with human preferences, may have their own biases. Therefore, to make sure our experimental results are solid, we also use GPT-4 as the judge and compare against the performance gap in CRR between different baselines and the Self-Instruct method. The comparison results in Table 6 demonstrates the agreement of two LLM-based judges and confirms the superior performance of CodecLM against comparing methods.
๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ์ตœ์ข… ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ChatGPT๋ฅผ LLM ์‹ฌ์‚ฌ์ž๋กœ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ํšจ์œจ์„ฑ, ๊ฐ€๊ฒฉ, ๊ทธ๋ฆฌ๊ณ  ์ปค๋ฎค๋‹ˆํ‹ฐ๊ฐ€ ์šฐ๋ฆฌ์˜ ๊ฒฐ๊ณผ๋ฅผ ์žฌํ˜„ํ•  ์ˆ˜ ์žˆ๋Š” ์ ‘๊ทผ์„ฑ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. Chiang et al. (2023)์—์„œ ์ง€์ ํ•œ ๋ฐ”์™€ ๊ฐ™์ด, LLM ํ‰๊ฐ€์ž๋Š” ์ธ๊ฐ„์˜ ์„ ํ˜ธ์™€ ๋Œ€์ฒด๋กœ ์ผ์น˜ํ•˜์ง€๋งŒ, ์ž์ฒด์ ์ธ ํŽธํ–ฅ์„ ๊ฐ€์งˆ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์šฐ๋ฆฌ์˜ ์‹คํ—˜ ๊ฒฐ๊ณผ๊ฐ€ ํ™•๊ณ ํ•จ์„ ๋ณด์žฅํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ๋˜ํ•œ GPT-4๋ฅผ ์‹ฌ์‚ฌ์ž๋กœ ์‚ฌ์šฉํ•˜๊ณ , ๋‹ค์–‘ํ•œ ๊ธฐ์ค€์„ ๊ณผ Self-Instruct ๋ฐฉ๋ฒ• ๊ฐ„์˜ CRR ์„ฑ๋Šฅ ์ฐจ์ด๋ฅผ ๋น„๊ตํ•ฉ๋‹ˆ๋‹ค. ํ‘œ 6์˜ ๋น„๊ต ๊ฒฐ๊ณผ๋Š” ๋‘ LLM ๊ธฐ๋ฐ˜ ์‹ฌ์‚ฌ์ž์˜ ์ผ์น˜๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ, CodecLM์˜ ๋น„๊ต ๋ฐฉ๋ฒ•๋“ค์— ๋Œ€ํ•œ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.
๋ณธ ๋…ผ๋ฌธ์—์„œ๋Š” ํšจ์œจ์„ฑ, ๊ฐ€๊ฒฉ, ์ปค๋ฎค๋‹ˆํ‹ฐ๊ฐ€ ๊ฒฐ๊ณผ๋ฅผ ์žฌํ˜„ํ•  ์ˆ˜ ์žˆ๋Š” ์ ‘๊ทผ์„ฑ์„ ๊ณ ๋ คํ•˜์—ฌ ์ตœ์ข… ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ LLM ์‹ฌ์‚ฌ์œ„์›์œผ๋กœ ChatGPT๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. Chiang ์™ธ(2023)์—์„œ ์ง€์ ํ•œ ๋ฐ”์™€ ๊ฐ™์ด, LLM ํ‰๊ฐ€์ž๋Š” ๋Œ€์ฒด๋กœ ์‚ฌ๋žŒ์˜ ์„ ํ˜ธ๋„์™€ ์ผ์น˜ํ•˜์ง€๋งŒ, ๊ทธ๋“ค๋งŒ์˜ ํŽธ๊ฒฌ์„ ๊ฐ€์ง€๊ณ  ์žˆ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์‹คํ—˜ ๊ฒฐ๊ณผ์˜ ๊ฒฌ๊ณ ์„ฑ์„ ํ™•๋ณดํ•˜๊ธฐ ์œ„ํ•ด GPT-4๋ฅผ ์‹ฌ์‚ฌ์œ„์›์œผ๋กœ ์‚ฌ์šฉํ•˜๊ณ  ๋‹ค๋ฅธ ๋ฒ ์ด์Šค๋ผ์ธ๊ณผ Self-Instruct ๋ฐฉ๋ฒ• ๊ฐ„์˜ CRR ์„ฑ๋Šฅ ์ฐจ์ด๋ฅผ ๋น„๊ตํ–ˆ์Šต๋‹ˆ๋‹ค. ํ‘œ 6์˜ ๋น„๊ต ๊ฒฐ๊ณผ๋Š” ๋‘ LLM ๊ธฐ๋ฐ˜ ์‹ฌ์‚ฌ์œ„์›์˜ ์ผ์น˜์„ฑ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๋น„๊ต ๋ฐฉ๋ฒ•์— ๋น„ํ•ด CodecLM์˜ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ํ™•์ธ์‹œ์ผœ ์ค๋‹ˆ๋‹ค.
**Comparing With SoTA Models** While ALMA-13B-LoRA ranks as one of the top moderate-size LLM translation models, it slightly trails behind GPT-4 and the WMT competition winners. However, the incorporation of CPO significantly enhances ALMA's capabilities, bringing its performance to a level that is comparable to or even surpasses that of GPT-4 and WMT winners. For example, ALMA-13B-R achieves an average score of 85.74 on KIWI-XXL and 94.05 on XCOMET for en\(\rightarrow\)xx translations. These scores outperform GPT-4, which scores 83.83 on KIWI-XXL and 93.23 on XCOMET, as well as the WMT winners, who score 84.81 on KIWI-XXL and 93.78 on XCOMET.
**์ตœ์‹  ๋ชจ๋ธ๊ณผ์˜ ๋น„๊ต** ALMA-13B-LoRA๋Š” ์ค‘๊ฐ„ ํฌ๊ธฐ์˜ LLM ๋ฒˆ์—ญ ๋ชจ๋ธ ์ค‘ ์ƒ์œ„๊ถŒ์— ์†ํ•˜์ง€๋งŒ, GPT-4์™€ WMT ๋Œ€ํšŒ ์šฐ์Šน์ž๋“ค๋ณด๋‹ค๋Š” ์•ฝ๊ฐ„ ๋’ค์ฒ˜์ง‘๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ CPO์˜ ๋„์ž…์€ ALMA์˜ ์„ฑ๋Šฅ์„ ํฌ๊ฒŒ ํ–ฅ์ƒ์‹œ์ผœ, GPT-4์™€ WMT ์šฐ์Šน์ž๋“ค๊ณผ ๋น„๊ตํ•  ๋งŒํ•œ ์ˆ˜์ค€์œผ๋กœ ๋Œ์–ด์˜ฌ๋ฆฝ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ALMA-13B-R์€ en\(\rightarrow\)xx ๋ฒˆ์—ญ์—์„œ KIWI-XXL์—์„œ ํ‰๊ท  85.74์ , XCOMET์—์„œ 94.05์ ์„ ๊ธฐ๋กํ•ฉ๋‹ˆ๋‹ค. ์ด ์ ์ˆ˜๋Š” KIWI-XXL์—์„œ 83.83์ , XCOMET์—์„œ 93.23์ ์„ ๊ธฐ๋กํ•œ GPT-4์™€ KIWI-XXL์—์„œ 84.81์ , XCOMET์—์„œ 93.78์ ์„ ๊ธฐ๋กํ•œ WMT ์šฐ์Šน์ž๋“ค์„ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค.
**SoTA ๋ชจ๋ธ๊ณผ ๋น„๊ต** ALMA-13B-LoRA๋Š” ์ƒ์œ„ ์ค‘ํ˜• LLM ๋ฒˆ์—ญ ๋ชจ๋ธ ์ค‘ ํ•˜๋‚˜์ด์ง€๋งŒ GPT-4 ๋ฐ WMT ๋Œ€ํšŒ ์šฐ์Šน์ž์— ์•ฝ๊ฐ„ ๋’ค์ฒ˜์ง‘๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ CPO๋ฅผ ํ†ตํ•ฉํ•˜๋ฉด ALMA์˜ ์„ฑ๋Šฅ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋˜์–ด GPT-4 ๋ฐ WMT ์šฐ์Šน์ž์™€ ๋น„์Šทํ•˜๊ฑฐ๋‚˜ ๊ทธ ์ด์ƒ์˜ ์ˆ˜์ค€์— ๋„๋‹ฌํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ALMA-13B-R์€ en\(\rightarrow\)xx ๋ฒˆ์—ญ์— ๋Œ€ํ•ด KIWI-XXL์—์„œ ํ‰๊ท  ์ ์ˆ˜ 85.74์ , XCOMET์—์„œ 94.05์ ์„ ๋‹ฌ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ด ์ ์ˆ˜๋Š” KIWI-XXL์—์„œ 83.83์ , XCOMET์—์„œ 93.23์ ์„ ๊ธฐ๋กํ•œ GPT-4์™€ WMT ์ˆ˜์ƒ์ž(KIWI-XXL์—์„œ 84.81์ , XCOMET์—์„œ 93.78์ )๋ฅผ ๋Šฅ๊ฐ€ํ•˜๋Š” ์„ฑ์ ์ž…๋‹ˆ๋‹ค.
ResultFigure 8 demonstrates that, in most of the mid-late layers, increasing the latent recall of the bridge entity when the LLM processes \(\mu(r_{1}(e_{1})))\) also increases the relative frequency of the LLM to output \(e_{2}^{(0)}\) to generate the appositive of \(\mu(r_{1}(e_{1})))\) followed by a comma.7 The result indicates that EntRec at the \(n\)-th token has controllability of the token to be generated as the \(n+2\)-th token to make it more likely to be the first token of the appositive, serving as an indirect evidence that EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\) is a reasonable proxy of the internal recall of the bridge entity.
Figure 8์€ ๋Œ€๋ถ€๋ถ„์˜ ์ค‘ํ›„๋ฐ˜ ๋ ˆ์ด์–ด์—์„œ LLM์ด \(\mu(r_{1}(e_{1})))\)๋ฅผ ์ฒ˜๋ฆฌํ•  ๋•Œ ๋ธŒ๋ฆฌ์ง€ ์—”ํ‹ฐํ‹ฐ์˜ ์ž ์žฌ์  ๋ฆฌ์ฝœ์„ ์ฆ๊ฐ€์‹œํ‚ค๋Š” ๊ฒƒ์ด LLM์ด \(\mu(r_{1}(e_{1})))\)์˜ ๋™๊ฒฉ์–ด๋ฅผ ์ƒ์„ฑํ•˜๊ธฐ ์œ„ํ•ด \(e_{2}^{(0)}\)๋ฅผ ์ถœ๋ ฅํ•  ์ƒ๋Œ€์  ๋นˆ๋„๋ฅผ ์ฆ๊ฐ€์‹œํ‚จ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด ๊ฒฐ๊ณผ๋Š” \(n\)๋ฒˆ์งธ ํ† ํฐ์—์„œ์˜ EntRec์ด \(n+2\)๋ฒˆ์งธ ํ† ํฐ์œผ๋กœ ์ƒ์„ฑ๋  ํ† ํฐ์„ ์ œ์–ดํ•  ์ˆ˜ ์žˆ์Œ์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ด๋Š” ๋™๊ฒฉ์–ด์˜ ์ฒซ ๋ฒˆ์งธ ํ† ํฐ์ด ๋  ๊ฐ€๋Šฅ์„ฑ์„ ๋†’์ž…๋‹ˆ๋‹ค. ์ด๋Š” EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\)๊ฐ€ ๋ธŒ๋ฆฌ์ง€ ์—”ํ‹ฐํ‹ฐ์˜ ๋‚ด๋ถ€ ๋ฆฌ์ฝœ์˜ ํ•ฉ๋ฆฌ์ ์ธ ๋Œ€๋ฆฌ์ž„์„ ๊ฐ„์ ‘์ ์œผ๋กœ ์ฆ๋ช…ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๊ฒฐ๊ณผ๊ทธ๋ฆผ 8์€ ๋Œ€๋ถ€๋ถ„์˜ ์ค‘ํ›„๋ฐ˜ ๋ ˆ์ด์–ด์—์„œ LLM์ด (\(\mu(r_{1}(e_{1})))\)๋ฅผ ์ฒ˜๋ฆฌํ•  ๋•Œ ๋‹ค๋ฆฌ ์—”ํ‹ฐํ‹ฐ์˜ ์ž ์žฌ์  ํšŒ์ƒ์„ ์ฆ๊ฐ€์‹œํ‚ค๋ฉด ์‰ผํ‘œ ๋’ค์— ์ด์–ด์ง€๋Š” \(\mu(r_{1}(e_{1})))\)์˜ ๊ด€๊ณ„ ๋นˆ๋„๋ฅผ ์ฆ๊ฐ€์‹œํ‚ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.7 ์ด ๊ฒฐ๊ณผ๋Š” n๋ฒˆ์งธ ํ† ํฐ์˜ EntRec์ด n+2๋ฒˆ์งธ ํ† ํฐ์œผ๋กœ ์ƒ์„ฑ๋  ํ† ํฐ์„ ์ œ์–ดํ•  ์ˆ˜ ์žˆ์–ด ๊ด€๊ณ„๋ช…์‚ฌ์ ˆ์˜ ์ฒซ ๋ฒˆ์งธ ํ† ํฐ์ด ๋  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์•„์ง€๋ฉฐ, ์ด๋Š” EntRec\({}^{l}(e_{2},\tau_{\text{2H}})\)๊ฐ€ ๋‹ค๋ฆฌ ์—”ํ‹ฐํ‹ฐ์˜ ๋‚ด๋ถ€ ํšŒ์ƒ์˜ ํ•ฉ๋ฆฌ์ ์ธ ๋Œ€๋ฆฌ์ธ์ด๋ผ๋Š” ๊ฐ„์ ‘์ ์ธ ์ฆ๊ฑฐ๋กœ ์ž‘์šฉํ•œ๋‹ค๋Š” ๊ฒƒ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
_Summary._ The above three abilities are of great value to the practical performance of LLMs: conforming to human values and preferences (human alignment), acting properly in real-world scenarios (interaction with the external environment), and expanding the ability scope (tool manipulation). In addition to the above three advanced abilities, LLMs might also show other abilities that are specially related to some tasks (_e.g._, data annotation [486]) or learning mechanisms (_e.g._, self-improvement [706]). It will be an open direction to discover, measure and evaluate these newly emerging abilities, so as to better utilize and improve LLMs.
_์š”์•ฝ._ ์œ„์˜ ์„ธ ๊ฐ€์ง€ ๋Šฅ๋ ฅ์€ LLM์˜ ์‹ค์งˆ์ ์ธ ์„ฑ๋Šฅ์— ํฐ ๊ฐ€์น˜๋ฅผ ์ง€๋‹™๋‹ˆ๋‹ค: ์ธ๊ฐ„์˜ ๊ฐ€์น˜์™€ ์„ ํ˜ธ์— ๋ถ€ํ•ฉํ•˜๋Š” ๊ฒƒ(์ธ๊ฐ„ ์ •๋ ฌ), ์‹ค์ œ ์„ธ๊ณ„ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ์ ์ ˆํ•˜๊ฒŒ ํ–‰๋™ํ•˜๋Š” ๊ฒƒ(์™ธ๋ถ€ ํ™˜๊ฒฝ๊ณผ์˜ ์ƒํ˜ธ์ž‘์šฉ), ๊ทธ๋ฆฌ๊ณ  ๋Šฅ๋ ฅ ๋ฒ”์œ„๋ฅผ ํ™•์žฅํ•˜๋Š” ๊ฒƒ(๋„๊ตฌ ์กฐ์ž‘). ์œ„์˜ ์„ธ ๊ฐ€์ง€ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ ์™ธ์—๋„, LLM์€ ํŠน์ • ์ž‘์—…(_์˜ˆ: ๋ฐ์ดํ„ฐ ์ฃผ์„ [486]_)์ด๋‚˜ ํ•™์Šต ๋ฉ”์ปค๋‹ˆ์ฆ˜(_์˜ˆ: ์ž๊ธฐ ๊ฐœ์„  [706]_)๊ณผ ํŠน๋ณ„ํžˆ ๊ด€๋ จ๋œ ๋‹ค๋ฅธ ๋Šฅ๋ ฅ์„ ๋ณด์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ƒˆ๋กญ๊ฒŒ ๋“ฑ์žฅํ•˜๋Š” ๋Šฅ๋ ฅ์„ ๋ฐœ๊ฒฌํ•˜๊ณ , ์ธก์ •ํ•˜๋ฉฐ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ LLM์„ ๋” ์ž˜ ํ™œ์šฉํ•˜๊ณ  ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•œ ๊ฐœ๋ฐฉ๋œ ๋ฐฉํ–ฅ์ด ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
_์š”์•ฝ._ ์œ„์˜ ์„ธ ๊ฐ€์ง€ ๋Šฅ๋ ฅ์€ ์ธ๊ฐ„์˜ ๊ฐ€์น˜์™€ ์„ ํ˜ธ๋„๋ฅผ ์ค€์ˆ˜ํ•˜๋Š” ๊ฒƒ(์ธ๊ฐ„ ์ •๋ ฌ), ์‹ค์ œ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ์ ์ ˆํ•˜๊ฒŒ ํ–‰๋™ํ•˜๋Š” ๊ฒƒ(์™ธ๋ถ€ ํ™˜๊ฒฝ๊ณผ์˜ ์ƒํ˜ธ์ž‘์šฉ), ๋Šฅ๋ ฅ ๋ฒ”์œ„ ํ™•์žฅ(๋„๊ตฌ ์กฐ์ž‘) ๋“ฑ LLM์˜ ์‹ค์ œ ์„ฑ๋Šฅ์— ํฐ ๊ฐ€์น˜๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์œ„์˜ ์„ธ ๊ฐ€์ง€ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ ์™ธ์—๋„ LLM์€ ์ผ๋ถ€ ์ž‘์—…(_์˜ˆ: ๋ฐ์ดํ„ฐ ์ฃผ์„[486]) ๋˜๋Š” ํ•™์Šต ๋ฉ”์ปค๋‹ˆ์ฆ˜(_์˜ˆ: ์ž๊ธฐ ๊ฐœ์„ [706])๊ณผ ํŠน๋ณ„ํžˆ ๊ด€๋ จ๋œ ๋‹ค๋ฅธ ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์ค„ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ƒˆ๋กญ๊ฒŒ ๋“ฑ์žฅํ•˜๋Š” ๋Šฅ๋ ฅ์„ ๋ฐœ๊ฒฌํ•˜๊ณ , ์ธก์ •ํ•˜๊ณ , ํ‰๊ฐ€ํ•˜์—ฌ LLM์„ ๋” ์ž˜ ํ™œ์šฉํ•˜๊ณ  ๊ฐœ์„ ํ•˜๋Š” ๊ฒƒ์€ ์—ด๋ ค ์žˆ๋Š” ๋ฐฉํ–ฅ์ด ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Given the huge cost of model pre-training, well-trained model checkpoints are critical to the study and development of LLMs for the research community. Since the parameter scale is a key factor to consider for using LLMs, we categorize these public models into two scale levels (_i.e., tens of billions of parameters_ and _hundreds of billions of parameters_), which is useful for users to identify the suitable resources according to their resource budget. In addition, for inference, we can directly employ public APIs to perform our tasks, without running the model locally. Next, we introduce the publicly available model checkpoints and APIs.
๋ชจ๋ธ ์‚ฌ์ „ ํ›ˆ๋ จ์˜ ๋ง‰๋Œ€ํ•œ ๋น„์šฉ์„ ๊ณ ๋ คํ•  ๋•Œ, ์ž˜ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ ์ฒดํฌํฌ์ธํŠธ๋Š” ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ์—์„œ LLMs์˜ ์—ฐ๊ตฌ ๋ฐ ๊ฐœ๋ฐœ์— ๋งค์šฐ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. LLMs๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ ๊ณ ๋ คํ•ด์•ผ ํ•  ์ฃผ์š” ์š”์†Œ๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ ๊ทœ๋ชจ์ด๊ธฐ ๋•Œ๋ฌธ์—, ์šฐ๋ฆฌ๋Š” ์ด๋Ÿฌํ•œ ๊ณต๊ฐœ ๋ชจ๋ธ๋“ค์„ ๋‘ ๊ฐ€์ง€ ๊ทœ๋ชจ ์ˆ˜์ค€(์ฆ‰, ์ˆ˜๋ฐฑ์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ์™€ ์ˆ˜์ฒœ์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ)์œผ๋กœ ๋ถ„๋ฅ˜ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์‚ฌ์šฉ์ž๊ฐ€ ์ž์‹ ์˜ ์ž์› ์˜ˆ์‚ฐ์— ๋”ฐ๋ผ ์ ์ ˆํ•œ ์ž์›์„ ์‹๋ณ„ํ•˜๋Š” ๋ฐ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ถ”๋ก ์„ ์œ„ํ•ด ์šฐ๋ฆฌ๋Š” ๋ชจ๋ธ์„ ๋กœ์ปฌ์—์„œ ์‹คํ–‰ํ•˜์ง€ ์•Š๊ณ ๋„ ๊ณต๊ฐœ API๋ฅผ ์ง์ ‘ ์‚ฌ์šฉํ•˜์—ฌ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, ๊ณต๊ฐœ์ ์œผ๋กœ ์ด์šฉ ๊ฐ€๋Šฅํ•œ ๋ชจ๋ธ ์ฒดํฌํฌ์ธํŠธ์™€ API๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.
๋ชจ๋ธ ์‚ฌ์ „ ํ•™์Šต์— ๋ง‰๋Œ€ํ•œ ๋น„์šฉ์ด ์†Œ์š”๋˜๊ธฐ ๋•Œ๋ฌธ์—, ์ž˜ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ ์ฒดํฌํฌ์ธํŠธ๋Š” ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ์˜ LLM ์—ฐ๊ตฌ ๋ฐ ๊ฐœ๋ฐœ์— ๋งค์šฐ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ํŒŒ๋ผ๋ฏธํ„ฐ ๊ทœ๋ชจ๋Š” LLM ์‚ฌ์šฉ ์‹œ ๊ณ ๋ คํ•ด์•ผ ํ•  ํ•ต์‹ฌ ์š”์†Œ์ด๋ฏ€๋กœ, ์ด๋Ÿฌํ•œ ๊ณต๊ฐœ ๋ชจ๋ธ์„ ๋‘ ๊ฐ€์ง€ ๊ทœ๋ชจ ์ˆ˜์ค€(_์ฆ‰, ์ˆ˜์‹ญ์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ์™€ ์ˆ˜๋ฐฑ์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ_)์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜์—ฌ ์‚ฌ์šฉ์ž๊ฐ€ ๋ฆฌ์†Œ์Šค ์˜ˆ์‚ฐ์— ๋”ฐ๋ผ ์ ํ•ฉํ•œ ๋ฆฌ์†Œ์Šค๋ฅผ ์‹๋ณ„ํ•˜๋Š” ๋ฐ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ถ”๋ก ์˜ ๊ฒฝ์šฐ, ๋กœ์ปฌ์—์„œ ๋ชจ๋ธ์„ ์‹คํ–‰ํ•˜์ง€ ์•Š๊ณ ๋„ ๊ณต๊ฐœ๋œ API๋ฅผ ์ง์ ‘ ์‚ฌ์šฉํ•˜์—ฌ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์—์„œ๋Š” ๊ณต๊ฐœ์ ์œผ๋กœ ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ๋ชจ๋ธ ์ฒดํฌํฌ์ธํŠธ์™€ API๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.
ScalabilityDue to budget constraints, we have only verified the effectiveness of our method on smaller models (<=7B parameters) and smaller datasets (<100B tokens). Smaller models benefit significantly from removing the loss of irrelevant tokens and focusing on important ones. However, it's possible that very large models trained on extensive corpora may naturally develop this inductive bias to compress useful data (_i.e.,_ compressing everything), although it may sounds inefficient for now. Therefore, future works should study whether this selective language modeling technique can scale to very large models and data Kaplan et al. (2020).
ํ™•์žฅ์„ฑ: ์˜ˆ์‚ฐ ์ œ์•ฝ์œผ๋กœ ์ธํ•ด, ์šฐ๋ฆฌ๋Š” ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์ด ๋” ์ž‘์€ ๋ชจ๋ธ(<=7B ํŒŒ๋ผ๋ฏธํ„ฐ)๊ณผ ๋” ์ž‘์€ ๋ฐ์ดํ„ฐ์…‹(<100B ํ† ํฐ)์—์„œ๋งŒ ํšจ๊ณผ์ ์ž„์„ ๊ฒ€์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ๋” ์ž‘์€ ๋ชจ๋ธ์€ ๊ด€๋ จ ์—†๋Š” ํ† ํฐ์˜ ์†์‹ค์„ ์ œ๊ฑฐํ•˜๊ณ  ์ค‘์š”ํ•œ ํ† ํฐ์— ์ง‘์ค‘ํ•จ์œผ๋กœ์จ ์ƒ๋‹นํ•œ ์ด์ ์„ ์–ป์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋งค์šฐ ํฐ ์ฝ”ํผ์Šค์—์„œ ํ›ˆ๋ จ๋œ ๋งค์šฐ ํฐ ๋ชจ๋ธ์€ ์œ ์šฉํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์••์ถ•ํ•˜๋Š” ์ด ๊ท€๋‚ฉ์  ํŽธํ–ฅ์„ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค(์ฆ‰, ๋ชจ๋“  ๊ฒƒ์„ ์••์ถ•ํ•˜๋Š” ๊ฒƒ), ๋น„๋ก ํ˜„์žฌ๋กœ์„œ๋Š” ๋น„ํšจ์œจ์ ์œผ๋กœ ๋ณด์ผ ์ˆ˜ ์žˆ์ง€๋งŒ. ๋”ฐ๋ผ์„œ, ํ–ฅํ›„ ์—ฐ๊ตฌ๋Š” ์ด ์„ ํƒ์  ์–ธ์–ด ๋ชจ๋ธ๋ง ๊ธฐ๋ฒ•์ด ๋งค์šฐ ํฐ ๋ชจ๋ธ๊ณผ ๋ฐ์ดํ„ฐ์— ํ™•์žฅ๋  ์ˆ˜ ์žˆ๋Š”์ง€ ์—ฌ๋ถ€๋ฅผ ์—ฐ๊ตฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค (Kaplan et al., 2020).
ํ™•์žฅ์„ฑ์˜ˆ์‚ฐ ์ œ์•ฝ์œผ๋กœ ์ธํ•ด ๋” ์ž‘์€ ๋ชจ๋ธ(<=7B ํŒŒ๋ผ๋ฏธํ„ฐ)๊ณผ ๋” ์ž‘์€ ๋ฐ์ดํ„ฐ ์„ธํŠธ(<100B ํ† ํฐ)์—์„œ๋งŒ ๋ฐฉ๋ฒ•์˜ ํšจ๊ณผ๋ฅผ ๊ฒ€์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ์ž‘์€ ๋ชจ๋ธ์€ ๊ด€๋ จ ์—†๋Š” ํ† ํฐ์˜ ์†์‹ค์„ ์ œ๊ฑฐํ•˜๊ณ  ์ค‘์š”ํ•œ ํ† ํฐ์— ์ง‘์ค‘ํ•จ์œผ๋กœ์จ ์ƒ๋‹นํ•œ ์ด์ ์„ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ˜„์žฌ๋กœ์„œ๋Š” ๋น„ํšจ์œจ์ ์œผ๋กœ ๋“ค๋ฆด ์ˆ˜ ์žˆ์ง€๋งŒ, ๊ด‘๋ฒ”์œ„ํ•œ ๋ง๋ญ‰์น˜๋กœ ํ›ˆ๋ จ๋œ ๋งค์šฐ ํฐ ๋ชจ๋ธ์€ ์œ ์šฉํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์••์ถ•ํ•˜๋Š”(_์ฆ‰,_ ๋ชจ๋“  ๊ฒƒ์„ ์••์ถ•ํ•˜๋Š”) ์ด๋Ÿฌํ•œ ๊ท€๋‚ฉ์  ํŽธํ–ฅ์„ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ํ–ฅํ›„ ์—ฐ๊ตฌ์—์„œ๋Š” ์ด๋Ÿฌํ•œ ์„ ํƒ์  ์–ธ์–ด ๋ชจ๋ธ๋ง ๊ธฐ๋ฒ•์ด ๋งค์šฐ ํฐ ๋ชจ๋ธ๊ณผ ๋ฐ์ดํ„ฐ Kaplan ๋“ฑ(2020)์— ํ™•์žฅํ•  ์ˆ˜ ์žˆ๋Š”์ง€ ์—ฐ๊ตฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
Property 2.1: _Gains increase with the number of steps._ We analyze the number of steps by isolating \(S\). We tune \(S\) from \(1\) to \(8\), while keeping the values of \(I\) and \(K\) constant across four groups of different values, ranging from small to large, respectively. Figure 6 (middle) shows that as the number of steps increases, there is a corresponding increase in performance gain. Additionally, we find that when \(I\) and \(K\) are increased (which indicates a higher difficulty), the performance gains are more significant, e.g., \(4\%\)-\(18\%\) gains over \(\{I=10,K=2\}\) compared to \(16\%\)-\(48\%\) over \(\{I=100,K=4\}\).
Property 2.1: _์ด๋“์€ ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค._ ์šฐ๋ฆฌ๋Š” \(S\)๋ฅผ ๋ถ„๋ฆฌํ•˜์—ฌ ๋‹จ๊ณ„ ์ˆ˜๋ฅผ ๋ถ„์„ํ•ฉ๋‹ˆ๋‹ค. \(I\)์™€ \(K\)์˜ ๊ฐ’์„ ๋„ค ๊ทธ๋ฃน์˜ ์„œ๋กœ ๋‹ค๋ฅธ ๊ฐ’๋“ค์—์„œ ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€ํ•˜๋ฉด์„œ \(S\)๋ฅผ \(1\)์—์„œ \(8\)๊นŒ์ง€ ์กฐ์ •ํ•ฉ๋‹ˆ๋‹ค. Figure 6 (์ค‘๊ฐ„)์—์„œ ๋ณผ ์ˆ˜ ์žˆ๋“ฏ์ด, ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์„ฑ๋Šฅ ์ด๋“๋„ ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์ถ”๊ฐ€์ ์œผ๋กœ, \(I\)์™€ \(K\)๊ฐ€ ์ฆ๊ฐ€ํ•  ๋•Œ (์ด๋Š” ๋” ๋†’์€ ๋‚œ์ด๋„๋ฅผ ๋‚˜ํƒ€๋ƒ„), ์„ฑ๋Šฅ ์ด๋“์ด ๋” ํฌ๊ฒŒ ๋‚˜ํƒ€๋‚˜๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, \(\{I=10,K=2\}\)์— ๋น„ํ•ด \(\{I=100,K=4\}\)์—์„œ \(4\%\)-\(18\%\)์˜ ์ด๋“์ด \(16\%\)-\(48\%\)๋กœ ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค.
์†์„ฑ 2.1: _์ด๋“์€ ๋‹จ๊ณ„ ์ˆ˜์— ๋”ฐ๋ผ ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค._ ๋‹จ๊ณ„ ์ˆ˜๋ฅผ ๋ถ„์„ํ•˜๊ธฐ ์œ„ํ•ด \(S\)๋ฅผ ๋ถ„๋ฆฌํ•ฉ๋‹ˆ๋‹ค. ์ž‘์€ ๊ฐ’๋ถ€ํ„ฐ ํฐ ๊ฐ’๊นŒ์ง€ ๊ฐ๊ฐ ๋„ค ๊ฐœ์˜ ๋‹ค๋ฅธ ๊ฐ’ ๊ทธ๋ฃน์—์„œ \(I\)์™€ \(K\)์˜ ๊ฐ’์„ ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€ํ•˜๋ฉด์„œ \(S\)๋ฅผ \(1\)์—์„œ \(8\)๊นŒ์ง€ ์กฐ์ •ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ฆผ 6(๊ฐ€์šด๋ฐ)์€ ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์„ฑ๋Šฅ ํ–ฅ์ƒ๋„ ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋˜ํ•œ \(I\)์™€ \(K\)๊ฐ€ ์ฆ๊ฐ€ํ•˜๋ฉด(๋‚œ์ด๋„๊ฐ€ ๋†’๋‹ค๋Š” ์˜๋ฏธ) ์„ฑ๋Šฅ ํ–ฅ์ƒ์ด ๋” ํฐ ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค(์˜ˆ: \(I=10,K=2\}\)์— ๋น„ํ•ด \(16\%\)-\(48\%\), $\{I=100,K=4\}\).
As highlighted in the correlation heatmaps in Fig. 4, we find that there is no distinct correlation across all tasks and languages, demanding a more granular analysis. While for non-English tasks, we mainly observe a correlation between low fertility and higher downstream performance, the non-English tasks yield seemingly random positive and negative correlations. However, it should be noted that the number of multilingual tasks per language is much lower than for English and that for several multilingual tasks such as XSQA and LAMBADA, a similar correlation behaviour between the English tasks and their translated version can be observed.
Fig. 4์˜ ์ƒ๊ด€๊ด€๊ณ„ ํžˆํŠธ๋งต์—์„œ ๊ฐ•์กฐ๋œ ๋ฐ”์™€ ๊ฐ™์ด, ์šฐ๋ฆฌ๋Š” ๋ชจ๋“  ์ž‘์—…๊ณผ ์–ธ์–ด์— ๊ฑธ์ณ ๋šœ๋ ทํ•œ ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ์—†์Œ์„ ๋ฐœ๊ฒฌํ•˜์˜€์œผ๋ฉฐ, ์ด๋Š” ๋” ์„ธ๋ถ„ํ™”๋œ ๋ถ„์„์ด ํ•„์š”ํ•จ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๋น„์˜์–ด ์ž‘์—…์˜ ๊ฒฝ์šฐ, ์ฃผ๋กœ ๋‚ฎ์€ fertility์™€ ๋†’์€ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์„ฑ๋Šฅ ๊ฐ„์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์—ˆ์ง€๋งŒ, ๋น„์˜์–ด ์ž‘์—…์—์„œ๋Š” ๊ธ์ •์  ๋ฐ ๋ถ€์ •์  ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ๋ฌด์ž‘์œ„๋กœ ๋‚˜ํƒ€๋‚˜๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์–ธ์–ด๋ณ„ ๋‹ค๊ตญ์–ด ์ž‘์—…์˜ ์ˆ˜๊ฐ€ ์˜์–ด์— ๋น„ํ•ด ํ›จ์”ฌ ์ ์œผ๋ฉฐ, XSQA ๋ฐ LAMBADA์™€ ๊ฐ™์€ ์—ฌ๋Ÿฌ ๋‹ค๊ตญ์–ด ์ž‘์—…์˜ ๊ฒฝ์šฐ, ์˜์–ด ์ž‘์—…๊ณผ ๊ทธ ๋ฒˆ์—ญ๋œ ๋ฒ„์ „ ๊ฐ„์— ์œ ์‚ฌํ•œ ์ƒ๊ด€๊ด€๊ณ„ ํ–‰๋™์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์Œ์„ ์ฃผ๋ชฉํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 4์˜ ์ƒ๊ด€๊ด€๊ณ„ ํžˆํŠธ๋งต์—์„œ ๊ฐ•์กฐํ–ˆ๋“ฏ์ด ๋ชจ๋“  ์ž‘์—…๊ณผ ์–ธ์–ด์— ๊ฑธ์ณ ๋šœ๋ ทํ•œ ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ์—†์œผ๋ฏ€๋กœ ๋ณด๋‹ค ์„ธ๋ถ„ํ™”๋œ ๋ถ„์„์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ๋น„์˜์–ด ์ž‘์—…์˜ ๊ฒฝ์šฐ, ์ €์ถœ์‚ฐ๊ณผ ํ•˜๋ฅ˜ ์„ฑ๋Šฅ์ด ๋†’์€ ๊ฒƒ ์‚ฌ์ด์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ์ฃผ๋กœ ๊ด€์ฐฐํ•˜๋Š” ๋ฐ˜๋ฉด, ๋น„์˜์–ด ์ž‘์—…์€ ๊ธ์ •์  ๋˜๋Š” ๋ถ€์ •์  ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ๋ฌด์ž‘์œ„๋กœ ๋‚˜ํƒ€๋‚˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณด์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์–ธ์–ด๋ณ„ ๋‹ค๊ตญ์–ด ์ž‘์—…์˜ ์ˆ˜๊ฐ€ ์˜์–ด๋ณด๋‹ค ํ›จ์”ฌ ์ ๊ณ , XSQA ๋ฐ LAMBADA์™€ ๊ฐ™์€ ์—ฌ๋Ÿฌ ๋‹ค๊ตญ์–ด ์ž‘์—…์˜ ๊ฒฝ์šฐ ์˜์–ด ์ž‘์—…๊ณผ ๋ฒˆ์—ญ๋œ ๋ฒ„์ „ ๊ฐ„์— ์œ ์‚ฌํ•œ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
**Improvements Compared with SFT.** The results presented in Tab. 1 illustrate that our approach improves performance no matter whether augmentation data is used, demonstrating effectiveness across datasets of varying quality and under two different base models. Notably, this enhancement is most significant in simpler or smaller datasets where the available useful signals in the data may be limited. Our approach demonstrates enhancements in the performance of the MAMnoTH dataset, which incorporates Program-of-Thought Data. This suggests that our method remains effective and is not adversely affected by the inclusion of programming tasks.
**SFT์™€ ๋น„๊ตํ•œ ๊ฐœ์„  ์‚ฌํ•ญ.** ํ‘œ 1์— ์ œ์‹œ๋œ ๊ฒฐ๊ณผ๋Š” ์ฆ๊ฐ• ๋ฐ์ดํ„ฐ ์‚ฌ์šฉ ์—ฌ๋ถ€์— ๊ด€๊ณ„์—†์ด ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์ด ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚จ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๋‹ค์–‘ํ•œ ํ’ˆ์งˆ์˜ ๋ฐ์ดํ„ฐ์…‹๊ณผ ๋‘ ๊ฐ€์ง€ ๋‹ค๋ฅธ ๊ธฐ๋ณธ ๋ชจ๋ธ์—์„œ ํšจ๊ณผ์ ์ž„์„ ์ž…์ฆํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ, ๋ฐ์ดํ„ฐ ๋‚ด ์œ ์šฉํ•œ ์‹ ํ˜ธ๊ฐ€ ์ œํ•œ๋  ์ˆ˜ ์žˆ๋Š” ๋” ๊ฐ„๋‹จํ•˜๊ฑฐ๋‚˜ ์ž‘์€ ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ด ํ–ฅ์ƒ์ด ๊ฐ€์žฅ ๋‘๋“œ๋Ÿฌ์ง‘๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ Program-of-Thought Data๋ฅผ ํฌํ•จํ•˜๋Š” MAMnoTH ๋ฐ์ดํ„ฐ์…‹์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋Š” ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์ด ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์ž‘์—…์˜ ํฌํ•จ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์—ฌ์ „ํžˆ ํšจ๊ณผ์ ์ด๋ฉฐ ๋ถ€์ •์ ์ธ ์˜ํ–ฅ์„ ๋ฐ›์ง€ ์•Š๋Š”๋‹ค๋Š” ๊ฒƒ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
**SFT์™€ ๋น„๊ตํ•œ ๊ฐœ์„  ์‚ฌํ•ญ**. ํ‘œ 1์— ์ œ์‹œ๋œ ๊ฒฐ๊ณผ๋Š” ์ฆ๊ฐ• ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜๋“  ์‚ฌ์šฉํ•˜์ง€ ์•Š๋“  ์„ฑ๋Šฅ์ด ํ–ฅ์ƒ๋˜๋ฉฐ, ๋‹ค์–‘ํ•œ ํ’ˆ์งˆ์˜ ๋ฐ์ดํ„ฐ ์„ธํŠธ์™€ ๋‘ ๊ฐ€์ง€ ๋‹ค๋ฅธ ๊ธฐ๋ณธ ๋ชจ๋ธ์—์„œ ํšจ๊ณผ๋ฅผ ์ž…์ฆํ•˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํŠนํžˆ, ์ด๋Ÿฌํ•œ ํ–ฅ์ƒ์€ ๋ฐ์ดํ„ฐ์—์„œ ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ์œ ์šฉํ•œ ์‹ ํ˜ธ๊ฐ€ ์ œํ•œ์ ์ผ ์ˆ˜ ์žˆ๋Š” ๋‹จ์ˆœํ•˜๊ฑฐ๋‚˜ ์ž‘์€ ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ๊ฐ€์žฅ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ ์ ‘๊ทผ ๋ฐฉ์‹์€ ์ƒ๊ฐ์˜ ํ”„๋กœ๊ทธ๋žจ ๋ฐ์ดํ„ฐ๋ฅผ ํ†ตํ•ฉํ•œ MAMnoTH ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋Š” ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์ด ํšจ๊ณผ์ ์ด๋ฉฐ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์ž‘์—…์˜ ํฌํ•จ์œผ๋กœ ์ธํ•ด ๋ถ€์ •์ ์ธ ์˜ํ–ฅ์„ ๋ฐ›์ง€ ์•Š๋Š”๋‹ค๋Š” ๊ฒƒ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
While passively collecting and transcribing real-world dialogues from in-person clinical visits is feasible, two substantial challenges limit its effectiveness in training LLMs for medical conversations: (1) existing real-world data often fails to capture the vast range of medical conditions and scenarios, hindering its scalability and comprehensiveness; (2) the data derived from real-world dialogue transcripts tends to be noisy, containing ambiguous language (including slang, jargon, and sarcasm), interruptions, ungrammatical utterances, and implicit references. This in turn, may limit AMIE's knowledge, capabilities, and applicability.
์‹ค์ œ ์ž„์ƒ ๋ฐฉ๋ฌธ์—์„œ์˜ ๋Œ€ํ™”๋ฅผ ์ˆ˜๋™์œผ๋กœ ์ˆ˜์ง‘ํ•˜๊ณ  ์ „์‚ฌํ•˜๋Š” ๊ฒƒ์€ ๊ฐ€๋Šฅํ•˜์ง€๋งŒ, ์˜๋ฃŒ ๋Œ€ํ™”๋ฅผ ์œ„ํ•œ LLMs ํ›ˆ๋ จ์˜ ํšจ๊ณผ๋ฅผ ์ œํ•œํ•˜๋Š” ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ๋„์ „ ๊ณผ์ œ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค: (1) ๊ธฐ์กด์˜ ์‹ค์ œ ๋ฐ์ดํ„ฐ๋Š” ๋‹ค์–‘ํ•œ ์˜๋ฃŒ ์ƒํƒœ์™€ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ํฌ์ฐฉํ•˜์ง€ ๋ชปํ•ด ํ™•์žฅ์„ฑ๊ณผ ํฌ๊ด„์„ฑ์ด ์ €ํ•ด๋ฉ๋‹ˆ๋‹ค; (2) ์‹ค์ œ ๋Œ€ํ™” ์ „์‚ฌ์—์„œ ํŒŒ์ƒ๋œ ๋ฐ์ดํ„ฐ๋Š” ๋ชจํ˜ธํ•œ ์–ธ์–ด(์†์–ด, ์ „๋ฌธ ์šฉ์–ด, ํ’์ž ํฌํ•จ), ์ค‘๋‹จ, ๋น„๋ฌธ๋ฒ•์  ๋ฐœํ™” ๋ฐ ์•”์‹œ์  ์ฐธ์กฐ๋ฅผ ํฌํ•จํ•˜์—ฌ ์žก์Œ์ด ๋งŽ์€ ๊ฒฝํ–ฅ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๊ฒฐ๊ณผ์ ์œผ๋กœ AMIE์˜ ์ง€์‹, ๋Šฅ๋ ฅ ๋ฐ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ์ œํ•œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋Œ€๋ฉด ์ž„์ƒ ๋ฐฉ๋ฌธ์—์„œ ์‹ค์ œ ๋Œ€ํ™”๋ฅผ ์ˆ˜๋™์œผ๋กœ ์ˆ˜์ง‘ํ•˜๊ณ  ํ•„์‚ฌํ•˜๋Š” ๊ฒƒ์€ ๊ฐ€๋Šฅํ•˜์ง€๋งŒ, ์˜๋ฃŒ ๋Œ€ํ™”๋ฅผ ์œ„ํ•œ LLM์„ ํ›ˆ๋ จํ•˜๋Š” ๋ฐ ์žˆ์–ด ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ๊ณผ์ œ๊ฐ€ ๊ทธ ํšจ๊ณผ๋ฅผ ์ œํ•œํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค: (1) ๊ธฐ์กด ์‹ค์ œ ๋ฐ์ดํ„ฐ๋Š” ๊ด‘๋ฒ”์œ„ํ•œ ์˜๋ฃŒ ์กฐ๊ฑด๊ณผ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ํฌ์ฐฉํ•˜์ง€ ๋ชปํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์•„ ํ™•์žฅ์„ฑ๊ณผ ํฌ๊ด„์„ฑ์„ ์ €ํ•ดํ•ฉ๋‹ˆ๋‹ค; (2) ์‹ค์ œ ๋Œ€ํ™” ๋…น์ทจ๋ก์—์„œ ํŒŒ์ƒ๋œ ๋ฐ์ดํ„ฐ๋Š” ๋ชจํ˜ธํ•œ ์–ธ์–ด(์†์–ด, ์ „๋ฌธ ์šฉ์–ด, ํ’์ž ๋“ฑ), ์ค‘๋‹จ, ๋น„๋ฌธ๋ฒ•์  ๋ฐœํ™”, ์•”์‹œ์  ์ฐธ์กฐ ๋“ฑ ๋…ธ์ด์ฆˆ๊ฐ€ ๋งŽ์ด ํฌํ•จ๋˜์–ด ์žˆ๋Š” ๊ฒฝํ–ฅ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๊ฒฐ๊ตญ AMIE์˜ ์ง€์‹, ์—ญ๋Ÿ‰ ๋ฐ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ์ œํ•œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
To evaluate different configurations of data selection strategies, we train OPT [59] models from scratch on the pruned versions of datasets. We use the standard model architectures and settings of Zhang et al. [59] and use MetaSeq [59] to train all our models. For 125M models, we train to \(T_{target}=3B\) tokens. For 1.3B parameter models, we train to target token count of \(T_{target}=40B\). For 6.7B parameter models, we train to \(T_{target}=100B\) tokens. We choose these by trimming down the token budgets suggested by Hoffmann et al. [20] to meet our compute limitations. We provide full details of our training setup in Section A.1.
๋‹ค์–‘ํ•œ ๋ฐ์ดํ„ฐ ์„ ํƒ ์ „๋žต์˜ ๊ตฌ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ๋ฐ์ดํ„ฐ์…‹์˜ ์ถ•์†Œ๋œ ๋ฒ„์ „์—์„œ OPT [59] ๋ชจ๋ธ์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. Zhang et al. [59]์˜ ํ‘œ์ค€ ๋ชจ๋ธ ์•„ํ‚คํ…์ฒ˜์™€ ์„ค์ •์„ ์‚ฌ์šฉํ•˜๊ณ , ๋ชจ๋“  ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๊ธฐ ์œ„ํ•ด MetaSeq [59]๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. 125M ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, \(T_{target}=3B\) ํ† ํฐ๊นŒ์ง€ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. 1.3B ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, ๋ชฉํ‘œ ํ† ํฐ ์ˆ˜ \(T_{target}=40B\)๊นŒ์ง€ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. 6.7B ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, \(T_{target}=100B\) ํ† ํฐ๊นŒ์ง€ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” Hoffmann et al. [20]์ด ์ œ์•ˆํ•œ ํ† ํฐ ์˜ˆ์‚ฐ์„ ์šฐ๋ฆฌ์˜ ๊ณ„์‚ฐ ํ•œ๊ณ„์— ๋งž์ถ”๊ธฐ ์œ„ํ•ด ์ถ•์†Œํ•˜์—ฌ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ํ›ˆ๋ จ ์„ค์ •์— ๋Œ€ํ•œ ์ „์ฒด ์„ธ๋ถ€ ์‚ฌํ•ญ์€ ์„น์…˜ A.1์— ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
๋ฐ์ดํ„ฐ ์„ ํƒ ์ „๋žต์˜ ๋‹ค์–‘ํ•œ ๊ตฌ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ์ž˜๋ฆฐ ๋ฒ„์ „์— ๋Œ€ํ•ด ์ฒ˜์Œ๋ถ€ํ„ฐ OPT [59] ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. Zhang et al. [59]์˜ ํ‘œ์ค€ ๋ชจ๋ธ ์•„ํ‚คํ…์ฒ˜์™€ ์„ค์ •์„ ์‚ฌ์šฉํ•˜๋ฉฐ, ๋ชจ๋“  ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๊ธฐ ์œ„ํ•ด MetaSeq [59]๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. 125M ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, \(T_{target}=3B\) ํ† ํฐ์œผ๋กœ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. 13์–ต ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, \(T_{target}=40B\) ํ† ํฐ์˜ ๋ชฉํ‘œ ํ† ํฐ ์ˆ˜๋กœ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. 67์–ต ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, \(T_{target}=100B\) ํ† ํฐ์œผ๋กœ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์ปดํ“จํŒ… ์ œํ•œ ์‚ฌํ•ญ์„ ์ถฉ์กฑํ•˜๊ธฐ ์œ„ํ•ด Hoffmann ๋“ฑ[20]์ด ์ œ์•ˆํ•œ ํ† ํฐ ์˜ˆ์‚ฐ์„ ์กฐ์ •ํ•˜์—ฌ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค. ํ›ˆ๋ จ ์„ค์ •์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ์„น์…˜ A.1์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**Reddit Links.** Reddit is a social media platform that enables users to submit links and text posts, which can be voted on by others through "upvotes" or "downvotes". Highly upvoted posts are often considered useful, and can be utilized to create high-quality datasets. WebText [26] is a well-known corpus composed of highly upvoted links from Reddit, but it is not publicly available. As a surrogate, there is a readily accessible open-source alternative called OpenWebText [157]. Another corpus extracted from Reddit is PushShift.io [158], a real-time updated dataset that consists of historical data from Reddit since its creation day.
**Reddit ๋งํฌ.** Reddit์€ ์‚ฌ์šฉ์ž๊ฐ€ ๋งํฌ์™€ ํ…์ŠคํŠธ ๊ฒŒ์‹œ๋ฌผ์„ ์ œ์ถœํ•  ์ˆ˜ ์žˆ๋Š” ์†Œ์…œ ๋ฏธ๋””์–ด ํ”Œ๋žซํผ์œผ๋กœ, ๋‹ค๋ฅธ ์‚ฌ์šฉ์ž๋“ค์ด "์—…๋ณดํŠธ" ๋˜๋Š” "๋‹ค์šด๋ณดํŠธ"๋ฅผ ํ†ตํ•ด ํˆฌํ‘œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋†’์€ ์—…๋ณดํŠธ๋ฅผ ๋ฐ›์€ ๊ฒŒ์‹œ๋ฌผ์€ ์ข…์ข… ์œ ์šฉํ•˜๋‹ค๊ณ  ๊ฐ„์ฃผ๋˜๋ฉฐ, ๊ณ ํ’ˆ์งˆ ๋ฐ์ดํ„ฐ์…‹์„ ๋งŒ๋“œ๋Š” ๋ฐ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. WebText [26]๋Š” Reddit์—์„œ ๋†’์€ ์—…๋ณดํŠธ๋ฅผ ๋ฐ›์€ ๋งํฌ๋กœ ๊ตฌ์„ฑ๋œ ์ž˜ ์•Œ๋ ค์ง„ ์ฝ”ํผ์Šค์ด์ง€๋งŒ, ๊ณต๊ฐœ์ ์œผ๋กœ ์ด์šฉํ•  ์ˆ˜๋Š” ์—†์Šต๋‹ˆ๋‹ค. ๋Œ€์•ˆ์œผ๋กœ, OpenWebText [157]๋ผ๋Š” ์‰ฝ๊ฒŒ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋Š” ์˜คํ”ˆ ์†Œ์Šค ๋Œ€์ฒด๋ฌผ์ด ์žˆ์Šต๋‹ˆ๋‹ค. Reddit์—์„œ ์ถ”์ถœ๋œ ๋˜ ๋‹ค๋ฅธ ์ฝ”ํผ์Šค๋Š” PushShift.io [158]๋กœ, Reddit ์ƒ์„ฑ์ผ ์ดํ›„์˜ ์—ญ์‚ฌ์  ๋ฐ์ดํ„ฐ๋ฅผ ํฌํ•จํ•˜๋Š” ์‹ค์‹œ๊ฐ„ ์—…๋ฐ์ดํŠธ ๋ฐ์ดํ„ฐ์…‹์ž…๋‹ˆ๋‹ค.
**Reddit ๋งํฌ**. Reddit์€ ์‚ฌ์šฉ์ž๊ฐ€ ๋งํฌ์™€ ํ…์ŠคํŠธ ๊ฒŒ์‹œ๋ฌผ์„ ์ œ์ถœํ•  ์ˆ˜ ์žˆ๋Š” ์†Œ์…œ ๋ฏธ๋””์–ด ํ”Œ๋žซํผ์œผ๋กœ, ๋‹ค๋ฅธ ์‚ฌ์šฉ์ž๊ฐ€ "์—…๋ณดํŠธ" ๋˜๋Š” "๋‹ค์šด๋ณดํŠธ"๋ฅผ ํ†ตํ•ด ํˆฌํ‘œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์—…๋ณดํŠธ ์ˆ˜๊ฐ€ ๋†’์€ ๊ฒŒ์‹œ๋ฌผ์€ ์œ ์šฉํ•œ ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผ๋˜๋ฉฐ, ๊ณ ํ’ˆ์งˆ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๋งŒ๋“œ๋Š” ๋ฐ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์›นํ…์ŠคํŠธ[26]๋Š” Reddit์—์„œ ๋†’์€ ์—…๋ณดํŠธ ์ˆ˜๋ฅผ ๋ฐ›์€ ๋งํฌ๋กœ ๊ตฌ์„ฑ๋œ ์ž˜ ์•Œ๋ ค์ง„ ์ฝ”ํผ์Šค์ด์ง€๋งŒ ๊ณต๊ฐœ์ ์œผ๋กœ ์‚ฌ์šฉํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ๋Œ€๋ฆฌ ์ฝ”ํผ์Šค๋กœ๋Š” OpenWebText [157]๋ผ๋Š” ์‰ฝ๊ฒŒ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋Š” ์˜คํ”ˆ ์†Œ์Šค ๋Œ€์•ˆ์ด ์žˆ์Šต๋‹ˆ๋‹ค. Reddit์—์„œ ์ถ”์ถœํ•œ ๋˜ ๋‹ค๋ฅธ ์ฝ”ํผ์Šค๋Š” ์ƒ์„ฑ ๋‹น์‹œ๋ถ€ํ„ฐ Reddit์˜ ๊ณผ๊ฑฐ ๋ฐ์ดํ„ฐ๋กœ ๊ตฌ์„ฑ๋œ ์‹ค์‹œ๊ฐ„ ์—…๋ฐ์ดํŠธ ๋ฐ์ดํ„ฐ ์„ธํŠธ์ธ PushShift.io [158]์ž…๋‹ˆ๋‹ค.
set, we generated 256 responses and used the number of steps in the GSM8k test set's CoT annotations as the theoretical CoT steps. We draw the curve to show the relationship between the number of CoT reasoning steps and mean final answer accuracy and show the fitted curve based on Equation. 3. We test Xwin-Math-7B models with different synthetic data, and the results are shown in Figure 4. The solid line is fitted using all seven points and Table 4 shows the estimated single-step accuracy when using different amounts of data using all data points, and it can be seen that the single-step accuracy improve significantly with more data.
GSM8k ํ…Œ์ŠคํŠธ ์„ธํŠธ์˜ CoT ์ฃผ์„์—์„œ ์ด๋ก ์ ์ธ CoT ๋‹จ๊ณ„๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ 256๊ฐœ์˜ ์‘๋‹ต์„ ์ƒ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค. CoT ์ถ”๋ก  ๋‹จ๊ณ„ ์ˆ˜์™€ ํ‰๊ท  ์ตœ์ข… ๋‹ต๋ณ€ ์ •ํ™•๋„ ๊ฐ„์˜ ๊ด€๊ณ„๋ฅผ ๋ณด์—ฌ์ฃผ๊ธฐ ์œ„ํ•ด ๊ณก์„ ์„ ๊ทธ๋ ธ์œผ๋ฉฐ, Equation 3์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์ ํ•ฉ๋œ ๊ณก์„ ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋‹ค์–‘ํ•œ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Xwin-Math-7B ๋ชจ๋ธ์„ ํ…Œ์ŠคํŠธํ–ˆ์œผ๋ฉฐ, ๊ฒฐ๊ณผ๋Š” Figure 4์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์„ ์€ ๋ชจ๋“  7๊ฐœ์˜ ์ ์„ ์‚ฌ์šฉํ•˜์—ฌ ์ ํ•ฉ๋˜์—ˆ์œผ๋ฉฐ, Table 4๋Š” ๋ชจ๋“  ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ค๋ฅธ ์–‘์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ์˜ ๋‹จ์ผ ๋‹จ๊ณ„ ์ •ํ™•๋„๋ฅผ ์ถ”์ •ํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋” ๋งŽ์€ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ• ์ˆ˜๋ก ๋‹จ์ผ ๋‹จ๊ณ„ ์ •ํ™•๋„๊ฐ€ ํฌ๊ฒŒ ํ–ฅ์ƒ๋˜๋Š” ๊ฒƒ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ง‘ํ•ฉ์„ ์ƒ์„ฑํ•˜๊ณ  GSM8k ํ…Œ์ŠคํŠธ ์„ธํŠธ์˜ CoT ์ฃผ์„์— ์žˆ๋Š” ๋‹จ๊ณ„ ์ˆ˜๋ฅผ ์ด๋ก ์  CoT ๋‹จ๊ณ„๋กœ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. CoT ์ถ”๋ก  ๋‹จ๊ณ„ ์ˆ˜์™€ ์ตœ์ข… ๋‹ต ์ •ํ™•๋„ ํ‰๊ท  ์‚ฌ์ด์˜ ๊ด€๊ณ„๋ฅผ ๋ณด์—ฌ์ฃผ๊ธฐ ์œ„ํ•ด ์ปค๋ธŒ๋ฅผ ๊ทธ๋ฆฌ๊ณ  ๋ฐฉ์ •์‹ 3์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ”ผํŒ…๋œ ์ปค๋ธŒ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋กœ Xwin-Math-7B ๋ชจ๋ธ์„ ํ…Œ์ŠคํŠธํ–ˆ์œผ๋ฉฐ ๊ทธ ๊ฒฐ๊ณผ๋Š” ๊ทธ๋ฆผ 4์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์„ ์€ 7๊ฐœ์˜ ์ ์„ ๋ชจ๋‘ ์‚ฌ์šฉํ•˜์—ฌ ํ”ผํŒ…ํ•œ ๊ฒƒ์ด๋ฉฐ, ํ‘œ 4๋Š” ๋ชจ๋“  ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ค์–‘ํ•œ ์–‘์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ ์˜ˆ์ƒ๋˜๋Š” ๋‹จ์ผ ๋‹จ๊ณ„ ์ •ํ™•๋„๋ฅผ ๋ณด์—ฌ ์ฃผ๋ฉฐ, ๋” ๋งŽ์€ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ• ์ˆ˜๋ก ๋‹จ์ผ ๋‹จ๊ณ„ ์ •ํ™•๋„๊ฐ€ ํฌ๊ฒŒ ํ–ฅ์ƒ๋˜๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
1. The ducks lay 16 eggs per day. 2. She eats 3 for breakfast every morning. 3. She bakes muffins for her friends every day with 4. 4. She sells the remainder at the farmers' market daily for $2 per fresh duck egg. 5. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 6. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 7. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 8. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 9. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 10. The number of eggs she sells at the farmers' market is 16 - 3 - 4. 11.
1. ์˜ค๋ฆฌ๋Š” ํ•˜๋ฃจ์— 16๊ฐœ์˜ ์•Œ์„ ๋‚ณ์Šต๋‹ˆ๋‹ค. 2. ๊ทธ๋…€๋Š” ๋งค์ผ ์•„์นจ 3๊ฐœ์˜ ์•Œ์„ ์•„์นจ ์‹์‚ฌ๋กœ ๋จน์Šต๋‹ˆ๋‹ค. 3. ๊ทธ๋…€๋Š” ๋งค์ผ ์นœ๊ตฌ๋“ค์„ ์œ„ํ•ด 4๊ฐœ์˜ ์•Œ๋กœ ๋จธํ•€์„ ๊ตฝ์Šต๋‹ˆ๋‹ค. 4. ๊ทธ๋…€๋Š” ๋‚˜๋จธ์ง€ ์•Œ์„ ๋งค์ผ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ์‹ ์„ ํ•œ ์˜ค๋ฆฌ ์•Œ ํ•˜๋‚˜๋‹น 2๋‹ฌ๋Ÿฌ์— ํŒ๋งคํ•ฉ๋‹ˆ๋‹ค. 5. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 6. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 7. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 8. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 9. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 10. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 11.
1. ์˜ค๋ฆฌ๋Š” ํ•˜๋ฃจ์— 16๊ฐœ์˜ ์•Œ์„ ๋‚ณ์Šต๋‹ˆ๋‹ค. 2. ๊ทธ๋…€๋Š” ๋งค์ผ ์•„์นจ 3๊ฐœ๋ฅผ ๋จน์Šต๋‹ˆ๋‹ค. 3. ๊ทธ๋…€๋Š” ๋งค์ผ ์นœ๊ตฌ๋“ค์„ ์œ„ํ•ด ๋จธํ•€์„ ๊ตฝ์Šต๋‹ˆ๋‹ค. 4. 4. ๊ทธ๋…€๋Š” ๋งค์ผ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ์‹ ์„ ํ•œ ์˜ค๋ฆฌ์•Œ ํ•œ ๊ฐœ๋‹น 2๋‹ฌ๋Ÿฌ์— ๋‚จ์€ ์•Œ์„ ํŒ๋งคํ•ฉ๋‹ˆ๋‹ค. 5. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 6. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 7. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 8. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 9. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 10. ๊ทธ๋…€๊ฐ€ ๋†์‚ฐ๋ฌผ ์‹œ์žฅ์—์„œ ํŒ๋งคํ•˜๋Š” ์•Œ์˜ ์ˆ˜๋Š” 16 - 3 - 4์ž…๋‹ˆ๋‹ค. 11.
Figure 1: **Overview of Evolutionary Model Merge**. Our approach encompasses (1) evolving the weights for mixing parameters at each layer in the parameter space (PS); (2) evolving layer permutations in the data flow space (DFS); and (3) an integrated strategy that combines both methods for merging in both PS and DFS. Notice that merging in the PS is not simple copying and stitching of the layers parameters, but also mixes the weights. This merging is akin to blending colors as illustrated here (e.g., red and blue becomes purple). Note that we translated the questions to English for the reader; the models operate on Japanese text.
๊ทธ๋ฆผ 1: **์ง„ํ™”์  ๋ชจ๋ธ ๋ณ‘ํ•ฉ ๊ฐœ์š”**. ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ (1) ๋งค๊ฐœ๋ณ€์ˆ˜ ๊ณต๊ฐ„(PS)์—์„œ ๊ฐ ์ธต์˜ ํ˜ผํ•ฉ ๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์ง„ํ™”์‹œํ‚ค๋Š” ๊ฒƒ; (2) ๋ฐ์ดํ„ฐ ํ๋ฆ„ ๊ณต๊ฐ„(DFS)์—์„œ ์ธต ์ˆœ์—ด์„ ์ง„ํ™”์‹œํ‚ค๋Š” ๊ฒƒ; (3) PS์™€ DFS ๋ชจ๋‘์—์„œ ๋ณ‘ํ•ฉ์„ ์œ„ํ•œ ๋‘ ๊ฐ€์ง€ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ฉํ•˜๋Š” ์ „๋žต์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. PS์—์„œ์˜ ๋ณ‘ํ•ฉ์€ ๋‹จ์ˆœํžˆ ์ธต ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๋ณต์‚ฌํ•˜๊ณ  ์—ฐ๊ฒฐํ•˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ, ๊ฐ€์ค‘์น˜๋ฅผ ํ˜ผํ•ฉํ•˜๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์—ฌ๊ธฐ์„œ ์„ค๋ช…๋œ ๊ฒƒ์ฒ˜๋Ÿผ ์ƒ‰์ƒ์„ ํ˜ผํ•ฉํ•˜๋Š” ๊ฒƒ๊ณผ ์œ ์‚ฌํ•ฉ๋‹ˆ๋‹ค(์˜ˆ: ๋นจ๊ฐ„์ƒ‰๊ณผ ํŒŒ๋ž€์ƒ‰์ด ๋ณด๋ผ์ƒ‰์ด ๋˜๋Š” ๊ฒƒ). ๋…์ž๋ฅผ ์œ„ํ•ด ์งˆ๋ฌธ์„ ์˜์–ด๋กœ ๋ฒˆ์—ญํ–ˆ์Œ์„ ์œ ์˜ํ•˜์‹ญ์‹œ์˜ค; ๋ชจ๋ธ์€ ์ผ๋ณธ์–ด ํ…์ŠคํŠธ๋ฅผ ์ฒ˜๋ฆฌํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 1: **์ง„ํ™” ๋ชจ๋ธ ๋ณ‘ํ•ฉ ๊ฐœ์š”**. ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ (1) ํŒŒ๋ผ๋ฏธํ„ฐ ๊ณต๊ฐ„(PS)์—์„œ ๊ฐ ๋ ˆ์ด์–ด์˜ ๋ฏน์‹ฑ ํŒŒ๋ผ๋ฏธํ„ฐ์— ๋Œ€ํ•œ ๊ฐ€์ค‘์น˜๋ฅผ ์ง„ํ™”์‹œํ‚ค๋Š” ๊ฒƒ, (2) ๋ฐ์ดํ„ฐ ํ๋ฆ„ ๊ณต๊ฐ„(DFS)์—์„œ ๋ ˆ์ด์–ด ์ˆœ์—ด์„ ์ง„ํ™”์‹œํ‚ค๋Š” ๊ฒƒ, (3) PS์™€ DFS ๋ชจ๋‘์—์„œ ๋ณ‘ํ•ฉ์„ ์œ„ํ•ด ๋‘ ๊ฐ€์ง€ ๋ฐฉ๋ฒ•์„ ๊ฒฐํ•ฉํ•˜๋Š” ํ†ตํ•ฉ ์ „๋žต์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. PS์—์„œ ๋ณ‘ํ•ฉํ•˜๋Š” ๊ฒƒ์€ ๋‹จ์ˆœํžˆ ๋ ˆ์ด์–ด ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๋ณต์‚ฌํ•˜์—ฌ ๊ฟฐ๋งค๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ ๊ฐ€์ค‘์น˜๋„ ํ˜ผํ•ฉํ•œ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•˜์„ธ์š”. ์ด ๋ณ‘ํ•ฉ์€ ์—ฌ๊ธฐ์— ์˜ˆ์‹œ๋œ ๊ฒƒ์ฒ˜๋Ÿผ ์ƒ‰์ƒ์„ ํ˜ผํ•ฉํ•˜๋Š” ๊ฒƒ๊ณผ ์œ ์‚ฌํ•ฉ๋‹ˆ๋‹ค(์˜ˆ: ๋นจ๊ฐ„์ƒ‰๊ณผ ํŒŒ๋ž€์ƒ‰์ด ๋ณด๋ผ์ƒ‰์ด ๋จ). ๋…์ž๋ฅผ ์œ„ํ•ด ์งˆ๋ฌธ์„ ์˜์–ด๋กœ ๋ฒˆ์—ญํ–ˆ์ง€๋งŒ, ๋ชจ๋ธ์€ ์ผ๋ณธ์–ด ํ…์ŠคํŠธ๋กœ ์ž‘๋™ํ•ฉ๋‹ˆ๋‹ค.
code. Its features include providing detailed explanations of error messages, suggesting potential fixes, and ensuring the accuracy of the code. PENTESTGPT [125] is a penetration testing tool based on LLMs, which can effectively identify common vulnerabilities, and interpret source code to develop exploits. DB-GPT [41] utilizes the capabilities of LLMs to systematically assess potential root causes of anomalies in databases. Through the implementation of a tree of thought approach, DB-GPT enables LLMs to backtrack to previous steps in case the current step proves unsuccessful, thus enhancing the accuracy of the diagnosis process.
์ฝ”๋“œ. ์ด ๊ธฐ๋Šฅ์—๋Š” ์˜ค๋ฅ˜ ๋ฉ”์‹œ์ง€์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์„ค๋ช… ์ œ๊ณต, ์ž ์žฌ์  ์ˆ˜์ • ์ œ์•ˆ, ์ฝ”๋“œ์˜ ์ •ํ™•์„ฑ ๋ณด์žฅ์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. PENTESTGPT [125]๋Š” LLMs ๊ธฐ๋ฐ˜์˜ ์นจํˆฌ ํ…Œ์ŠคํŠธ ๋„๊ตฌ๋กœ, ์ผ๋ฐ˜์ ์ธ ์ทจ์•ฝ์ ์„ ํšจ๊ณผ์ ์œผ๋กœ ์‹๋ณ„ํ•˜๊ณ  ์†Œ์Šค ์ฝ”๋“œ๋ฅผ ํ•ด์„ํ•˜์—ฌ ์ต์Šคํ”Œ๋กœ์ž‡์„ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. DB-GPT [41]๋Š” LLMs์˜ ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์˜ ์ด์ƒ ํ˜„์ƒ์˜ ์ž ์žฌ์  ๊ทผ๋ณธ ์›์ธ์„ ์ฒด๊ณ„์ ์œผ๋กœ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์ƒ๊ฐ์˜ ๋‚˜๋ฌด ์ ‘๊ทผ ๋ฐฉ์‹์„ ๊ตฌํ˜„ํ•จ์œผ๋กœ์จ, DB-GPT๋Š” ํ˜„์žฌ ๋‹จ๊ณ„๊ฐ€ ์‹คํŒจํ•  ๊ฒฝ์šฐ ์ด์ „ ๋‹จ๊ณ„๋กœ ๋˜๋Œ์•„๊ฐˆ ์ˆ˜ ์žˆ์–ด ์ง„๋‹จ ๊ณผ์ •์˜ ์ •ํ™•์„ฑ์„ ๋†’์ž…๋‹ˆ๋‹ค.
์˜ค๋ฅ˜ ๋ฉ”์‹œ์ง€์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์„ค๋ช… ์ œ๊ณต, ์ž ์žฌ์ ์ธ ์ˆ˜์ • ์‚ฌํ•ญ ์ œ์•ˆ, ์ฝ”๋“œ์˜ ์ •ํ™•์„ฑ ๋ณด์žฅ ๋“ฑ์˜ ๊ธฐ๋Šฅ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. PENTESTGPT [125]๋Š” LLM์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์นจํˆฌ ํ…Œ์ŠคํŠธ ๋„๊ตฌ๋กœ, ์ผ๋ฐ˜์ ์ธ ์ทจ์•ฝ์ ์„ ํšจ๊ณผ์ ์œผ๋กœ ์‹๋ณ„ํ•˜๊ณ  ์†Œ์Šค ์ฝ”๋“œ๋ฅผ ํ•ด์„ํ•˜์—ฌ ์ต์Šคํ”Œ๋กœ์ž‡์„ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. DB-GPT [41]๋Š” LLM์˜ ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์—์„œ ์ด์ƒ ์ง•ํ›„์˜ ์ž ์žฌ์ ์ธ ๊ทผ๋ณธ ์›์ธ์„ ์ฒด๊ณ„์ ์œผ๋กœ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. DB-GPT๋Š” ์ƒ๊ฐ์˜ ํŠธ๋ฆฌ ์ ‘๊ทผ ๋ฐฉ์‹์„ ๊ตฌํ˜„ํ•˜์—ฌ LLM์ด ํ˜„์žฌ ๋‹จ๊ณ„๊ฐ€ ์‹คํŒจํ•œ ๊ฒฝ์šฐ ์ด์ „ ๋‹จ๊ณ„๋กœ ์—ญ์ถ”์ ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜์—ฌ ์ง„๋‹จ ํ”„๋กœ์„ธ์Šค์˜ ์ •ํ™•์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ต๋‹ˆ๋‹ค.
The strengths and weaknesses of different ethical theories can be evaluated based on their ability to address the ethical dilemma at hand. For example, deontological ethical theories may be strong in providing clear guidelines for moral decision making, but may not take into account the potential consequences of the action. Teleological ethical theories may be strong in their focus on outcomes, but may not consider the intentions or character of the person performing the action. Virtue ethical theories may be strong in promoting moral character, but may not provide clear guidelines for moral decision making in complex situations.
๋‹ค์–‘ํ•œ ์œค๋ฆฌ ์ด๋ก ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์€ ๋‹น๋ฉดํ•œ ์œค๋ฆฌ์  ๋”œ๋ ˆ๋งˆ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” ๋Šฅ๋ ฅ์— ๋”ฐ๋ผ ํ‰๊ฐ€๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์˜๋ฌด๋ก ์  ์œค๋ฆฌ ์ด๋ก ์€ ๋„๋•์  ์˜์‚ฌ๊ฒฐ์ •์„ ์œ„ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•˜๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ํ–‰๋™์˜ ์ž ์žฌ์  ๊ฒฐ๊ณผ๋ฅผ ๊ณ ๋ คํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ชฉ์ ๋ก ์  ์œค๋ฆฌ ์ด๋ก ์€ ๊ฒฐ๊ณผ์— ์ค‘์ ์„ ๋‘๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ํ–‰๋™์„ ์ˆ˜ํ–‰ํ•˜๋Š” ์‚ฌ๋žŒ์˜ ์˜๋„๋‚˜ ์„ฑ๊ฒฉ์„ ๊ณ ๋ คํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋• ์œค๋ฆฌ ์ด๋ก ์€ ๋„๋•์  ์„ฑํ’ˆ์„ ์ด‰์ง„ํ•˜๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ๋ณต์žกํ•œ ์ƒํ™ฉ์—์„œ ๋„๋•์  ์˜์‚ฌ๊ฒฐ์ •์„ ์œ„ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋‹ค๋ฅธ ์œค๋ฆฌ ์ด๋ก ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์€ ๋‹น๋ฉดํ•œ ์œค๋ฆฌ์  ๋”œ๋ ˆ๋งˆ๋ฅผ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์— ๋”ฐ๋ผ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์˜๋ฌด๋ก ์  ์œค๋ฆฌ ์ด๋ก ์€ ๋„๋•์  ์˜์‚ฌ ๊ฒฐ์ •์— ๋Œ€ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•˜๋Š” ๋ฐ ๊ฐ•ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ ํ–‰๋™์˜ ์ž ์žฌ์  ๊ฒฐ๊ณผ๋ฅผ ๊ณ ๋ คํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ชฉ์ ๋ก ์  ์œค๋ฆฌ ์ด๋ก ์€ ๊ฒฐ๊ณผ์— ์ดˆ์ ์„ ๋งž์ถ”๋Š” ๋ฐ ๊ฐ•ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ ํ–‰๋™์„ ์ˆ˜ํ–‰ํ•˜๋Š” ์‚ฌ๋žŒ์˜ ์˜๋„๋‚˜ ์„ฑ๊ฒฉ์„ ๊ณ ๋ คํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฏธ๋• ์œค๋ฆฌ ์ด๋ก ์€ ๋„๋•์  ์ธ์„ฑ์„ ์žฅ๋ คํ•˜๋Š” ๋ฐ ๊ฐ•ํ•  ์ˆ˜ ์žˆ์ง€๋งŒ ๋ณต์žกํ•œ ์ƒํ™ฉ์—์„œ ๋„๋•์  ์˜์‚ฌ ๊ฒฐ์ •์— ๋Œ€ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•˜์ง€ ๋ชปํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**Reference [1]** Ronaldo Assis Moreira (born 21 March 1980), commonly known as Ronaldinho Gaichco or simply Ronaldinho, is a Brazilian retired professional footballer who played mostly as an attacking midfielder, but was also deployed as a winger. Considered one of the best players of his generation and regarded by many as one of the greatest of all time, Ronaldinho won two FIFA World Player of the Year awards and a Ballon dโ€™Or. A global icon of the sport, he was renowned for his technical skills, creativity, dribbling ability and accuracy from free-kicks, as well as his use of tricks, feints, no-look passes and overhead...
**Reference [1]** Ronaldo Assis Moreira (1980๋…„ 3์›” 21์ผ ์ถœ์ƒ), ์ผ๋ฐ˜์ ์œผ๋กœ Ronaldinho Gaichco ๋˜๋Š” ๋‹จ์ˆœํžˆ Ronaldinho๋กœ ์•Œ๋ ค์ง„ ๊ทธ๋Š” ์ฃผ๋กœ ๊ณต๊ฒฉํ˜• ๋ฏธ๋“œํ•„๋”๋กœ ํ™œ์•ฝํ–ˆ์ง€๋งŒ ์œ™์–ด๋กœ๋„ ๋ฐฐ์น˜๋œ ๋ธŒ๋ผ์งˆ์˜ ์€ํ‡ดํ•œ ํ”„๋กœ ์ถ•๊ตฌ ์„ ์ˆ˜์ž…๋‹ˆ๋‹ค. ๊ทธ์˜ ์„ธ๋Œ€์—์„œ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ์—ฌ๊ฒจ์ง€๋ฉฐ, ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์— ์˜ํ•ด ์—ญ๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ํ‰๊ฐ€๋ฐ›๋Š” Ronaldinho๋Š” ๋‘ ๋ฒˆ์˜ FIFA ์˜ฌํ•ด์˜ ์„ ์ˆ˜์ƒ๊ณผ ๋ฐœ๋กฑ๋„๋ฅด๋ฅผ ์ˆ˜์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ์Šคํฌ์ธ ์˜ ๊ธ€๋กœ๋ฒŒ ์•„์ด์ฝ˜์œผ๋กœ์„œ ๊ทธ๋Š” ๊ธฐ์ˆ ์  ๋Šฅ๋ ฅ, ์ฐฝ์˜์„ฑ, ๋“œ๋ฆฌ๋ธ” ๋Šฅ๋ ฅ, ํ”„๋ฆฌํ‚ฅ ์ •ํ™•๋„, ๊ทธ๋ฆฌ๊ณ  ํŠธ๋ฆญ, ํŽ˜์ธํŠธ, ๋…ธ๋ฃฉ ํŒจ์Šค, ์˜ค๋ฒ„ํ—ค๋“œ ํ‚ฅ ๋“ฑ์˜ ์‚ฌ์šฉ์œผ๋กœ ์œ ๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
**์ฐธ๊ณ  [1]** ํ˜ธ๋‚˜์šฐ์ง€๋‰ด ๊ฐ€์ด์น˜์ฝ” ๋˜๋Š” ๊ฐ„๋‹จํžˆ ํ˜ธ๋‚˜์šฐ์ง€๋‰ด๋กœ ์•Œ๋ ค์ง„ ํ˜ธ๋‚˜์šฐ์ง€๋‰ด ์•„์‹œ์Šค ๋ชจ๋ ˆ์ด๋ผ(1980๋…„ 3์›” 21์ผ์ƒ)๋Š” ๋ธŒ๋ผ์งˆ์˜ ์€ํ‡ดํ•œ ํ”„๋กœ ์ถ•๊ตฌ ์„ ์ˆ˜๋กœ ์ฃผ๋กœ ๊ณต๊ฒฉํ˜• ๋ฏธ๋“œํ•„๋”๋กœ ํ™œ์•ฝํ–ˆ์ง€๋งŒ ์œ™์–ด๋กœ๋„ ํ™œ์•ฝํ–ˆ์Šต๋‹ˆ๋‹ค. ๋‹น๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ๊ผฝํžˆ๋ฉฐ ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์—ญ๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ๊ผฝ๋Š” ํ˜ธ๋‚˜์šฐ์ง€๋‰ด๋Š” FIFA ์˜ฌํ•ด์˜ ์„ธ๊ณ„ ์„ ์ˆ˜์ƒ์„ ๋‘ ์ฐจ๋ก€, ๋ฐœ๋กฑ๋„๋ฅด๋ฅผ ํ•œ ์ฐจ๋ก€ ์ˆ˜์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ์ถ•๊ตฌ์˜ ๊ธ€๋กœ๋ฒŒ ์•„์ด์ฝ˜์ธ ๊ทธ๋Š” ๊ธฐ์ˆ ๋ ฅ, ์ฐฝ์˜๋ ฅ, ๋“œ๋ฆฌ๋ธ” ๋Šฅ๋ ฅ, ํ”„๋ฆฌํ‚ฅ ์ •ํ™•๋„, ํŠธ๋ฆญ, ํŽ˜์ดํŠธ, ๋…ธ๋ฃฉ ํŒจ์Šค, ์˜ค๋ฒ„ํ—ค๋“œ ๋“ฑ์œผ๋กœ ์œ ๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
**Reference [1]** Ronaldo Assis Moreira (born 21 March 1980), commonly known as Ronaldinho Gaichco or simply Ronaldinho, is a Brazilian retired professional footballer who played mostly as an attacking midfielder, but was also deployed as a winger. Considered one of the best players of his generation and regarded by many as one of the greatest of all time, Ronaldinho won two FIFA World Player of the Year awards and a Ballon d'Or. A global icon of the sport, he was renowned for his technical skills, creativity, dribbling ability and accuracy from free-kicks, as well as his use of tricks, feints, no-look passes and overhead...
**Reference [1]** Ronaldo Assis Moreira (1980๋…„ 3์›” 21์ผ ์ถœ์ƒ), ์ผ๋ฐ˜์ ์œผ๋กœ Ronaldinho Gaichco ๋˜๋Š” ๋‹จ์ˆœํžˆ Ronaldinho๋กœ ์•Œ๋ ค์ง„ ๊ทธ๋Š” ๋ธŒ๋ผ์งˆ์˜ ์€ํ‡ดํ•œ ํ”„๋กœ ์ถ•๊ตฌ ์„ ์ˆ˜๋กœ ์ฃผ๋กœ ๊ณต๊ฒฉํ˜• ๋ฏธ๋“œํ•„๋”๋กœ ํ™œ์•ฝํ–ˆ์ง€๋งŒ, ์œ™์–ด๋กœ๋„ ๋ฐฐ์น˜๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ์˜ ์„ธ๋Œ€์—์„œ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ์—ฌ๊ฒจ์ง€๋ฉฐ, ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์— ์˜ํ•ด ์—ญ๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ํ‰๊ฐ€๋ฐ›๋Š” Ronaldinho๋Š” ๋‘ ๋ฒˆ์˜ FIFA ์˜ฌํ•ด์˜ ์„ ์ˆ˜์ƒ๊ณผ ํ•œ ๋ฒˆ์˜ Ballon d'Or์„ ์ˆ˜์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๊ธฐ์ˆ ์  ๋Šฅ๋ ฅ, ์ฐฝ์˜์„ฑ, ๋“œ๋ฆฌ๋ธ” ๋Šฅ๋ ฅ, ํ”„๋ฆฌํ‚ฅ ์ •ํ™•๋„, ๊ทธ๋ฆฌ๊ณ  ํŠธ๋ฆญ, ํŽ˜์ธํŠธ, ๋…ธ๋ฃฉ ํŒจ์Šค, ์˜ค๋ฒ„ํ—ค๋“œ ํ‚ฅ ๋“ฑ์˜ ์‚ฌ์šฉ์œผ๋กœ ์œ ๋ช…ํ•œ ๊ธ€๋กœ๋ฒŒ ์Šคํฌ์ธ  ์•„์ด์ฝ˜์ด์—ˆ์Šต๋‹ˆ๋‹ค.
**์ฐธ๊ณ  [1]** ํ˜ธ๋‚˜์šฐ์ง€๋‰ด ๊ฐ€์ด์น˜์ฝ” ๋˜๋Š” ๊ฐ„๋‹จํžˆ ํ˜ธ๋‚˜์šฐ์ง€๋‰ด๋กœ ์•Œ๋ ค์ง„ ํ˜ธ๋‚˜์šฐ์ง€๋‰ด ์•„์‹œ์Šค ๋ชจ๋ ˆ์ด๋ผ(1980๋…„ 3์›” 21์ผ์ƒ)๋Š” ๋ธŒ๋ผ์งˆ์˜ ์€ํ‡ดํ•œ ํ”„๋กœ ์ถ•๊ตฌ ์„ ์ˆ˜๋กœ ์ฃผ๋กœ ๊ณต๊ฒฉํ˜• ๋ฏธ๋“œํ•„๋”๋กœ ํ™œ์•ฝํ–ˆ์ง€๋งŒ ์œ™์–ด๋กœ๋„ ํ™œ์•ฝํ–ˆ์Šต๋‹ˆ๋‹ค. ๋‹น๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ๊ผฝํžˆ๋ฉฐ ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์—ญ๋Œ€ ์ตœ๊ณ ์˜ ์„ ์ˆ˜ ์ค‘ ํ•œ ๋ช…์œผ๋กœ ๊ผฝ๋Š” ํ˜ธ๋‚˜์šฐ์ง€๋‰ด๋Š” FIFA ์˜ฌํ•ด์˜ ์„ธ๊ณ„ ์„ ์ˆ˜์ƒ์„ ๋‘ ์ฐจ๋ก€, ๋ฐœ๋กฑ๋„๋ฅด๋ฅผ ํ•œ ์ฐจ๋ก€ ์ˆ˜์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ์ถ•๊ตฌ์˜ ๊ธ€๋กœ๋ฒŒ ์•„์ด์ฝ˜์ธ ๊ทธ๋Š” ๊ธฐ์ˆ ๋ ฅ, ์ฐฝ์˜๋ ฅ, ๋“œ๋ฆฌ๋ธ” ๋Šฅ๋ ฅ, ํ”„๋ฆฌํ‚ฅ ์ •ํ™•๋„, ํŠธ๋ฆญ, ํŽ˜์ดํŠธ, ๋…ธ๋ฃฉ ํŒจ์Šค, ์˜ค๋ฒ„ํ—ค๋“œ ๋“ฑ์œผ๋กœ ์œ ๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
**Patient Actor Ratings.** Figure 4 presents the various conversation qualities patient actors assessed following their consultations with the OSCE agents. Overall, AMIE's consultations were rated significantly better (\(p<0.05\)) by patient actors than those from PCPs across 24 of 26 axes. No significant differences in ratings were detected for the two PCCBP axes "Respecting Patient's Privacy" (N=108) and "Acknowledging Mistakes" (N=41). For the latter criterion, the number of exclusions was substantially higher since the question applied only when mistakes were made by the OSCE agent and pointed out in the conversation.
**ํ™˜์ž ๋ฐฐ์šฐ ํ‰๊ฐ€.** Figure 4๋Š” OSCE ์—์ด์ „ํŠธ์™€์˜ ์ƒ๋‹ด ํ›„ ํ™˜์ž ๋ฐฐ์šฐ๋“ค์ด ํ‰๊ฐ€ํ•œ ๋‹ค์–‘ํ•œ ๋Œ€ํ™” ํ’ˆ์งˆ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ „์ฒด์ ์œผ๋กœ, AMIE์˜ ์ƒ๋‹ด์€ 26๊ฐœ์˜ ์ถ• ์ค‘ 24๊ฐœ ์ถ•์—์„œ PCPs์˜ ์ƒ๋‹ด๋ณด๋‹ค ํ™˜์ž ๋ฐฐ์šฐ๋“ค์— ์˜ํ•ด ์œ ์˜๋ฏธํ•˜๊ฒŒ ๋” ๋†’์€ ํ‰๊ฐ€๋ฅผ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค (\(p<0.05\)). "ํ™˜์ž์˜ ํ”„๋ผ์ด๋ฒ„์‹œ ์กด์ค‘" (N=108)๊ณผ "์‹ค์ˆ˜ ์ธ์ •" (N=41)์ด๋ผ๋Š” ๋‘ PCCBP ์ถ•์—์„œ๋Š” ํ‰๊ฐ€์—์„œ ์œ ์˜๋ฏธํ•œ ์ฐจ์ด๊ฐ€ ๋ฐœ๊ฒฌ๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ํ›„์ž์˜ ๊ธฐ์ค€์—์„œ๋Š”, ์งˆ๋ฌธ์ด OSCE ์—์ด์ „ํŠธ๊ฐ€ ์‹ค์ˆ˜๋ฅผ ์ €์ง€๋ฅด๊ณ  ๋Œ€ํ™”์—์„œ ์ง€์ ๋œ ๊ฒฝ์šฐ์—๋งŒ ์ ์šฉ๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์— ์ œ์™ธ๋œ ์ˆ˜๊ฐ€ ์ƒ๋‹นํžˆ ๋งŽ์•˜์Šต๋‹ˆ๋‹ค.
**ํ™˜์ž ๋ฐฐ์šฐ ํ‰๊ฐ€.** ๊ทธ๋ฆผ 4๋Š” ํ™˜์ž ๋ฐฐ์šฐ๊ฐ€ OSCE ์—์ด์ „ํŠธ์™€์˜ ์ƒ๋‹ด ํ›„ ํ‰๊ฐ€ํ•œ ๋‹ค์–‘ํ•œ ๋Œ€ํ™” ํ’ˆ์งˆ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ „๋ฐ˜์ ์œผ๋กœ AMIE์˜ ์ƒ๋‹ด์€ 26๊ฐœ ์ถ• ์ค‘ 24๊ฐœ ์ถ•์—์„œ PCP๋ณด๋‹ค ํ™˜์ž ๋ฐฐ์šฐ๊ฐ€ ํ›จ์”ฌ ๋” ๋†’์€ ํ‰๊ฐ€๋ฅผ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค. ๋‘ PCCBP ์ถ•์ธ "ํ™˜์ž์˜ ํ”„๋ผ์ด๋ฒ„์‹œ ์กด์ค‘"(N=108)๊ณผ "์‹ค์ˆ˜ ์ธ์ •"(N=41)์— ๋Œ€ํ•ด์„œ๋Š” ์œ ์˜๋ฏธํ•œ ํ‰๊ฐ€ ์ฐจ์ด๊ฐ€ ๋ฐœ๊ฒฌ๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ํ›„์ž์˜ ๊ธฐ์ค€์˜ ๊ฒฝ์šฐ, ์ด ์งˆ๋ฌธ์€ OSCE ์—์ด์ „ํŠธ๊ฐ€ ์‹ค์ˆ˜๋ฅผ ๋ฒ”ํ•˜๊ณ  ๋Œ€ํ™”์—์„œ ์‹ค์ˆ˜๋ฅผ ์ง€์ ํ•œ ๊ฒฝ์šฐ์—๋งŒ ์ ์šฉ๋˜๊ธฐ ๋•Œ๋ฌธ์— ์ œ์™ธ๋œ ์ˆ˜๊ฐ€ ํ›จ์”ฌ ๋” ๋งŽ์•˜์Šต๋‹ˆ๋‹ค.
The simplest technique to find duplicate examples would be to perform exact string matching between all example pairs, but as we will show, this is insufficient. We introduce two complementary methods for performing deduplication. First, using a suffix array Manber and Myers (1993), we remove duplicate substrings from the dataset if they occur verbatim in more than one example. Second, we use MinHash Broder (1997), an efficient algorithm for estimating the \(n\)-gram similarity between all pairs of examples in a corpus, to remove entire examples from the dataset if they have high \(n\)-gram overlap with any other example.
์ค‘๋ณต ์˜ˆ์ œ๋ฅผ ์ฐพ๋Š” ๊ฐ€์žฅ ๊ฐ„๋‹จํ•œ ๋ฐฉ๋ฒ•์€ ๋ชจ๋“  ์˜ˆ์ œ ์Œ ์‚ฌ์ด์—์„œ ์ •ํ™•ํ•œ ๋ฌธ์ž์—ด ์ผ์น˜๋ฅผ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฒƒ์ด์ง€๋งŒ, ์šฐ๋ฆฌ๊ฐ€ ๋ณด์—ฌ์ค„ ๊ฒƒ์ฒ˜๋Ÿผ ์ด๋Š” ๋ถˆ์ถฉ๋ถ„ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ค‘๋ณต ์ œ๊ฑฐ๋ฅผ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•œ ๋‘ ๊ฐ€์ง€ ๋ณด์™„์ ์ธ ๋ฐฉ๋ฒ•์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ฒซ์งธ, Manber์™€ Myers (1993)์˜ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด์„ ์‚ฌ์šฉํ•˜์—ฌ, ํ•œ ์˜ˆ์ œ์—์„œ ๊ทธ๋Œ€๋กœ ์—ฌ๋Ÿฌ ๋ฒˆ ๋ฐœ์ƒํ•˜๋Š” ์ค‘๋ณต ๋ฌธ์ž์—ด์„ ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ๋‘˜์งธ, Broder (1997)์˜ MinHash๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ, ์ฝ”ํผ์Šค ๋‚ด ๋ชจ๋“  ์˜ˆ์ œ ์Œ ์‚ฌ์ด์˜ \(n\)-๊ทธ๋žจ ์œ ์‚ฌ์„ฑ์„ ํšจ์œจ์ ์œผ๋กœ ์ถ”์ •ํ•˜๊ณ , ๋‹ค๋ฅธ ์˜ˆ์ œ์™€ ๋†’์€ \(n\)-๊ทธ๋žจ ์ค‘๋ณต์„ ๊ฐ€์ง€๋Š” ์ „์ฒด ์˜ˆ์ œ๋ฅผ ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค.
์ค‘๋ณต ์˜ˆ์ œ๋ฅผ ์ฐพ๋Š” ๊ฐ€์žฅ ๊ฐ„๋‹จํ•œ ๋ฐฉ๋ฒ•์€ ๋ชจ๋“  ์˜ˆ์ œ ์Œ ๊ฐ„์— ์ •ํ™•ํ•œ ๋ฌธ์ž์—ด ์ผ์น˜๋ฅผ ์ˆ˜ํ–‰ํ•˜๋Š” ๊ฒƒ์ด์ง€๋งŒ, ๋‹ค์Œ์—์„œ ๋ณด์—ฌ๋“œ๋ฆฌ๋“ฏ์ด ์ด๊ฒƒ๋งŒ์œผ๋กœ๋Š” ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์ค‘๋ณต ์ œ๊ฑฐ๋ฅผ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•œ ๋‘ ๊ฐ€์ง€ ์ƒํ˜ธ ๋ณด์™„์ ์ธ ๋ฐฉ๋ฒ•์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ฒซ์งธ, ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด Manber์™€ Myers(1993)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์—์„œ ๋‘ ๊ฐœ ์ด์ƒ์˜ ์˜ˆ์ œ์—์„œ ์ถ•์–ด์ ์œผ๋กœ ๋ฐœ์ƒํ•˜๋Š” ์ค‘๋ณต ํ•˜์œ„ ๋ฌธ์ž์—ด์„ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ๋‘˜์งธ, ๋ง๋ญ‰์น˜์˜ ๋ชจ๋“  ์˜ˆ์ œ ์Œ ๊ฐ„์˜ \(n\)-๊ทธ๋žจ ์œ ์‚ฌ์„ฑ์„ ์ถ”์ •ํ•˜๋Š” ํšจ์œจ์ ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ธ MinHash Broder(1997)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ค๋ฅธ ์˜ˆ์ œ์™€ \(n\)-๊ทธ๋žจ์ด ๋งŽ์ด ๊ฒน์น˜๋Š” ๊ฒฝ์šฐ ์ „์ฒด ์˜ˆ์ œ๋ฅผ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์—์„œ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค.
**Data** we consider the high-quality and human-written FLORES-200 dataset (NLLB TEAM et al., 2022), comprising both development and test data, amounting to a total of 2009 samples for each language direction, to compare the gold references with the outputs generated by the models. We employed ALMA-13B-LoRA and GPT-4 to perform translations across five English-centric language pairs, covering both translations from and to English. These pairs include German (de), Czech (cs), Icelandic (is), Chinese (zh), and Russian (ru), with Icelandic (is) categorized as a low-resource language and the others as high-resource languages.
**๋ฐ์ดํ„ฐ** ์šฐ๋ฆฌ๋Š” ๊ณ ํ’ˆ์งˆ์˜ ์ธ๊ฐ„์ด ์ž‘์„ฑํ•œ FLORES-200 ๋ฐ์ดํ„ฐ์…‹(NLLB TEAM et al., 2022)์„ ๊ณ ๋ คํ•˜๋ฉฐ, ๊ฐœ๋ฐœ ๋ฐ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ๋ฅผ ํฌํ•จํ•˜์—ฌ ๊ฐ ์–ธ์–ด ๋ฐฉํ–ฅ์— ๋Œ€ํ•ด ์ด 2009๊ฐœ์˜ ์ƒ˜ํ”Œ์„ ๋น„๊ตํ•˜์—ฌ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ์ถœ๋ ฅ๊ณผ ๊ธˆ ํ‘œ์ค€ ์ฐธ์กฐ๋ฅผ ๋น„๊ตํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ALMA-13B-LoRA์™€ GPT-4๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์˜์–ด ์ค‘์‹ฌ์˜ ๋‹ค์„ฏ ๊ฐ€์ง€ ์–ธ์–ด ์Œ์— ๋Œ€ํ•ด ๋ฒˆ์—ญ์„ ์ˆ˜ํ–‰ํ–ˆ์œผ๋ฉฐ, ์—ฌ๊ธฐ์—๋Š” ์˜์–ด๋กœ๋ถ€ํ„ฐ์˜ ๋ฒˆ์—ญ๊ณผ ์˜์–ด๋กœ์˜ ๋ฒˆ์—ญ์ด ๋ชจ๋‘ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ์ด ์–ธ์–ด ์Œ์—๋Š” ๋…์ผ์–ด(de), ์ฒด์ฝ”์–ด(cs), ์•„์ด์Šฌ๋ž€๋“œ์–ด(is), ์ค‘๊ตญ์–ด(zh), ๋Ÿฌ์‹œ์•„์–ด(ru)๊ฐ€ ํฌํ•จ๋˜๋ฉฐ, ์•„์ด์Šฌ๋ž€๋“œ์–ด(is)๋Š” ์ €์ž์› ์–ธ์–ด๋กœ ๋ถ„๋ฅ˜๋˜๊ณ  ๋‚˜๋จธ์ง€๋Š” ๊ณ ์ž์› ์–ธ์–ด๋กœ ๋ถ„๋ฅ˜๋ฉ๋‹ˆ๋‹ค.
**๋ฐ์ดํ„ฐ** ๊ณ ํ’ˆ์งˆ์˜ ์ธ๊ฐ„์ด ์ž‘์„ฑํ•œ FLORES-200 ๋ฐ์ดํ„ฐ ์„ธํŠธ(NLLB TEAM, 2022)๋ฅผ ๊ณ ๋ คํ•˜๋ฉฐ, ๊ฐœ๋ฐœ ๋ฐ ํ…Œ์ŠคํŠธ ๋ฐ์ดํ„ฐ๊ฐ€ ๋ชจ๋‘ ํฌํ•จ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๊ฐ ์–ธ์–ด ๋ฐฉํ–ฅ์— ๋Œ€ํ•ด ์ด 2009๊ฐœ์˜ ์ƒ˜ํ”Œ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ธˆ๋ณธ ์ฐธ์กฐ์™€ ๋ชจ๋ธ์—์„œ ์ƒ์„ฑ๋œ ๊ฒฐ๊ณผ๋ฅผ ๋น„๊ตํ•˜๊ธฐ ์œ„ํ•ด ๊ณ ๋ คํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜์–ด ์ค‘์‹ฌ์˜ 5๊ฐœ ์–ธ์–ด ์Œ์— ๋Œ€ํ•ด ์˜์–ด์—์„œ ์˜์–ด๋กœ์˜ ๋ฒˆ์—ญ์„ ๋ชจ๋‘ ํฌํ•จํ•˜์—ฌ ALMA-13B-LoRA์™€ GPT-4๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฒˆ์—ญ์„ ์ˆ˜ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์Œ์—๋Š” ๋…์ผ์–ด(de), ์ฒด์ฝ”์–ด(cs), ์•„์ด์Šฌ๋ž€๋“œ์–ด(is), ์ค‘๊ตญ์–ด(zh), ๋Ÿฌ์‹œ์•„์–ด(ru)๊ฐ€ ํฌํ•จ๋˜๋ฉฐ, ์•„์ด์Šฌ๋ž€๋“œ์–ด(is)๋Š” ์ €์ž์› ์–ธ์–ด๋กœ ๋ถ„๋ฅ˜๋˜๊ณ  ๋‹ค๋ฅธ ์–ธ์–ด๋Š” ๊ณ ์ž์› ์–ธ์–ด๋กœ ๋ถ„๋ฅ˜๋ฉ๋‹ˆ๋‹ค.
Figure 4: **Evolved Configurations for DFS Merging of models \(A\) and \(B\).** The three figures depict the evolution of the inference path on the MGSM-JA task. The y-axis represents the layer index \(l\in[1,M]\), and the x-axis corresponds to the path index \(t\in[1,T]\). Blue markers indicate path steps utilizing layers from model \(A\), and red markers denotes those from \(B\). Marker size reflects the magnitude of the scaling factor \(W_{ij}\). The evolutionary search result includes most layers in \(A\) at an early stage and then alternates between layers from both models. This result is from our 10B model (PS+DFS).
Figure 4: **๋ชจ๋ธ \(A\)์™€ \(B\)์˜ DFS ๋ณ‘ํ•ฉ์„ ์œ„ํ•œ ์ง„ํ™”๋œ ๊ตฌ์„ฑ.** ์„ธ ๊ทธ๋ฆผ์€ MGSM-JA ์ž‘์—…์—์„œ ์ถ”๋ก  ๊ฒฝ๋กœ์˜ ์ง„ํ™”๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. y์ถ•์€ ๋ ˆ์ด์–ด ์ธ๋ฑ์Šค \(l\in[1,M]\)๋ฅผ ๋‚˜ํƒ€๋‚ด๊ณ , x์ถ•์€ ๊ฒฝ๋กœ ์ธ๋ฑ์Šค \(t\in[1,T]\)์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค. ํŒŒ๋ž€์ƒ‰ ๋งˆ์ปค๋Š” ๋ชจ๋ธ \(A\)์˜ ๋ ˆ์ด์–ด๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ๋กœ ๋‹จ๊ณ„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๋นจ๊ฐ„์ƒ‰ ๋งˆ์ปค๋Š” ๋ชจ๋ธ \(B\)์˜ ๋ ˆ์ด์–ด๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ๋งˆ์ปค ํฌ๊ธฐ๋Š” ์Šค์ผ€์ผ๋ง ํŒฉํ„ฐ \(W_{ij}\)์˜ ํฌ๊ธฐ๋ฅผ ๋ฐ˜์˜ํ•ฉ๋‹ˆ๋‹ค. ์ง„ํ™”์  ๊ฒ€์ƒ‰ ๊ฒฐ๊ณผ๋Š” ์ดˆ๊ธฐ ๋‹จ๊ณ„์—์„œ ๋Œ€๋ถ€๋ถ„ ๋ชจ๋ธ \(A\)์˜ ๋ ˆ์ด์–ด๋ฅผ ํฌํ•จํ•˜๊ณ , ์ดํ›„ ๋‘ ๋ชจ๋ธ์˜ ๋ ˆ์ด์–ด๋ฅผ ๋ฒˆ๊ฐˆ์•„ ๊ฐ€๋ฉฐ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ฒฐ๊ณผ๋Š” ์šฐ๋ฆฌ์˜ 10B ๋ชจ๋ธ (PS+DFS)์—์„œ ๋‚˜์˜จ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 4: **๋ชจ๋ธ \(A\)์™€ \(B\)์˜ DFS ๋ณ‘ํ•ฉ์„ ์œ„ํ•œ ์ง„ํ™”๋œ ๊ตฌ์„ฑ.** ์„ธ ๊ฐœ์˜ ๊ทธ๋ฆผ์€ MGSM-JA ์ž‘์—…์—์„œ ์ถ”๋ก  ๊ฒฝ๋กœ์˜ ์ง„ํ™”๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. Y์ถ•์€ ๋ ˆ์ด์–ด ์ธ๋ฑ์Šค \(l\in[1,M]\), X์ถ•์€ ๊ฒฝ๋กœ ์ธ๋ฑ์Šค \(t\in[1,T]\). ํŒŒ๋ž€์ƒ‰ ๋งˆ์ปค๋Š” ๋ชจ๋ธ \(A\)์˜ ๋ ˆ์ด์–ด๋ฅผ ํ™œ์šฉํ•˜๋Š” ๊ฒฝ๋กœ ๋‹จ๊ณ„๋ฅผ ๋‚˜ํƒ€๋‚ด๊ณ  ๋นจ๊ฐ„์ƒ‰ ๋งˆ์ปค๋Š” \(B\)์˜ ๋ ˆ์ด์–ด๋ฅผ ํ™œ์šฉํ•˜๋Š” ๊ฒฝ๋กœ ๋‹จ๊ณ„๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ๋งˆ์ปค ํฌ๊ธฐ๋Š” ์Šค์ผ€์ผ๋ง ๊ณ„์ˆ˜ \(W_{ij}\)์˜ ํฌ๊ธฐ๋ฅผ ๋ฐ˜์˜ํ•ฉ๋‹ˆ๋‹ค. ์ง„ํ™” ๊ฒ€์ƒ‰ ๊ฒฐ๊ณผ๋Š” ์ดˆ๊ธฐ ๋‹จ๊ณ„์—์„œ ๋Œ€๋ถ€๋ถ„์˜ ๋ ˆ์ด์–ด๋ฅผ \(A\)์— ํฌํ•จ์‹œํ‚จ ๋‹ค์Œ ๋‘ ๋ชจ๋ธ์˜ ๋ ˆ์ด์–ด๋ฅผ ๋ฒˆ๊ฐˆ์•„ ๊ฐ€๋ฉฐ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ฒฐ๊ณผ๋Š” 10B ๋ชจ๋ธ(PS+DFS)์—์„œ ์–ป์€ ๊ฒฐ๊ณผ์ž…๋‹ˆ๋‹ค.
A notable feature distinguishing the MiniCPM series from other robust SLMs like Phi (Li et al., 2023) by Microsoft is that we train MiniCPM on general-purpose corpora, which guarantees the models' versatility and generalization across a variety of tasks. In this section, we present several intriguing generation instances, typically achievable only by larger models, to demonstrate the capabilities of MiniCPM. Although we do not conduct extensive checks on whether specific cases exist in the training set, due to the diverse ways they can be expressed, we ensure that no data similar to the test cases is intentionally added.
MiniCPM ์‹œ๋ฆฌ์ฆˆ๋ฅผ Microsoft์˜ Phi (Li et al., 2023)์™€ ๊ฐ™์€ ๋‹ค๋ฅธ ๊ฐ•๋ ฅํ•œ SLM๊ณผ ๊ตฌ๋ณ„ํ•˜๋Š” ์ฃผ๋ชฉํ•  ๋งŒํ•œ ํŠน์ง•์€ MiniCPM์„ ๋ฒ”์šฉ ์ฝ”ํผ์Šค์—์„œ ํ›ˆ๋ จ์‹œํ‚จ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์˜ ๋‹ค์–‘ํ•œ ์ž‘์—…์— ๋Œ€ํ•œ ๋‹ค์žฌ๋‹ค๋Šฅ์„ฑ๊ณผ ์ผ๋ฐ˜ํ™”๋ฅผ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ๋” ํฐ ๋ชจ๋ธ์—์„œ๋งŒ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋Š” ์—ฌ๋Ÿฌ ํฅ๋ฏธ๋กœ์šด ์ƒ์„ฑ ์‚ฌ๋ก€๋ฅผ ์ œ์‹œํ•˜์—ฌ MiniCPM์˜ ๋Šฅ๋ ฅ์„ ์ž…์ฆํ•ฉ๋‹ˆ๋‹ค. ํŠน์ • ์‚ฌ๋ก€๊ฐ€ ํ›ˆ๋ จ ์„ธํŠธ์— ์กด์žฌํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ๊ด‘๋ฒ”์œ„ํ•œ ๊ฒ€์‚ฌ๋ฅผ ์ˆ˜ํ–‰ํ•˜์ง€๋Š” ์•Š์ง€๋งŒ, ๋‹ค์–‘ํ•œ ํ‘œํ˜„ ๋ฐฉ์‹ ๋•Œ๋ฌธ์— ํ…Œ์ŠคํŠธ ์‚ฌ๋ก€์™€ ์œ ์‚ฌํ•œ ๋ฐ์ดํ„ฐ๊ฐ€ ์˜๋„์ ์œผ๋กœ ์ถ”๊ฐ€๋˜์ง€ ์•Š๋„๋ก ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
Microsoft์˜ Phi(Li et al., 2023)์™€ ๊ฐ™์€ ๋‹ค๋ฅธ ๊ฐ•๋ ฅํ•œ SLM๊ณผ MiniCPM ์‹œ๋ฆฌ์ฆˆ๋ฅผ ๊ตฌ๋ณ„ํ•˜๋Š” ์ฃผ๋ชฉํ•  ๋งŒํ•œ ํŠน์ง•์€ ๋ฒ”์šฉ ์ฝ”ํผ๋ผ์— ๋Œ€ํ•ด MiniCPM์„ ํ›ˆ๋ จํ•˜์—ฌ ๋‹ค์–‘ํ•œ ์ž‘์—…์—์„œ ๋ชจ๋ธ์˜ ๋‹ค์šฉ๋„์„ฑ๊ณผ ์ผ๋ฐ˜ํ™”๋ฅผ ๋ณด์žฅํ•œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” MiniCPM์˜ ๊ธฐ๋Šฅ์„ ๋ณด์—ฌ์ฃผ๊ธฐ ์œ„ํ•ด ์ผ๋ฐ˜์ ์œผ๋กœ ๋” ํฐ ๋ชจ๋ธ์—์„œ๋งŒ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋Š” ๋ช‡ ๊ฐ€์ง€ ํฅ๋ฏธ๋กœ์šด ์ƒ์„ฑ ์‚ฌ๋ก€๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ํŠน์ • ์‚ฌ๋ก€๊ฐ€ ํ›ˆ๋ จ ์„ธํŠธ์— ์กด์žฌํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ๊ด‘๋ฒ”์œ„ํ•œ ๊ฒ€์‚ฌ๋Š” ์ˆ˜ํ–‰ํ•˜์ง€ ์•Š์ง€๋งŒ, ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์œผ๋กœ ํ‘œํ˜„๋  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์— ํ…Œ์ŠคํŠธ ์‚ฌ๋ก€์™€ ์œ ์‚ฌํ•œ ๋ฐ์ดํ„ฐ๊ฐ€ ์˜๋„์ ์œผ๋กœ ์ถ”๊ฐ€๋˜์ง€ ์•Š์•˜๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.
Using PLND, we feed a corpus in a specific language to LLMs and identify the neurons that are consistently activated. These language-specific neurons are responsible for processing queries in that language. To ascertain whether these neurons are genuinely language-specific, we assess the performance of LLMs in corresponding languages when these neurons are deactivated versus when the same number of randomly sampled neurons is deactivated. We utilize the summarization task with XLSum dataset Hasan et al. (2021) to showcase the capabilities of LLMs, aiming to demonstrate that the resulting output can be devoid of meaning.
PLND๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํŠน์ • ์–ธ์–ด์˜ ์ฝ”ํผ์Šค๋ฅผ LLM์— ์ž…๋ ฅํ•˜๊ณ  ์ผ๊ด€๋˜๊ฒŒ ํ™œ์„ฑํ™”๋˜๋Š” ๋‰ด๋Ÿฐ์„ ์‹๋ณ„ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์–ธ์–ด ํŠนํ™” ๋‰ด๋Ÿฐ์€ ํ•ด๋‹น ์–ธ์–ด์˜ ์ฟผ๋ฆฌ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋‰ด๋Ÿฐ์ด ์ง„์ •์œผ๋กœ ์–ธ์–ด ํŠนํ™”์ธ์ง€ ํ™•์ธํ•˜๊ธฐ ์œ„ํ•ด, ํ•ด๋‹น ๋‰ด๋Ÿฐ์ด ๋น„ํ™œ์„ฑํ™”๋˜์—ˆ์„ ๋•Œ์™€ ๋™์ผํ•œ ์ˆ˜์˜ ๋ฌด์ž‘์œ„๋กœ ์ƒ˜ํ”Œ๋ง๋œ ๋‰ด๋Ÿฐ์ด ๋น„ํ™œ์„ฑํ™”๋˜์—ˆ์„ ๋•Œ์˜ LLM ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” XLSum ๋ฐ์ดํ„ฐ์…‹ Hasan et al. (2021)์„ ์‚ฌ์šฉํ•œ ์š”์•ฝ ์ž‘์—…์„ ํ†ตํ•ด LLM์˜ ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๊ฒฐ๊ณผ ์ถœ๋ ฅ์ด ์˜๋ฏธ๋ฅผ ์žƒ์„ ์ˆ˜ ์žˆ์Œ์„ ์ž…์ฆํ•˜๋ ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค.
PLND๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํŠน์ • ์–ธ์–ด์˜ ๋ง๋ญ‰์น˜๋ฅผ LLM์— ๊ณต๊ธ‰ํ•˜๊ณ  ์ผ๊ด€๋˜๊ฒŒ ํ™œ์„ฑํ™”๋˜๋Š” ๋‰ด๋Ÿฐ์„ ์‹๋ณ„ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์–ธ์–ด๋ณ„ ๋‰ด๋Ÿฐ์€ ํ•ด๋‹น ์–ธ์–ด๋กœ ๋œ ์ฟผ๋ฆฌ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋‰ด๋Ÿฐ์ด ์ง„์ •์œผ๋กœ ์–ธ์–ด๋ณ„์ธ์ง€ ํ™•์ธํ•˜๊ธฐ ์œ„ํ•ด ์ด๋Ÿฌํ•œ ๋‰ด๋Ÿฐ์„ ๋น„ํ™œ์„ฑํ™”ํ–ˆ์„ ๋•Œ์™€ ๋ฌด์ž‘์œ„๋กœ ์ƒ˜ํ”Œ๋งํ•œ ๋™์ผํ•œ ์ˆ˜์˜ ๋‰ด๋Ÿฐ์„ ๋น„ํ™œ์„ฑํ™”ํ–ˆ์„ ๋•Œ ํ•ด๋‹น ์–ธ์–ด์—์„œ LLM์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. XLSum ๋ฐ์ดํ„ฐ ์„ธํŠธ Hasan et al. (2021)์˜ ์š”์•ฝ ์ž‘์—…์„ ํ™œ์šฉํ•˜์—ฌ LLM์˜ ๊ธฐ๋Šฅ์„ ๋ณด์—ฌ์ฃผ๊ณ , ๊ทธ ๊ฒฐ๊ณผ๋ฌผ์ด ์˜๋ฏธ๊ฐ€ ์—†์„ ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์ฃผ๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
Large Language Models.LLMs have achieved substantial advancements within the realm of Natural Language Processing (NLP), providing a valuable and task-agnostic foundation for widespread applications. These models typically encompass parameter counts reaching into the hundreds of billions, which are trained on extensive large-scale corpuses of textual data. The prominent instancesentail OpenAI's GPT3&4[3, 51], Anthropic's Claude7, Google's PaLM [7, 44], Bard11, DeepMind's Chinchilla [47], and Gopher [52]. However none of them have been open-sourced so far, and some of them can only be exclusively accessible through APIs.
๋Œ€ํ˜• ์–ธ์–ด ๋ชจ๋ธ(LLM)์€ ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ(NLP) ๋ถ„์•ผ์—์„œ ์ƒ๋‹นํ•œ ๋ฐœ์ „์„ ์ด๋ฃจ์–ด๋‚ด๋ฉฐ, ๊ด‘๋ฒ”์œ„ํ•œ ์‘์šฉ์„ ์œ„ํ•œ ์œ ์šฉํ•˜๊ณ  ์ž‘์—…์— ๊ตฌ์• ๋ฐ›์ง€ ์•Š๋Š” ๊ธฐ์ดˆ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ชจ๋ธ๋“ค์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์ˆ˜๋ฐฑ์–ต ๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ํฌํ•จํ•˜๋ฉฐ, ๋ฐฉ๋Œ€ํ•œ ๋Œ€๊ทœ๋ชจ ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ฝ”ํผ์Šค์—์„œ ํ›ˆ๋ จ๋ฉ๋‹ˆ๋‹ค. ๋Œ€ํ‘œ์ ์ธ ์˜ˆ๋กœ๋Š” OpenAI์˜ GPT-3 ๋ฐ GPT-4 [3, 51], Anthropic์˜ Claude7, Google์˜ PaLM [7, 44], Bard11, DeepMind์˜ Chinchilla [47], ๊ทธ๋ฆฌ๊ณ  Gopher [52]๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋“ค ์ค‘ ์–ด๋Š ๊ฒƒ๋„ ์•„์ง ์˜คํ”ˆ ์†Œ์Šค๋กœ ๊ณต๊ฐœ๋˜์ง€ ์•Š์•˜์œผ๋ฉฐ, ์ผ๋ถ€๋Š” API๋ฅผ ํ†ตํ•ด์„œ๋งŒ ๋…์ ์ ์œผ๋กœ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ.LLM์€ ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ(NLP) ์˜์—ญ์—์„œ ์ƒ๋‹นํ•œ ๋ฐœ์ „์„ ์ด๋ฃจ์–ด ํญ๋„“์€ ์‘์šฉ ๋ถ„์•ผ์— ์œ ์šฉํ•˜๊ณ  ์ž‘์—…์— ๊ตฌ์• ๋ฐ›์ง€ ์•Š๋Š” ๊ธฐ๋ฐ˜์„ ์ œ๊ณตํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ชจ๋ธ์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์ˆ˜์ฒœ์–ต ๊ฐœ์— ๋‹ฌํ•˜๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜๋ฅผ ํฌํ•จํ•˜๋ฉฐ, ๋Œ€๊ทœ๋ชจ ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ ์ฝ”ํผ์Šค๋ฅผ ๋Œ€์ƒ์œผ๋กœ ํ•™์Šตํ•ฉ๋‹ˆ๋‹ค. ๋Œ€ํ‘œ์ ์ธ ์‚ฌ๋ก€๋กœ๋Š” OpenAI์˜ GPT3&4[3, 51], Anthropic์˜ Claude7, Google์˜ PaLM [7, 44], Bard11, DeepMind์˜ Chinchilla [47], Gopher [52] ๋“ฑ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ง€๊ธˆ๊นŒ์ง€ ์˜คํ”ˆ ์†Œ์Šคํ™”๋œ ์‚ฌ๋ก€๋Š” ์—†์œผ๋ฉฐ, ์ผ๋ถ€๋Š” API๋ฅผ ํ†ตํ•ด์„œ๋งŒ ๋…์ ์ ์œผ๋กœ ์•ก์„ธ์Šคํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Property 3: _The performance increases with the prior probability._ We investigate the influence of prior probability on performance by tuning the parameter \(K\), while maintaining constant values for \(I\) and \(K\). As \(K\) represents the number of intervals, the prior probability is defined as \(1/K\). We vary \(K\) from \(4\) to \(32\), which equivalently alters the prior probability from \(1/4\) to \(1/32\). Through four experimental groups illustrated in Figure 6 (right), each characterized by different configurations of \(I\) and \(S\), we find that as the prior probability increases, so does the performance.
Property 3: _์„ฑ๋Šฅ์€ ์‚ฌ์ „ ํ™•๋ฅ ๊ณผ ํ•จ๊ป˜ ์ฆ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค._ ์šฐ๋ฆฌ๋Š” \(I\)์™€ \(K\) ๊ฐ’์„ ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€ํ•˜๋ฉด์„œ ๋งค๊ฐœ๋ณ€์ˆ˜ \(K\)๋ฅผ ์กฐ์ •ํ•˜์—ฌ ์‚ฌ์ „ ํ™•๋ฅ ์ด ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. \(K\)๋Š” ๊ฐ„๊ฒฉ์˜ ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฏ€๋กœ, ์‚ฌ์ „ ํ™•๋ฅ ์€ \(1/K\)๋กœ ์ •์˜๋ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” \(K\)๋ฅผ \(4\)์—์„œ \(32\)๊นŒ์ง€ ๋ณ€ํ™”์‹œ์ผœ ์‚ฌ์ „ ํ™•๋ฅ ์„ \(1/4\)์—์„œ \(1/32\)๋กœ ๋ณ€๊ฒฝํ•ฉ๋‹ˆ๋‹ค. Figure 6 (์˜ค๋ฅธ์ชฝ)์— ๋‚˜ํƒ€๋‚œ ๋„ค ๊ฐœ์˜ ์‹คํ—˜ ๊ทธ๋ฃน์„ ํ†ตํ•ด, ๊ฐ ๊ทธ๋ฃน์€ \(I\)์™€ \(S\)์˜ ๋‹ค๋ฅธ ๊ตฌ์„ฑ์œผ๋กœ ํŠน์ง•์ง€์–ด์ง€๋ฉฐ, ์‚ฌ์ „ ํ™•๋ฅ ์ด ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์„ฑ๋Šฅ๋„ ์ฆ๊ฐ€ํ•œ๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค.
์†์„ฑ 3: _์‚ฌ์ „ ํ™•๋ฅ ์ด ๋†’์„์ˆ˜๋ก ์„ฑ๋Šฅ์ด ํ–ฅ์ƒ๋ฉ๋‹ˆ๋‹ค._ ์šฐ๋ฆฌ๋Š” \(I\)์™€ \(K\)์˜ ๊ฐ’์„ ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€ํ•˜๋ฉด์„œ ํŒŒ๋ผ๋ฏธํ„ฐ \(K\)๋ฅผ ์กฐ์ •ํ•˜์—ฌ ์‚ฌ์ „ ํ™•๋ฅ ์ด ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. K๋Š” ๊ฐ„๊ฒฉ์˜ ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฏ€๋กœ ์‚ฌ์ „ ํ™•๋ฅ ์€ \(1/K\)๋กœ ์ •์˜๋ฉ๋‹ˆ๋‹ค. K\)๋ฅผ \(4\)์—์„œ \(32\)๊นŒ์ง€ ๋ณ€๊ฒฝํ•˜๋ฉด ์‚ฌ์ „ ํ™•๋ฅ ์ด \(1/4\)์—์„œ \(1/32\)๋กœ ๋™์ผํ•˜๊ฒŒ ๋ณ€๊ฒฝ๋ฉ๋‹ˆ๋‹ค. ๊ทธ๋ฆผ 6(์˜ค๋ฅธ์ชฝ)์— ๋‚˜์™€ ์žˆ๋Š” ๋„ค ๊ฐ€์ง€ ์‹คํ—˜ ๊ทธ๋ฃน์„ ํ†ตํ•ด ๊ฐ๊ฐ ๋‹ค๋ฅธ \(I\) ๋ฐ \(S\) ๊ตฌ์„ฑ์œผ๋กœ ํŠน์ง•์ง€์–ด์ง€๋Š” ์„ฑ๋Šฅ์ด ์‚ฌ์ „ ํ™•๋ฅ ์ด ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์ฆ๊ฐ€ํ•œ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**RQ2. How often does an LLM perform the second hop of reasoning while processing two-hop prompts?** We view the second-hop reasoning as the LLM's utilization of the first-hop reasoning for the second hop. Therefore, we examine the frequency with which enhancing the LLM's recall of the bridge entity for its descriptive mention improves its use of the knowledge about the bridge entity to answer the two-hop prompt. For example, we investigate if increasing the internal recall of Stevie Wonder for _"the singer of 'Superstition'"_ makes the LLM better utilize its knowledge of Stevie Wonder's mother to complete the prompt.
RQ2. LLM์ด ๋‘ ๋‹จ๊ณ„์˜ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ฒ˜๋ฆฌํ•  ๋•Œ ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์˜ ์ถ”๋ก ์„ ์–ผ๋งˆ๋‚˜ ์ž์ฃผ ์ˆ˜ํ–‰ํ•˜๋Š”๊ฐ€? ์šฐ๋ฆฌ๋Š” ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์˜ ์ถ”๋ก ์„ ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„์˜ ์ถ”๋ก ์„ ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์— ํ™œ์šฉํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋ด…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ, ๋‹ค๋ฆฌ ์—ญํ• ์„ ํ•˜๋Š” ์—”ํ‹ฐํ‹ฐ์— ๋Œ€ํ•œ ์„ค๋ช…์  ์–ธ๊ธ‰์„ ๊ธฐ์–ตํ•˜๋Š” LLM์˜ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์ด ๋‘ ๋‹จ๊ณ„ ํ”„๋กฌํ”„ํŠธ์— ๋‹ตํ•˜๊ธฐ ์œ„ํ•ด ๋‹ค๋ฆฌ ์—”ํ‹ฐํ‹ฐ์— ๋Œ€ํ•œ ์ง€์‹์„ ์‚ฌ์šฉํ•˜๋Š” ๋นˆ๋„๋ฅผ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, _"Superstition์˜ ๊ฐ€์ˆ˜"_์— ๋Œ€ํ•œ Stevie Wonder์˜ ๋‚ด๋ถ€ ๊ธฐ์–ต์„ ์ฆ๊ฐ€์‹œํ‚ค๋Š” ๊ฒƒ์ด LLM์ด Stevie Wonder์˜ ์–ด๋จธ๋‹ˆ์— ๋Œ€ํ•œ ์ง€์‹์„ ๋” ์ž˜ ํ™œ์šฉํ•˜์—ฌ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์™„์„ฑํ•˜๋Š”์ง€ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
**RQ2. LLM์ด ๋‘ ํ™‰ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ๋™์•ˆ ๋‘ ๋ฒˆ์งธ ํ™‰ ์ถ”๋ก ์„ ์–ผ๋งˆ๋‚˜ ์ž์ฃผ ์ˆ˜ํ–‰ํ•˜๋‚˜์š”?** ๋‘ ๋ฒˆ์งธ ํ™‰ ์ถ”๋ก ์€ LLM์ด ์ฒซ ๋ฒˆ์งธ ํ™‰ ์ถ”๋ก ์„ ๋‘ ๋ฒˆ์งธ ํ™‰์— ํ™œ์šฉํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ LLM์ด ๋‘ ํ™‰ ํ”„๋กฌํ”„ํŠธ์— ๋‹ตํ•˜๊ธฐ ์œ„ํ•ด ๊ต๋Ÿ‰ ๊ฐœ์ฒด์— ๋Œ€ํ•œ ์ง€์‹์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐ ์žˆ์–ด, ๊ต๋Ÿ‰ ๊ฐœ์ฒด์— ๋Œ€ํ•œ ์„ค๋ช…์  ์–ธ๊ธ‰์— ๋Œ€ํ•œ LLM์˜ ๋ฆฌ์ฝœ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์ด ์–ผ๋งˆ๋‚˜ ์ž์ฃผ ๊ฐœ์„ ๋˜๋Š”์ง€ ์‚ดํŽด๋ด…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, _"๋ฏธ์น˜๊ด‘์ด์˜ ๊ฐ€์ˆ˜"_์— ๋Œ€ํ•œ ์Šคํ‹ฐ๋น„ ์›๋”์˜ ๋‚ด๋ถ€ ๋ฆฌ์ฝœ์„ ๋†’์ด๋ฉด LLM์ด ์Šคํ‹ฐ๋น„ ์›๋”์˜ ์–ด๋จธ๋‹ˆ์— ๋Œ€ํ•œ ์ง€์‹์„ ๋” ์ž˜ ํ™œ์šฉํ•˜์—ฌ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์™„๋ฃŒํ•  ์ˆ˜ ์žˆ๋Š”์ง€ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
**Are reference Gold or Gilded?** We conducted an in-depth analysis of the training data (FLORES-200 data) utilized by the ALMA model. We meticulously compared the quality of the reference translations with those generated by strong translation models. Our findings reveal that, in numerous instances, the quality of human-written parallel data is even inferior to that of system-generated translations. This observation underscores a critical insight: training models exclusively towards replicating reference translations may not be the most effective approach, and reliance on reference-based evaluation could be flawed.
**์ฐธ์กฐ ๋ฒˆ์—ญ์€ Gold์ธ๊ฐ€ Gilded์ธ๊ฐ€?** ์šฐ๋ฆฌ๋Š” ALMA ๋ชจ๋ธ์ด ์‚ฌ์šฉํ•œ ํ•™์Šต ๋ฐ์ดํ„ฐ(FLORES-200 ๋ฐ์ดํ„ฐ)์— ๋Œ€ํ•œ ์‹ฌ์ธต ๋ถ„์„์„ ์ˆ˜ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ฐธ์กฐ ๋ฒˆ์—ญ์˜ ํ’ˆ์งˆ์„ ๊ฐ•๋ ฅํ•œ ๋ฒˆ์—ญ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ๋ฒˆ์—ญ๊ณผ ์„ธ์‹ฌํ•˜๊ฒŒ ๋น„๊ตํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ, ๋งŽ์€ ๊ฒฝ์šฐ์— ์ธ๊ฐ„์ด ์ž‘์„ฑํ•œ ๋ณ‘๋ ฌ ๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ์ด ์‹œ์Šคํ…œ์ด ์ƒ์„ฑํ•œ ๋ฒˆ์—ญ๋ณด๋‹ค ์˜คํžˆ๋ ค ์—ด๋“ฑํ•˜๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ด€์ฐฐ์€ ์ค‘์š”ํ•œ ํ†ต์ฐฐ์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค: ์ฐธ์กฐ ๋ฒˆ์—ญ์„ ๋ณต์ œํ•˜๋Š” ๊ฒƒ์—๋งŒ ๋ชจ๋ธ์„ ํ›ˆ๋ จ์‹œํ‚ค๋Š” ๊ฒƒ์ด ๊ฐ€์žฅ ํšจ๊ณผ์ ์ธ ์ ‘๊ทผ๋ฒ•์ด ์•„๋‹ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ฐธ์กฐ ๊ธฐ๋ฐ˜ ํ‰๊ฐ€์— ์˜์กดํ•˜๋Š” ๊ฒƒ์ด ๊ฒฐํ•จ์ด ์žˆ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**์ฐธ์กฐ ๊ณจ๋“œ ๋˜๋Š” ๊ธˆ๋ฐ• ์ฒ˜๋ฆฌ๋˜์–ด ์žˆ๋‚˜์š”?** ALMA ๋ชจ๋ธ์—์„œ ํ™œ์šฉํ•˜๋Š” ํ•™์Šต ๋ฐ์ดํ„ฐ(FLORES-200 ๋ฐ์ดํ„ฐ)์— ๋Œ€ํ•œ ์‹ฌ์ธต ๋ถ„์„์„ ์ˆ˜ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ฐธ์กฐ ๋ฒˆ์—ญ์˜ ํ’ˆ์งˆ์„ ๊ฐ•๋ ฅํ•œ ๋ฒˆ์—ญ ๋ชจ๋ธ์—์„œ ์ƒ์„ฑํ•œ ๋ฒˆ์—ญ๊ณผ ๊ผผ๊ผผํ•˜๊ฒŒ ๋น„๊ตํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ ๊ฒฐ๊ณผ, ์ˆ˜๋งŽ์€ ๊ฒฝ์šฐ์— ๊ฑธ์ณ ์‚ฌ๋žŒ์ด ์ž‘์„ฑํ•œ ๋ณ‘๋ ฌ ๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ์ด ์‹œ์Šคํ…œ์—์„œ ์ƒ์„ฑํ•œ ๋ฒˆ์—ญ๋ณด๋‹ค ๋–จ์–ด์ง€๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ด€์ฐฐ์€ ์ค‘์š”ํ•œ ํ†ต์ฐฐ๋ ฅ์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค. ์ฐธ์กฐ ๋ฒˆ์—ญ์„ ๋ณต์ œํ•˜๋Š” ๊ฒƒ๋งŒ์„ ๋ชฉํ‘œ๋กœ ๋ชจ๋ธ์„ ํ•™์Šต์‹œํ‚ค๋Š” ๊ฒƒ์ด ๊ฐ€์žฅ ํšจ๊ณผ์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด ์•„๋‹ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ฐธ์กฐ ๊ธฐ๋ฐ˜ ํ‰๊ฐ€์— ์˜์กดํ•˜๋Š” ๊ฒƒ์€ ๊ฒฐํ•จ์ด ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
We conducted a comprehensive evaluation of the model across various math reasoning benchmarks, encompassing a range of difficulties from elementary to university level, multiple mathematical domains, and diverse question types including multiple-choice and open-ended questions. Our benchmarks include GSM8k [Cobbe et al., 2021], MATH [Hendrycks et al., 2021], GSM-Hard [Gao et al., 2022], SVAMP [Patel et al., 2021], ASDIV [Miao et al., 2020], MAWPS [Koncel-Kedziorski et al., 2016], TabMWP (TAB) [Lu et al., 2023], MathQA (MQA) [Amini et al., 2019], MMLU-STEM [Hendrycks et al., 2020], and SAT [Azerbayev et al., 2023].
์šฐ๋ฆฌ๋Š” ์ดˆ๋“ฑํ•™๊ต๋ถ€ํ„ฐ ๋Œ€ํ•™๊ต ์ˆ˜์ค€๊นŒ์ง€ ๋‹ค์–‘ํ•œ ๋‚œ์ด๋„์˜ ์ˆ˜ํ•™ ์ถ”๋ก  ๋ฒค์น˜๋งˆํฌ, ์—ฌ๋Ÿฌ ์ˆ˜ํ•™์  ์˜์—ญ, ๊ทธ๋ฆฌ๊ณ  ๊ฐ๊ด€์‹ ๋ฐ ์ฃผ๊ด€์‹ ์งˆ๋ฌธ ์œ ํ˜•์„ ํฌํ•จํ•œ ํฌ๊ด„์ ์ธ ๋ชจ๋ธ ํ‰๊ฐ€๋ฅผ ์ˆ˜ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ๋ฒค์น˜๋งˆํฌ์—๋Š” GSM8k [Cobbe et al., 2021], MATH [Hendrycks et al., 2021], GSM-Hard [Gao et al., 2022], SVAMP [Patel et al., 2021], ASDIV [Miao et al., 2020], MAWPS [Koncel-Kedziorski et al., 2016], TabMWP (TAB) [Lu et al., 2023], MathQA (MQA) [Amini et al., 2019], MMLU-STEM [Hendrycks et al., 2020], ๊ทธ๋ฆฌ๊ณ  SAT [Azerbayev et al., 2023]๊ฐ€ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค.
์ดˆ๋“ฑํ•™๊ต๋ถ€ํ„ฐ ๋Œ€ํ•™ ์ˆ˜์ค€๊นŒ์ง€ ๋‹ค์–‘ํ•œ ๋‚œ์ด๋„, ์—ฌ๋Ÿฌ ์ˆ˜ํ•™ ์˜์—ญ, ๊ฐ๊ด€์‹ ๋ฐ ์ฃผ๊ด€์‹ ๋ฌธ์ œ ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฌธ์ œ ์œ ํ˜•์„ ํฌํ•จํ•˜๋Š” ๋‹ค์–‘ํ•œ ์ˆ˜ํ•™ ์ถ”๋ก  ๋ฒค์น˜๋งˆํฌ์— ๊ฑธ์ณ ๋ชจ๋ธ์— ๋Œ€ํ•œ ์ข…ํ•ฉ์ ์ธ ํ‰๊ฐ€๋ฅผ ์‹ค์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฒค์น˜๋งˆํฌ์—๋Š” GSM8k [Cobbe et al., 2021], MATH [Hendrycks et al., 2021], GSM-Hard [Gao et al., 2022], SVAMP [Patel et al., 2021], ASDIV [Miao et al., 2020], MAWPS [Koncel-Kedziorski et al., 2016], TabMWP(TAB) [Lu et al., 2023], MathQA(MQA) [Amini et al., 2019], MMLU-STEM [Hendrycks et al., 2020], SAT [Azerbayev et al., 2023] ๋“ฑ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
Indian patient actors played the roles in all India scenario packs and 7 of the 14 UK scenario packs. Canadian patient actors participated in scenario packs for both Canada and the other half of UK-based scenario packs. This assignment process resulted in 149 distinct simulated patients ("scenarios"). Below, we use the term "OSCE agent" to refer to the conversational counterpart interviewing the patient actor, i.e., either PCP or AMIE. Table 1 summarizes the OSCE assignment information across three geographical locations. Each of the 149 simulated patients completed the three-step study flow depicted in Figure 2.
์ธ๋„ ํ™˜์ž ๋ฐฐ์šฐ๋“ค์€ ๋ชจ๋“  ์ธ๋„ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ๊ณผ 14๊ฐœ์˜ ์˜๊ตญ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ ์ค‘ 7๊ฐœ์—์„œ ์—ญํ• ์„ ๋งก์•˜์Šต๋‹ˆ๋‹ค. ์บ๋‚˜๋‹ค ํ™˜์ž ๋ฐฐ์šฐ๋“ค์€ ์บ๋‚˜๋‹ค์™€ ๋‚˜๋จธ์ง€ ์ ˆ๋ฐ˜์˜ ์˜๊ตญ ๊ธฐ๋ฐ˜ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ์—์„œ ์—ญํ• ์„ ๋งก์•˜์Šต๋‹ˆ๋‹ค. ์ด ํ• ๋‹น ๊ณผ์ •์€ 149๊ฐœ์˜ ๋…ํŠนํ•œ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜์ž("์‹œ๋‚˜๋ฆฌ์˜ค")๋ฅผ ๋งŒ๋“ค์–ด๋ƒˆ์Šต๋‹ˆ๋‹ค. ์•„๋ž˜์—์„œ๋Š” "OSCE ์—์ด์ „ํŠธ"๋ผ๋Š” ์šฉ์–ด๋ฅผ ํ™˜์ž ๋ฐฐ์šฐ์™€ ์ธํ„ฐ๋ทฐํ•˜๋Š” ๋Œ€ํ™” ์ƒ๋Œ€, ์ฆ‰ PCP ๋˜๋Š” AMIE๋ฅผ ์ง€์นญํ•˜๋Š” ๋ฐ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ํ‘œ 1์€ ์„ธ ๊ฐœ์˜ ์ง€๋ฆฌ์  ์œ„์น˜์— ๊ฑธ์นœ OSCE ํ• ๋‹น ์ •๋ณด๋ฅผ ์š”์•ฝํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. 149๋ช…์˜ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜์ž ๊ฐ๊ฐ์€ ๊ทธ๋ฆผ 2์— ๋ฌ˜์‚ฌ๋œ 3๋‹จ๊ณ„ ์—ฐ๊ตฌ ํ๋ฆ„์„ ์™„๋ฃŒํ–ˆ์Šต๋‹ˆ๋‹ค.
์ธ๋„ ํ™˜์ž ๋ฐฐ์šฐ๋Š” ์ธ๋„ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ ์ „๋ฐ˜๊ณผ 14๊ฐœ์˜ ์˜๊ตญ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ ์ค‘ 7๊ฐœ์—์„œ ์—ญํ• ์„ ๋งก์•˜์Šต๋‹ˆ๋‹ค. ์บ๋‚˜๋‹ค ํ™˜์ž ๋ฐฐ์šฐ๋Š” ์บ๋‚˜๋‹ค ๋ฐ ๋‚˜๋จธ์ง€ ์˜๊ตญ ๊ธฐ๋ฐ˜ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ์˜ ์ ˆ๋ฐ˜์— ๋Œ€ํ•œ ์‹œ๋‚˜๋ฆฌ์˜ค ํŒฉ์— ์ฐธ์—ฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ํ• ๋‹น ํ”„๋กœ์„ธ์Šค๋ฅผ ํ†ตํ•ด 149๊ฐœ์˜ ๊ณ ์œ ํ•œ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜์ž("์‹œ๋‚˜๋ฆฌ์˜ค")๊ฐ€ ์ƒ์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์•„๋ž˜์—์„œ๋Š” ํ™˜์ž ๋ฐฐ์šฐ์™€ ์ธํ„ฐ๋ทฐํ•˜๋Š” ๋Œ€ํ™” ์ƒ๋Œ€, ์ฆ‰ PCP ๋˜๋Š” AMIE๋ฅผ ์ง€์นญํ•˜๊ธฐ ์œ„ํ•ด "OSCE ์—์ด์ „ํŠธ"๋ผ๋Š” ์šฉ์–ด๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ํ‘œ 1์€ ์„ธ ์ง€์—ญ์— ๊ฑธ์นœ OSCE ํ• ๋‹น ์ •๋ณด๋ฅผ ์š”์•ฝํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. 149๋ช…์˜ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ™˜์ž๋Š” ๊ทธ๋ฆผ 2์— ํ‘œ์‹œ๋œ 3๋‹จ๊ณ„ ์—ฐ๊ตฌ ํ๋ฆ„์„ ๋ชจ๋‘ ์™„๋ฃŒํ–ˆ์Šต๋‹ˆ๋‹ค.
Memorizing training data.The privacy risks of data memorization, for example the ability to extract sensitive data such as valid phone numbers and IRC usernames, are highlighted by Carlini et al. (2020). While their paper finds 604 samples that GPT-2 emitted from its training set, we show that _over_\(1\%\) of the data most models emit is memorized training data. In computer vision, memorization of training data has been studied from various angles for both discriminative and generative models (e.g. Arpit et al., 2017; Webster et al., 2019; Feldman and Zhang, 2020; Stephenson et al., 2021; Tetewak et al., 2021).
ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์˜ ์•”๊ธฐ. ๋ฐ์ดํ„ฐ ์•”๊ธฐ์˜ ํ”„๋ผ์ด๋ฒ„์‹œ ์œ„ํ—˜, ์˜ˆ๋ฅผ ๋“ค์–ด ์œ ํšจํ•œ ์ „ํ™”๋ฒˆํ˜ธ๋‚˜ IRC ์‚ฌ์šฉ์ž ์ด๋ฆ„๊ณผ ๊ฐ™์€ ๋ฏผ๊ฐํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ถ”์ถœํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์€ Carlini et al. (2020)์— ์˜ํ•ด ๊ฐ•์กฐ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋“ค์˜ ๋…ผ๋ฌธ์—์„œ๋Š” GPT-2๊ฐ€ ํ›ˆ๋ จ ์„ธํŠธ์—์„œ ๋ฐฉ์ถœํ•œ 604๊ฐœ์˜ ์ƒ˜ํ”Œ์„ ๋ฐœ๊ฒฌํ–ˆ์ง€๋งŒ, ์šฐ๋ฆฌ๋Š” ๋Œ€๋ถ€๋ถ„์˜ ๋ชจ๋ธ์ด ๋ฐฉ์ถœํ•˜๋Š” ๋ฐ์ดํ„ฐ์˜ _1\%_ ์ด์ƒ์ด ์•”๊ธฐ๋œ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์ž„์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ปดํ“จํ„ฐ ๋น„์ „ ๋ถ„์•ผ์—์„œ๋Š”, ์•”๊ธฐ๋œ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ๊ฐ€ ํŒ๋ณ„ ๋ชจ๋ธ๊ณผ ์ƒ์„ฑ ๋ชจ๋ธ ๋ชจ๋‘์—์„œ ๋‹ค์–‘ํ•œ ๊ฐ๋„๋กœ ์—ฐ๊ตฌ๋˜์—ˆ์Šต๋‹ˆ๋‹ค (์˜ˆ: Arpit et al., 2017; Webster et al., 2019; Feldman and Zhang, 2020; Stephenson et al., 2021; Tetewak et al., 2021).
ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ๋ฅผ ์•”๊ธฐํ•ฉ๋‹ˆ๋‹ค.์œ ํšจํ•œ ์ „ํ™”๋ฒˆํ˜ธ ๋ฐ IRC ์‚ฌ์šฉ์ž ์ด๋ฆ„๊ณผ ๊ฐ™์€ ๋ฏผ๊ฐํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ถ”์ถœํ•˜๋Š” ๊ธฐ๋Šฅ๊ณผ ๊ฐ™์€ ๋ฐ์ดํ„ฐ ์•”๊ธฐ์˜ ๊ฐœ์ธ์ •๋ณด ๋ณดํ˜ธ ์œ„ํ—˜์€ Carlini ๋“ฑ(2020)์— ์˜ํ•ด ๊ฐ•์กฐ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ณด๊ณ ์„œ์—์„œ GPT-2๊ฐ€ ํ•™์Šต ์„ธํŠธ์—์„œ ์ถœ๋ ฅํ•œ 604๊ฐœ์˜ ์ƒ˜ํ”Œ์„ ๋ฐœ๊ฒฌํ–ˆ์ง€๋งŒ, ๋Œ€๋ถ€๋ถ„์˜ ๋ชจ๋ธ์—์„œ ์ถœ๋ ฅํ•˜๋Š” ๋ฐ์ดํ„ฐ์˜ \(1\%\) ์ด์ƒ์ด ์•”๊ธฐ๋œ ํ•™์Šต ๋ฐ์ดํ„ฐ๋ผ๋Š” ๊ฒƒ์„ ๋ณด์—ฌ ์ค๋‹ˆ๋‹ค. ์ปดํ“จํ„ฐ ๋น„์ „์—์„œ๋Š” ํŒ๋ณ„ ๋ชจ๋ธ๊ณผ ์ƒ์„ฑ ๋ชจ๋ธ ๋ชจ๋‘์— ๋Œ€ํ•ด ๋‹ค์–‘ํ•œ ๊ฐ๋„์—์„œ ํ•™์Šต ๋ฐ์ดํ„ฐ์˜ ์•”๊ธฐ๊ฐ€ ์—ฐ๊ตฌ๋˜์–ด ์™”์Šต๋‹ˆ๋‹ค(์˜ˆ: Arpit et al., 2017; Webster et al., 2019; Feldman and Zhang, 2020; Stephenson et al., 2021; Tetewak et al., 2021).
This paper proposes the AIOS architecture, demonstrating the potential to facilitate the development and deployment of LLM-based agents, fostering a more cohesive, effective and efficient AIOS-Agent ecosystem. The insights and methodologies presented herein contribute to the ongoing discourse in both AI and system research, offering a viable solution to the integration challenges posed by the diverse landscape of AI Agents. Diverse future work can be built upon this foundation, exploring innovative ways to refine and expand the AIOS architecture to meet the evolving needs of developing and deploying LLM agents.
์ด ๋…ผ๋ฌธ์€ LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ์˜ ๊ฐœ๋ฐœ ๋ฐ ๋ฐฐํฌ๋ฅผ ์ด‰์ง„ํ•  ์ˆ˜ ์žˆ๋Š” ์ž ์žฌ๋ ฅ์„ ๋ณด์—ฌ์ฃผ๋Š” AIOS ์•„ํ‚คํ…์ฒ˜๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๋” ํ†ตํ•ฉ์ ์ด๊ณ  ํšจ๊ณผ์ ์ด๋ฉฐ ํšจ์œจ์ ์ธ AIOS-Agent ์ƒํƒœ๊ณ„๋ฅผ ์กฐ์„ฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ ์ œ์‹œ๋œ ํ†ต์ฐฐ๋ ฅ๊ณผ ๋ฐฉ๋ฒ•๋ก ์€ AI ๋ฐ ์‹œ์Šคํ…œ ์—ฐ๊ตฌ ๋ชจ๋‘์—์„œ ์ง„ํ–‰ ์ค‘์ธ ๋…ผ์˜์— ๊ธฐ์—ฌํ•˜๋ฉฐ, ๋‹ค์–‘ํ•œ AI ์—์ด์ „ํŠธ์˜ ํ†ตํ•ฉ ๋ฌธ์ œ์— ๋Œ€ํ•œ ์‹ค์งˆ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ธฐ์ดˆ ์œ„์— ๋‹ค์–‘ํ•œ ๋ฏธ๋ž˜ ์—ฐ๊ตฌ๊ฐ€ ๊ตฌ์ถ•๋  ์ˆ˜ ์žˆ์œผ๋ฉฐ, LLM ์—์ด์ „ํŠธ์˜ ๊ฐœ๋ฐœ ๋ฐ ๋ฐฐํฌ์˜ ์ง„ํ™”ํ•˜๋Š” ์š”๊ตฌ๋ฅผ ์ถฉ์กฑํ•˜๊ธฐ ์œ„ํ•ด AIOS ์•„ํ‚คํ…์ฒ˜๋ฅผ ์ •์ œํ•˜๊ณ  ํ™•์žฅํ•˜๋Š” ํ˜์‹ ์ ์ธ ๋ฐฉ๋ฒ•์„ ํƒ๊ตฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ด ๋ฐฑ์„œ์—์„œ๋Š” LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ์˜ ๊ฐœ๋ฐœ ๋ฐ ๋ฐฐํฌ๋ฅผ ์ด‰์ง„ํ•˜์—ฌ ๋ณด๋‹ค ์‘์ง‘๋ ฅ ์žˆ๊ณ  ํšจ๊ณผ์ ์ด๋ฉฐ ํšจ์œจ์ ์ธ AIOS ์—์ด์ „ํŠธ ์—์ฝ”์‹œ์Šคํ…œ์„ ์กฐ์„ฑํ•  ์ˆ˜ ์žˆ๋Š” ์ž ์žฌ๋ ฅ์„ ๋ณด์—ฌ์ฃผ๋Š” AIOS ์•„ํ‚คํ…์ฒ˜๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์— ์ œ์‹œ๋œ ์ธ์‚ฌ์ดํŠธ์™€ ๋ฐฉ๋ฒ•๋ก ์€ AI ๋ฐ ์‹œ์Šคํ…œ ์—ฐ๊ตฌ ๋ถ„์•ผ์˜ ์ง€์†์ ์ธ ๋…ผ์˜์— ๊ธฐ์—ฌํ•˜๋ฉฐ, ๋‹ค์–‘ํ•œ AI ์—์ด์ „ํŠธ ํ™˜๊ฒฝ์ด ์ œ๊ธฐํ•˜๋Š” ํ†ตํ•ฉ ๋ฌธ์ œ์— ๋Œ€ํ•œ ์‹คํ–‰ ๊ฐ€๋Šฅํ•œ ์†”๋ฃจ์…˜์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ํ–ฅํ›„ ์—ฐ๊ตฌ๋Š” ์ด ๊ธฐ๋ฐ˜์„ ๋ฐ”ํƒ•์œผ๋กœ LLM ์—์ด์ „ํŠธ ๊ฐœ๋ฐœ ๋ฐ ๋ฐฐํฌ์˜ ์ง„ํ™”ํ•˜๋Š” ์š”๊ตฌ ์‚ฌํ•ญ์„ ์ถฉ์กฑํ•˜๊ธฐ ์œ„ํ•ด AIOS ์•„ํ‚คํ…์ฒ˜๋ฅผ ๊ฐœ์„ ํ•˜๊ณ  ํ™•์žฅํ•˜๋Š” ํ˜์‹ ์ ์ธ ๋ฐฉ๋ฒ•์„ ๋ชจ์ƒ‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**Metrics.** Considering the limitations of 2-billion parameter models, our evaluation criteria go beyond just the accuracy of responses. We additionally consider factors such as usefulness, relevance, accuracy, depth, creativity, and the level of detail in the model's answers. This comprehensive method allows for a detailed evaluation of the model's response quality. Specifically, We use GPT-4 (Achiam et al., 2023) to score responses from tested LLMs in specific problem contexts, with the scoring prompt available in the Appendix. C.2. We translate the score assignment prompt template from Zheng et al. (2024).
**ํ‰๊ฐ€์ง€ํ‘œ.** 20์–ต ๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๊ฐ€์ง„ ๋ชจ๋ธ์˜ ํ•œ๊ณ„๋ฅผ ๊ณ ๋ คํ•˜์—ฌ, ์šฐ๋ฆฌ์˜ ํ‰๊ฐ€ ๊ธฐ์ค€์€ ๋‹จ์ˆœํžˆ ์‘๋‹ต์˜ ์ •ํ™•์„ฑ์— ๊ตญํ•œ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์œ ์šฉ์„ฑ, ๊ด€๋ จ์„ฑ, ์ •ํ™•์„ฑ, ๊นŠ์ด, ์ฐฝ์˜์„ฑ, ๊ทธ๋ฆฌ๊ณ  ๋ชจ๋ธ ์‘๋‹ต์˜ ์„ธ๋ถ€ ์ˆ˜์ค€๊ณผ ๊ฐ™์€ ์š”์†Œ๋“ค๋„ ์ถ”๊ฐ€๋กœ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค. ์ด ํฌ๊ด„์ ์ธ ๋ฐฉ๋ฒ•์€ ๋ชจ๋ธ ์‘๋‹ต์˜ ํ’ˆ์งˆ์„ ์ƒ์„ธํžˆ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ์šฐ๋ฆฌ๋Š” ํŠน์ • ๋ฌธ์ œ ์ƒํ™ฉ์—์„œ ํ…Œ์ŠคํŠธ๋œ LLM์˜ ์‘๋‹ต์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด GPT-4 (Achiam et al., 2023)๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, ํ‰๊ฐ€ ํ”„๋กฌํ”„ํŠธ๋Š” ๋ถ€๋ก C.2์— ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” Zheng et al. (2024)์˜ ์ ์ˆ˜ ํ• ๋‹น ํ”„๋กฌํ”„ํŠธ ํ…œํ”Œ๋ฆฟ์„ ๋ฒˆ์—ญํ•˜์—ฌ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.
**๋ฉ”ํŠธ๋ฆญ.** 20์–ต ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ์˜ ํ•œ๊ณ„๋ฅผ ๊ณ ๋ คํ•˜์—ฌ, ์ €ํฌ์˜ ํ‰๊ฐ€ ๊ธฐ์ค€์€ ์‘๋‹ต์˜ ์ •ํ™•์„ฑ ๊ทธ ์ด์ƒ์ž…๋‹ˆ๋‹ค. ์œ ์šฉ์„ฑ, ๊ด€๋ จ์„ฑ, ์ •ํ™•์„ฑ, ๊นŠ์ด, ์ฐฝ์˜์„ฑ, ๋ชจ๋ธ ๋‹ต๋ณ€์˜ ์„ธ๋ถ€ ์ˆ˜์ค€ ๋“ฑ์˜ ์š”์†Œ๋„ ์ถ”๊ฐ€๋กœ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ข…ํ•ฉ์ ์ธ ๋ฐฉ๋ฒ•์„ ํ†ตํ•ด ๋ชจ๋ธ์˜ ์‘๋‹ต ํ’ˆ์งˆ์„ ์ƒ์„ธํ•˜๊ฒŒ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ํŠน์ • ๋ฌธ์ œ ๋งฅ๋ฝ์—์„œ ํ…Œ์ŠคํŠธ๋œ LLM์˜ ์‘๋‹ต์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด GPT-4(Achiam et al., 2023)๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, ์ฑ„์  ํ”„๋กฌํ”„ํŠธ๋Š” ๋ถ€๋ก C.2์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. C.2. ์ ์ˆ˜ ํ• ๋‹น ํ”„๋กฌํ”„ํŠธ ํ…œํ”Œ๋ฆฟ์€ Zheng et al.(2024)์—์„œ ๋ฒˆ์—ญํ–ˆ์Šต๋‹ˆ๋‹ค.
Despite focusing on different aspects of examination-style questions and encompassing both English and Chinese languages, MMLU, CMMLU, and CEVAL appear to assess overlapping capabilities of the models, leading to similar performance trends. Furthermore, the training process of models demonstrates that the behavior of known tasks within a domain can predict the behavior of yet-to-be-encountered tasks. This suggests that understanding a model's performance on familiar tasks can offer valuable insights into its potential performance on analogous tasks that have not been previously explored within the same domain.
์‹œํ—˜ ์Šคํƒ€์ผ ์งˆ๋ฌธ์˜ ๋‹ค์–‘ํ•œ ์ธก๋ฉด์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์˜์–ด์™€ ์ค‘๊ตญ์–ด๋ฅผ ๋ชจ๋‘ ํฌํ•จํ•˜๊ณ  ์žˆ์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , MMLU, CMMLU, CEVAL์€ ๋ชจ๋ธ์˜ ์ค‘๋ณต๋˜๋Š” ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณด์ด๋ฉฐ, ์ด๋Š” ์œ ์‚ฌํ•œ ์„ฑ๋Šฅ ๊ฒฝํ–ฅ์„ ์ดˆ๋ž˜ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋ชจ๋ธ์˜ ํ›ˆ๋ จ ๊ณผ์ •์€ ๋„๋ฉ”์ธ ๋‚ด์—์„œ ์•Œ๋ ค์ง„ ์ž‘์—…์˜ ํ–‰๋™์ด ์•„์ง ์ ‘ํ•˜์ง€ ์•Š์€ ์ž‘์—…์˜ ํ–‰๋™์„ ์˜ˆ์ธกํ•  ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์ด ์ต์ˆ™ํ•œ ์ž‘์—…์—์„œ์˜ ์„ฑ๋Šฅ์„ ์ดํ•ดํ•˜๋Š” ๊ฒƒ์ด ๋™์ผํ•œ ๋„๋ฉ”์ธ ๋‚ด์—์„œ ์ด์ „์— ํƒ๊ตฌ๋˜์ง€ ์•Š์€ ์œ ์‚ฌํ•œ ์ž‘์—…์—์„œ์˜ ์ž ์žฌ์  ์„ฑ๋Šฅ์— ๋Œ€ํ•œ ๊ท€์ค‘ํ•œ ํ†ต์ฐฐ๋ ฅ์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Œ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
์‹œํ—˜ ์Šคํƒ€์ผ์˜ ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ค์–‘ํ•œ ์ธก๋ฉด์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์˜์–ด์™€ ์ค‘๊ตญ์–ด๋ฅผ ๋ชจ๋‘ ํฌ๊ด„ํ•˜์ง€๋งŒ, MMLU, CMMLU ๋ฐ CEVAL์€ ๋ชจ๋ธ์˜ ์ค‘๋ณต๋œ ๊ธฐ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณด์ด๋ฉฐ, ์ด๋กœ ์ธํ•ด ์œ ์‚ฌํ•œ ์„ฑ๋Šฅ ์ถ”์„ธ๋กœ ์ด์–ด์ง‘๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ชจ๋ธ์˜ ํ•™์Šต ๊ณผ์ •์„ ํ†ตํ•ด ํ•œ ๋„๋ฉ”์ธ ๋‚ด์—์„œ ์•Œ๋ ค์ง„ ์ž‘์—…์˜ ํ–‰๋™์ด ์•„์ง ์ ‘ํ•˜์ง€ ์•Š์€ ์ž‘์—…์˜ ํ–‰๋™์„ ์˜ˆ์ธกํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์นœ์ˆ™ํ•œ ์ž‘์—…์— ๋Œ€ํ•œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ์ดํ•ดํ•˜๋ฉด ๋™์ผํ•œ ๋„๋ฉ”์ธ ๋‚ด์—์„œ ์ด์ „์— ํƒ์ƒ‰ํ•˜์ง€ ์•Š์€ ์œ ์‚ฌํ•œ ์ž‘์—…์— ๋Œ€ํ•œ ์„ฑ๋Šฅ์„ ์˜ˆ์ธกํ•˜๋Š” ๋ฐ ์œ ์šฉํ•œ ์ธ์‚ฌ์ดํŠธ๋ฅผ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Œ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
Related work.Two works published during the responsible disclosure period use a similar procedure, and deal with numerical issues in different ways. (Chiu, 2024) start with a low \(B\) for the whole vocabulary, then increase \(B\) and ask for all tokens that haven't appeared before, and repeat until all tokens are covered. (Hayase et al., 2024) use the method in Appendix E.1, and set \(B=-\hat{z}_{i}\), where \(\hat{z}_{i}\) is an estimate of \(z_{i}\) inherent to their application. It is possible variants of this method have been discussed before our or these works, but we are not aware of further references.
๊ด€๋ จ ์—ฐ๊ตฌ. ์ฑ…์ž„ ์žˆ๋Š” ๊ณต๊ฐœ ๊ธฐ๊ฐ„ ๋™์•ˆ ๋ฐœํ‘œ๋œ ๋‘ ์—ฐ๊ตฌ๋Š” ์œ ์‚ฌํ•œ ์ ˆ์ฐจ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, ์„œ๋กœ ๋‹ค๋ฅธ ๋ฐฉ์‹์œผ๋กœ ์ˆ˜์น˜์  ๋ฌธ์ œ๋ฅผ ๋‹ค๋ฃน๋‹ˆ๋‹ค. (Chiu, 2024)๋Š” ์ „์ฒด ์–ดํœ˜์— ๋Œ€ํ•ด ๋‚ฎ์€ \(B\)๋กœ ์‹œ์ž‘ํ•œ ํ›„, \(B\)๋ฅผ ์ฆ๊ฐ€์‹œํ‚ค๊ณ  ์ด์ „์— ๋‚˜ํƒ€๋‚˜์ง€ ์•Š์€ ๋ชจ๋“  ํ† ํฐ์„ ์š”์ฒญํ•˜๋ฉฐ, ๋ชจ๋“  ํ† ํฐ์ด ํฌํ•จ๋  ๋•Œ๊นŒ์ง€ ์ด๋ฅผ ๋ฐ˜๋ณตํ•ฉ๋‹ˆ๋‹ค. (Hayase et al., 2024)๋Š” ๋ถ€๋ก E.1์— ์žˆ๋Š” ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜๋ฉฐ, \(B=-\hat{z}_{i}\)๋กœ ์„ค์ •ํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ \(\hat{z}_{i}\)๋Š” ๊ทธ๋“ค์˜ ์‘์šฉ์— ๋‚ด์žฌ๋œ \(z_{i}\)์˜ ์ถ”์ •์น˜์ž…๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์˜ ๋ณ€ํ˜•์ด ์šฐ๋ฆฌ ์—ฐ๊ตฌ๋‚˜ ์ด๋“ค ์—ฐ๊ตฌ ์ด์ „์— ๋…ผ์˜๋˜์—ˆ์„ ๊ฐ€๋Šฅ์„ฑ์€ ์žˆ์ง€๋งŒ, ์ถ”๊ฐ€์ ์ธ ์ฐธ๊ณ  ๋ฌธํ—Œ์€ ์•Œ์ง€ ๋ชปํ•ฉ๋‹ˆ๋‹ค.
๊ด€๋ จ ์ž‘์—….์ฑ…์ž„ ๊ณต๊ฐœ ๊ธฐ๊ฐ„ ๋™์•ˆ ๋ฐœํ‘œ๋œ ๋‘ ๊ฐœ์˜ ์ž‘์—…์€ ์œ ์‚ฌํ•œ ์ ˆ์ฐจ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, ์ˆ˜์น˜ ๋ฌธ์ œ๋ฅผ ๋‹ค๋ฅด๊ฒŒ ๋‹ค๋ฃน๋‹ˆ๋‹ค. (Chiu, 2024)๋Š” ์ „์ฒด ์–ดํœ˜์— ๋Œ€ํ•ด ๋‚ฎ์€ \(B\)๋กœ ์‹œ์ž‘ํ•˜์—ฌ \(B\)๋ฅผ ์ฆ๊ฐ€์‹œํ‚ค๊ณ  ์ด์ „์— ๋‚˜ํƒ€๋‚˜์ง€ ์•Š์€ ๋ชจ๋“  ํ† ํฐ์„ ์š”์ฒญํ•œ ๋‹ค์Œ ๋ชจ๋“  ํ† ํฐ์ด ํฌํ•จ๋  ๋•Œ๊นŒ์ง€ ๋ฐ˜๋ณตํ•ฉ๋‹ˆ๋‹ค. (Hayase et al., 2024)๋Š” ๋ถ€๋ก E.1์˜ ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ \(B=-\hat{z}_{i}\), ์—ฌ๊ธฐ์„œ \(\hat{z}_{i}\)๋Š” ํ•ด๋‹น ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— ๋‚ด์žฌ๋œ \(z_{i}\)์˜ ์ถ”์ •์น˜์ž…๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์˜ ๋ณ€ํ˜•์ด ์ €ํฌ๋‚˜ ๋‹ค๋ฅธ ์—ฐ๊ตฌ์ž๋“ค์— ์˜ํ•ด ๋…ผ์˜๋˜์—ˆ์„ ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ์ง€๋งŒ, ์ถ”๊ฐ€์ ์ธ ์ฐธ๊ณ  ์ž๋ฃŒ๋Š” ์—†์Šต๋‹ˆ๋‹ค.
Both our hidden-dimension-stealing and entire-layer-stealing attack worked for all five of these models. The size we recover from the model perfectly matches the actual size of the original model, as confirmed by OpenAI. For the first three models, we report in Table 4 the size we recover because (1) the sizes of these models was never previously confirmed, but (2) they have now been deprecated and so disclosing the size is not harmful. In discussions with OpenAI, we decided to withhold disclosure of the size of gpt-3.5-turbo models, but we confirmed with them that the number our attack reported was accurate.
์šฐ๋ฆฌ์˜ hidden-dimension-stealing ๋ฐ entire-layer-stealing ๊ณต๊ฒฉ์€ ์ด ๋‹ค์„ฏ ๊ฐ€์ง€ ๋ชจ๋ธ ๋ชจ๋‘์—์„œ ์ž‘๋™ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ์—์„œ ๋ณต๊ตฌํ•œ ํฌ๊ธฐ๋Š” OpenAI์— ์˜ํ•ด ํ™•์ธ๋œ ๋ฐ”์™€ ๊ฐ™์ด ์›๋ž˜ ๋ชจ๋ธ์˜ ์‹ค์ œ ํฌ๊ธฐ์™€ ์™„๋ฒฝํ•˜๊ฒŒ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. ์ฒซ ์„ธ ๋ชจ๋ธ์— ๋Œ€ํ•ด์„œ๋Š”, (1) ์ด ๋ชจ๋ธ๋“ค์˜ ํฌ๊ธฐ๊ฐ€ ์ด์ „์— ํ™•์ธ๋œ ์ ์ด ์—†์—ˆ๊ณ , (2) ์ด์ œ๋Š” ์‚ฌ์šฉ ์ค‘๋‹จ๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์— ํฌ๊ธฐ๋ฅผ ๊ณต๊ฐœํ•˜๋Š” ๊ฒƒ์ด ํ•ด๋กญ์ง€ ์•Š๋‹ค๋Š” ์ด์œ ๋กœ Table 4์— ๋ณต๊ตฌํ•œ ํฌ๊ธฐ๋ฅผ ๋ณด๊ณ ํ•ฉ๋‹ˆ๋‹ค. OpenAI์™€์˜ ๋…ผ์˜์—์„œ, ์šฐ๋ฆฌ๋Š” gpt-3.5-turbo ๋ชจ๋ธ์˜ ํฌ๊ธฐ ๊ณต๊ฐœ๋ฅผ ๋ณด๋ฅ˜ํ•˜๊ธฐ๋กœ ๊ฒฐ์ •ํ–ˆ์ง€๋งŒ, ์šฐ๋ฆฌ์˜ ๊ณต๊ฒฉ์ด ๋ณด๊ณ ํ•œ ์ˆซ์ž๊ฐ€ ์ •ํ™•ํ•˜๋‹ค๋Š” ๊ฒƒ์„ ๊ทธ๋“ค๊ณผ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค.
์ด ๋‹ค์„ฏ ๊ฐ€์ง€ ๋ชจ๋ธ ๋ชจ๋‘์— ๋Œ€ํ•ด ์ˆจ๊ฒจ์ง„ ์ฐจ์› ํ›”์น˜๊ธฐ์™€ ์ „์ฒด ๋ ˆ์ด์–ด ํ›”์น˜๊ธฐ ๊ณต๊ฒฉ์ด ์„ฑ๊ณตํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ์—์„œ ๋ณต๊ตฌํ•œ ํฌ๊ธฐ๋Š” OpenAI๊ฐ€ ํ™•์ธํ•œ ๋ฐ”์™€ ๊ฐ™์ด ์›๋ž˜ ๋ชจ๋ธ์˜ ์‹ค์ œ ํฌ๊ธฐ์™€ ์™„๋ฒฝํ•˜๊ฒŒ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. ์ฒ˜์Œ ์„ธ ๊ฐ€์ง€ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ (1) ์ด ๋ชจ๋ธ์˜ ํฌ๊ธฐ๊ฐ€ ์ด์ „์— ํ™•์ธ๋œ ์ ์ด ์—†์ง€๋งŒ (2) ์ด์ œ ๋” ์ด์ƒ ์‚ฌ์šฉ๋˜์ง€ ์•Š์œผ๋ฏ€๋กœ ํฌ๊ธฐ๋ฅผ ๊ณต๊ฐœํ•ด๋„ ๋ฌธ์ œ๊ฐ€ ์—†๊ธฐ ๋•Œ๋ฌธ์— ๋ณต๊ตฌํ•œ ํฌ๊ธฐ๋ฅผ ํ‘œ 4์— ๋ณด๊ณ ํ•ฉ๋‹ˆ๋‹ค. OpenAI์™€์˜ ๋…ผ์˜์—์„œ gpt-3.5-turbo ๋ชจ๋ธ์˜ ํฌ๊ธฐ ๊ณต๊ฐœ๋ฅผ ๋ณด๋ฅ˜ํ•˜๊ธฐ๋กœ ๊ฒฐ์ •ํ–ˆ์ง€๋งŒ, ๊ณต๊ฒฉ์ด ๋ณด๊ณ ํ•œ ์ˆซ์ž๊ฐ€ ์ •ํ™•ํ•˜๋‹ค๋Š” ๊ฒƒ์„ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค.
Model merging holds significant potential and democratizes the model-building process, making it accessible to a broad range of participants. However, it heavily relies on intuition and domain knowledge. Human intuition, however, has its limits. With the growing diversity of open models and tasks, we need a more systematic approach. We believe evolutionary algorithms, inspired by natural selection, can unlock more effective merging solutions. These algorithms can explore a vast space of possibilities, discovering novel and counter-intuitive combinations that traditional methods and human intuition might miss.
๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ์ƒ๋‹นํ•œ ์ž ์žฌ๋ ฅ์„ ์ง€๋‹ˆ๊ณ  ์žˆ์œผ๋ฉฐ, ๋ชจ๋ธ ๊ตฌ์ถ• ๊ณผ์ •์„ ๋ฏผ์ฃผํ™”ํ•˜์—ฌ ๋‹ค์–‘ํ•œ ์ฐธ๊ฐ€์ž๋“ค์ด ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋Š” ์ง๊ด€๊ณผ ๋„๋ฉ”์ธ ์ง€์‹์— ํฌ๊ฒŒ ์˜์กดํ•ฉ๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ธ๊ฐ„์˜ ์ง๊ด€์—๋Š” ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐœ๋ฐฉํ˜• ๋ชจ๋ธ๊ณผ ์ž‘์—…์˜ ๋‹ค์–‘์„ฑ์ด ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ, ์šฐ๋ฆฌ๋Š” ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ž์—ฐ ์„ ํƒ์—์„œ ์˜๊ฐ์„ ๋ฐ›์€ ์ง„ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋” ํšจ๊ณผ์ ์ธ ๋ณ‘ํ•ฉ ์†”๋ฃจ์…˜์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๊ด‘๋ฒ”์œ„ํ•œ ๊ฐ€๋Šฅ์„ฑ์˜ ๊ณต๊ฐ„์„ ํƒ์ƒ‰ํ•˜์—ฌ ์ „ํ†ต์ ์ธ ๋ฐฉ๋ฒ•๊ณผ ์ธ๊ฐ„์˜ ์ง๊ด€์ด ๋†“์น  ์ˆ˜ ์žˆ๋Š” ์ƒˆ๋กœ์šด ๋ฐ ๋ฐ˜์ง๊ด€์ ์ธ ์กฐํ•ฉ์„ ๋ฐœ๊ฒฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ์ƒ๋‹นํ•œ ์ž ์žฌ๋ ฅ์„ ๊ฐ€์ง€๊ณ  ์žˆ์œผ๋ฉฐ ๋ชจ๋ธ ๊ตฌ์ถ• ํ”„๋กœ์„ธ์Šค๋ฅผ ๋ฏผ์ฃผํ™”ํ•˜์—ฌ ๊ด‘๋ฒ”์œ„ํ•œ ์ฐธ์—ฌ์ž๊ฐ€ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ด์ค๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ง๊ด€๊ณผ ๋„๋ฉ”์ธ ์ง€์‹์— ํฌ๊ฒŒ ์˜์กดํ•ฉ๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ธ๊ฐ„์˜ ์ง๊ด€์—๋Š” ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜คํ”ˆ ๋ชจ๋ธ๊ณผ ํƒœ์Šคํฌ์˜ ๋‹ค์–‘์„ฑ์ด ์ปค์ง€๋ฉด์„œ ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ž์—ฐ ์„ ํƒ์—์„œ ์˜๊ฐ์„ ์–ป์€ ์ง„ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋ณด๋‹ค ํšจ๊ณผ์ ์ธ ๋ณ‘ํ•ฉ ์†”๋ฃจ์…˜์„ ์ž ๊ธˆ ํ•ด์ œํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ฐฉ๋Œ€ํ•œ ๊ฐ€๋Šฅ์„ฑ์˜ ๊ณต๊ฐ„์„ ํƒ์ƒ‰ํ•˜์—ฌ ๊ธฐ์กด ๋ฐฉ๋ฒ•๊ณผ ์ธ๊ฐ„์˜ ์ง๊ด€์ด ๋†“์น  ์ˆ˜ ์žˆ๋Š” ์ƒˆ๋กญ๊ณ  ์ง๊ด€์ ์ด์ง€ ์•Š์€ ์กฐํ•ฉ์„ ๋ฐœ๊ฒฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Language models produce a distribution over the next token; can we use this to recover the prompt tokens? We consider the problem of language model inversion and show that next-token probabilities contain a surprising amount of information about the preceding text. Often we can recover the text in cases where it is hidden from the user, motivating a method for recovering unknown prompts given only the model's current distribution output. We consider a variety of model access scenarios, and show how even without predictions for every token in the vocabulary we can recover the probability vector through search.
์–ธ์–ด ๋ชจ๋ธ์€ ๋‹ค์Œ ํ† ํฐ์— ๋Œ€ํ•œ ๋ถ„ํฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํ”„๋กฌํ”„ํŠธ ํ† ํฐ์„ ๋ณต์›ํ•  ์ˆ˜ ์žˆ์„๊นŒ์š”? ์šฐ๋ฆฌ๋Š” ์–ธ์–ด ๋ชจ๋ธ ์—ญ์ „ ๋ฌธ์ œ๋ฅผ ๊ณ ๋ คํ•˜๊ณ , ๋‹ค์Œ ํ† ํฐ ํ™•๋ฅ ์ด ์ด์ „ ํ…์ŠคํŠธ์— ๋Œ€ํ•œ ๋†€๋ผ์šด ์–‘์˜ ์ •๋ณด๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ข…์ข… ์‚ฌ์šฉ์ž๊ฐ€ ํ…์ŠคํŠธ๋ฅผ ๋ณผ ์ˆ˜ ์—†๋Š” ๊ฒฝ์šฐ์—๋„ ํ…์ŠคํŠธ๋ฅผ ๋ณต์›ํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” ๋ชจ๋ธ์˜ ํ˜„์žฌ ๋ถ„ํฌ ์ถœ๋ ฅ๋งŒ์œผ๋กœ ์•Œ๋ ค์ง€์ง€ ์•Š์€ ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋ณต์›ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ๋ชจ๋ธ ์ ‘๊ทผ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๊ณ ๋ คํ•˜๊ณ , ์–ดํœ˜์˜ ๋ชจ๋“  ํ† ํฐ์— ๋Œ€ํ•œ ์˜ˆ์ธก์ด ์—†์–ด๋„ ๊ฒ€์ƒ‰์„ ํ†ตํ•ด ํ™•๋ฅ  ๋ฒกํ„ฐ๋ฅผ ๋ณต์›ํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
์–ธ์–ด ๋ชจ๋ธ์€ ๋‹ค์Œ ํ† ํฐ์— ๋Œ€ํ•œ ๋ถ„ํฌ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํ”„๋กฌํ”„ํŠธ ํ† ํฐ์„ ๋ณต๊ตฌํ•  ์ˆ˜ ์žˆ์„๊นŒ์š”? ์–ธ์–ด ๋ชจ๋ธ ๋ฐ˜์ „ ๋ฌธ์ œ๋ฅผ ๊ณ ๋ คํ•˜๊ณ  ๋‹ค์Œ ํ† ํฐ ํ™•๋ฅ ์— ์ด์ „ ํ…์ŠคํŠธ์— ๋Œ€ํ•œ ๋†€๋ผ์šด ์–‘์˜ ์ •๋ณด๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ข…์ข… ์‚ฌ์šฉ์ž์—๊ฒŒ ์ˆจ๊ฒจ์ ธ ์žˆ๋Š” ๊ฒฝ์šฐ์—๋„ ํ…์ŠคํŠธ๋ฅผ ๋ณต๊ตฌํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” ๋ชจ๋ธ์˜ ํ˜„์žฌ ๋ถ„ํฌ ์ถœ๋ ฅ๋งŒ ์ฃผ์–ด์กŒ์„ ๋•Œ ์•Œ ์ˆ˜ ์—†๋Š” ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋ณต๊ตฌํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๋™๊ธฐํ™”ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ๋ชจ๋ธ ์•ก์„ธ์Šค ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๊ณ ๋ คํ•˜๊ณ  ์–ดํœ˜์˜ ๋ชจ๋“  ํ† ํฐ์— ๋Œ€ํ•œ ์˜ˆ์ธก์ด ์—†๋”๋ผ๋„ ๊ฒ€์ƒ‰์„ ํ†ตํ•ด ํ™•๋ฅ  ๋ฒกํ„ฐ๋ฅผ ๋ณต๊ตฌํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
* **JA-VG-VQA-500:** A 500-sample test set extracted from the Japanese Visual Genome VQA dataset [42]. * **JA-VLM-Bench-In-the-Wild:** A Japanese version of LLaVA-Bench-In-the-Wild [32]. We compiled a rich collection of 42 images, accompanied by a total of 50 questions, featuring a variety of Japanese cultural elements and objects found in Japan. The QAs were crafted with the assistance of GPT-4V [1] and underwent a human-in-the-loop filtering process to eliminate nonsensical outcomes. Compared to the JA-VG-VQA-500 dataset, our set poses more complex challenges, demanding more nuanced and detailed responses.
* **JA-VG-VQA-500:** ์ผ๋ณธ์–ด Visual Genome VQA ๋ฐ์ดํ„ฐ์…‹ [42]์—์„œ ์ถ”์ถœํ•œ 500๊ฐœ์˜ ์ƒ˜ํ”Œ ํ…Œ์ŠคํŠธ ์„ธํŠธ์ž…๋‹ˆ๋‹ค. * **JA-VLM-Bench-In-the-Wild:** LLaVA-Bench-In-the-Wild [32]์˜ ์ผ๋ณธ์–ด ๋ฒ„์ „์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ผ๋ณธ์˜ ๋‹ค์–‘ํ•œ ๋ฌธํ™” ์š”์†Œ์™€ ๋ฌผ์ฒด๋ฅผ ํŠน์ง•์œผ๋กœ ํ•˜๋Š” ์ด 50๊ฐœ์˜ ์งˆ๋ฌธ๊ณผ ํ•จ๊ป˜ 42๊ฐœ์˜ ์ด๋ฏธ์ง€๋ฅผ ํ’๋ถ€ํ•˜๊ฒŒ ์ˆ˜์ง‘ํ–ˆ์Šต๋‹ˆ๋‹ค. QAs๋Š” GPT-4V [1]์˜ ๋„์›€์„ ๋ฐ›์•„ ์ œ์ž‘๋˜์—ˆ์œผ๋ฉฐ, ๋น„ํ•ฉ๋ฆฌ์ ์ธ ๊ฒฐ๊ณผ๋ฅผ ์ œ๊ฑฐํ•˜๊ธฐ ์œ„ํ•ด ์ธ๊ฐ„์ด ์ฐธ์—ฌํ•˜๋Š” ํ•„ํ„ฐ๋ง ๊ณผ์ •์„ ๊ฑฐ์ณค์Šต๋‹ˆ๋‹ค. JA-VG-VQA-500 ๋ฐ์ดํ„ฐ์…‹๊ณผ ๋น„๊ตํ•  ๋•Œ, ์šฐ๋ฆฌ์˜ ์„ธํŠธ๋Š” ๋” ๋ณต์žกํ•œ ๋„์ „์„ ์ œ์‹œํ•˜๋ฉฐ, ๋” ๋ฏธ๋ฌ˜ํ•˜๊ณ  ์ƒ์„ธํ•œ ์‘๋‹ต์„ ์š”๊ตฌํ•ฉ๋‹ˆ๋‹ค.
* **JA-VG-VQA-500:** ์ผ๋ณธ ๋น„์ฃผ์–ผ ์ง€๋…ธ๋ฉ” VQA ๋ฐ์ดํ„ฐ ์„ธํŠธ[42]์—์„œ ์ถ”์ถœํ•œ 500๊ฐœ ์ƒ˜ํ”Œ ํ…Œ์ŠคํŠธ ์„ธํŠธ์ž…๋‹ˆ๋‹ค. * **JA-VLM-Bench-In-the-Wild:** LLaVA-Bench-In-the-Wild [32]์˜ ์ผ๋ณธ์–ด ๋ฒ„์ „์ž…๋‹ˆ๋‹ค. ์ผ๋ณธ์—์„œ ๋ฐœ๊ฒฌ๋˜๋Š” ๋‹ค์–‘ํ•œ ์ผ๋ณธ ๋ฌธํ™” ์š”์†Œ์™€ ์‚ฌ๋ฌผ์ด ๋“ฑ์žฅํ•˜๋Š” ์ด 50๊ฐœ์˜ ์งˆ๋ฌธ์ด ํฌํ•จ๋œ 42๊ฐœ์˜ ์ด๋ฏธ์ง€๋ฅผ ๋ชจ์•˜์Šต๋‹ˆ๋‹ค. ์งˆ๋ฌธ๊ณผ ๋‹ต๋ณ€์€ GPT-4V [1]์˜ ๋„์›€์„ ๋ฐ›์•„ ์ž‘์„ฑ๋˜์—ˆ์œผ๋ฉฐ, ๋ง๋„ ์•ˆ ๋˜๋Š” ๊ฒฐ๊ณผ๋ฅผ ์ œ๊ฑฐํ•˜๊ธฐ ์œ„ํ•ด ์ธ๊ฐ„์ด ์ฐธ์—ฌํ•˜๋Š” ํ•„ํ„ฐ๋ง ๊ณผ์ •์„ ๊ฑฐ์ณค์Šต๋‹ˆ๋‹ค. JA-VG-VQA-500 ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋น„ํ•ด ๋” ๋ณต์žกํ•œ ๊ณผ์ œ๋ฅผ ์ œ์‹œํ•˜์—ฌ ๋ณด๋‹ค ๋ฏธ๋ฌ˜ํ•˜๊ณ  ์„ธ๋ฐ€ํ•œ ๋‹ต๋ณ€์„ ์š”๊ตฌํ•ฉ๋‹ˆ๋‹ค.
Examining the parameters individually, \(E\) and \(\beta\) show highly significant differences, with \(p\)-values of \(1.5\times 10^{-6}\) and \(4.3\times 10^{-5}\), respectively. These small \(p\)-values indicate that the estimated values for \(E\) and \(\beta\) are significantly different from their respective true values, suggesting substantial deviations in these parameters' estimates from expected outcomes based on the model. On the other hand, the parameters \(A\), \(B\), and \(\alpha\) have estimated values that are not statistically significantly different from the values reported in Hoffmann et al.
๊ฐœ๋ณ„์ ์œผ๋กœ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๊ฒ€ํ† ํ•œ ๊ฒฐ๊ณผ, \(E\)์™€ \(\beta\)๋Š” ๊ฐ๊ฐ \(p\)-๊ฐ’์ด \(1.5\times 10^{-6}\) ๋ฐ \(4.3\times 10^{-5}\)๋กœ ๋งค์šฐ ์œ ์˜๋ฏธํ•œ ์ฐจ์ด๋ฅผ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ž‘์€ \(p\)-๊ฐ’์€ \(E\)์™€ \(\beta\)์˜ ์ถ”์ •๊ฐ’์ด ๊ฐ๊ฐ์˜ ์‹ค์ œ ๊ฐ’๊ณผ ์œ ์˜๋ฏธํ•˜๊ฒŒ ๋‹ค๋ฅด๋‹ค๋Š” ๊ฒƒ์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ด๋Š” ๋ชจ๋ธ์— ๊ธฐ๋ฐ˜ํ•œ ์˜ˆ์ƒ ๊ฒฐ๊ณผ์™€ ๋น„๊ตํ–ˆ์„ ๋•Œ ์ด๋“ค ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ์ถ”์ •์น˜์— ์ƒ๋‹นํ•œ ํŽธ์ฐจ๊ฐ€ ์žˆ์Œ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, \(A\), \(B\), \(\alpha\) ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ์ถ”์ •๊ฐ’์€ Hoffmann ๋“ฑ์—์„œ ๋ณด๊ณ ๋œ ๊ฐ’๊ณผ ํ†ต๊ณ„์ ์œผ๋กœ ์œ ์˜๋ฏธํ•˜๊ฒŒ ๋‹ค๋ฅด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๊ฐœ๋ณ„์ ์œผ๋กœ ์‚ดํŽด๋ณด๋ฉด, \(E\)์™€ \(\beta\)๋Š” ๊ฐ๊ฐ \(1.5\๋ฐฐ 10^{-6}\)์™€ \(4.3\๋ฐฐ 10^{-5}\)์˜ \(p\)-๊ฐ’์œผ๋กœ ๋งค์šฐ ์œ ์˜๋ฏธํ•œ ์ฐจ์ด๋ฅผ ๋ณด์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ž‘์€ \(p\)-๊ฐ’์€ \(E\)์™€ \(\beta\)์˜ ์ถ”์ •๊ฐ’์ด ๊ฐ๊ฐ์˜ ์‹ค์ œ ๊ฐ’๊ณผ ์ƒ๋‹นํžˆ ๋‹ค๋ฅด๋‹ค๋Š” ๊ฒƒ์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ด๋Š” ๋ชจ๋ธ์— ๊ธฐ๋ฐ˜ํ•œ ์˜ˆ์ƒ ๊ฒฐ๊ณผ์—์„œ ์ด๋Ÿฌํ•œ ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ์ถ”์ •์น˜๊ฐ€ ์ƒ๋‹นํ•œ ํŽธ์ฐจ๋ฅผ ๋ณด์ž„์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ๋งค๊ฐœ๋ณ€์ˆ˜ \(A\), \(B\), \(\alpha\)์˜ ์ถ”์ •๊ฐ’์€ Hoffmann ๋“ฑ์—์„œ ๋ณด๊ณ ํ•œ ๊ฐ’๊ณผ ํ†ต๊ณ„์ ์œผ๋กœ ์œ ์˜๋ฏธํ•œ ์ฐจ์ด๊ฐ€ ์—†๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค.
Advances in long-context language models have sparked interest in summarizing book-length documents (\(>\)100K tokens). Despite the importance of faithfulness and content relevance for summary quality, recent work in this regime focuses only on input-agnostic aspects like coherence (Chang et al., 2023b). This is due to the length and complexity of the input documents: hiring human annotators to read and understand them is expensive and time-consuming. Our work fills this gap by presenting the first large-scale human evaluation of faithfulness and other content selection errors in book-length summarization.
๊ธด ๋ฌธ๋งฅ ์–ธ์–ด ๋ชจ๋ธ์˜ ๋ฐœ์ „์€ ์ฑ… ๊ธธ์ด์˜ ๋ฌธ์„œ(100K ํ† ํฐ ์ด์ƒ)๋ฅผ ์š”์•ฝํ•˜๋Š” ๋ฐ ๋Œ€ํ•œ ๊ด€์‹ฌ์„ ๋ถˆ๋Ÿฌ์ผ์œผ์ผฐ์Šต๋‹ˆ๋‹ค. ์š”์•ฝ ํ’ˆ์งˆ์— ์žˆ์–ด ์ถฉ์‹ค์„ฑ๊ณผ ๋‚ด์šฉ ๊ด€๋ จ์„ฑ์˜ ์ค‘์š”์„ฑ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , ์ตœ๊ทผ์˜ ์—ฐ๊ตฌ๋Š” ์ผ๊ด€์„ฑ ๊ฐ™์€ ์ž…๋ ฅ๊ณผ ๋ฌด๊ด€ํ•œ ์ธก๋ฉด์—๋งŒ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค(Chang et al., 2023b). ์ด๋Š” ์ž…๋ ฅ ๋ฌธ์„œ์˜ ๊ธธ์ด์™€ ๋ณต์žก์„ฑ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค: ์‚ฌ๋žŒ ํ‰๊ฐ€์ž๋ฅผ ๊ณ ์šฉํ•˜์—ฌ ์ด๋ฅผ ์ฝ๊ณ  ์ดํ•ดํ•˜๋Š” ๊ฒƒ์€ ๋น„์šฉ์ด ๋งŽ์ด ๋“ค๊ณ  ์‹œ๊ฐ„์ด ๋งŽ์ด ์†Œ์š”๋ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ๋Š” ์ฑ… ๊ธธ์ด ์š”์•ฝ์—์„œ ์ถฉ์‹ค์„ฑ๊ณผ ๊ธฐํƒ€ ๋‚ด์šฉ ์„ ํƒ ์˜ค๋ฅ˜์— ๋Œ€ํ•œ ์ตœ์ดˆ์˜ ๋Œ€๊ทœ๋ชจ ์ธ๊ฐ„ ํ‰๊ฐ€๋ฅผ ์ œ์‹œํ•จ์œผ๋กœ์จ ์ด ๊ฒฉ์ฐจ๋ฅผ ๋ฉ”์›๋‹ˆ๋‹ค.
๊ธด ๋ฌธ๋งฅ ์–ธ์–ด ๋ชจ๋ธ์˜ ๋ฐœ์ „์œผ๋กœ ์ฑ… ๊ธธ์ด์˜ ๋ฌธ์„œ(\(>\)100K ํ† ํฐ)๋ฅผ ์š”์•ฝํ•˜๋Š” ๊ฒƒ์— ๋Œ€ํ•œ ๊ด€์‹ฌ์ด ๋†’์•„์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์š”์•ฝ ํ’ˆ์งˆ์— ์žˆ์–ด ์ถฉ์‹ค์„ฑ๊ณผ ์ฝ˜ํ…์ธ  ๊ด€๋ จ์„ฑ์ด ์ค‘์š”ํ•จ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , ์ตœ๊ทผ ์ด ๋ถ„์•ผ์˜ ์—ฐ๊ตฌ๋Š” ์ผ๊ด€์„ฑ(Chang et al., 2023b)๊ณผ ๊ฐ™์€ ์ž…๋ ฅ์— ๊ตฌ์• ๋ฐ›์ง€ ์•Š๋Š” ์ธก๋ฉด์—๋งŒ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ž…๋ ฅ ๋ฌธ์„œ์˜ ๊ธธ์ด์™€ ๋ณต์žก์„ฑ ๋•Œ๋ฌธ์ธ๋ฐ, ์ธ๊ฐ„ ์ฃผ์„๊ฐ€๊ฐ€ ์ด๋ฅผ ์ฝ๊ณ  ์ดํ•ดํ•˜๋„๋ก ๊ณ ์šฉํ•˜๋Š” ๊ฒƒ์€ ๋น„์šฉ๊ณผ ์‹œ๊ฐ„์ด ๋งŽ์ด ์†Œ์š”๋ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฉ์ฐจ๋ฅผ ๋ฉ”์šฐ๊ธฐ ์œ„ํ•ด ์ €์ž๋Š” ์ฑ… ๊ธธ์ด์˜ ์š”์•ฝ์—์„œ ์ถฉ์‹ค์„ฑ ๋ฐ ๊ธฐํƒ€ ์ฝ˜ํ…์ธ  ์„ ํƒ ์˜ค๋ฅ˜์— ๋Œ€ํ•œ ์ตœ์ดˆ์˜ ๋Œ€๊ทœ๋ชจ ์ธ๊ฐ„ ํ‰๊ฐ€๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.
We choose \(11000\) clusters following previous work [1] and we note that this choice sticks to the heuristic that the number of clusters should roughly be the square root of the number of total points being clustered. We also note that in initial experiments for data selection at the 125M OPT model scale, we did not find a significant effect of number of clusters on the performance of our data selection methods (see Figure A1) this finding agrees with Abbas et al. [1] who notice significant overlap between datasets selected by SemDeDup with different number of clusters (see Figure A2 in Abbas et al. [1]).
์ด์ „ ์—ฐ๊ตฌ [1]๋ฅผ ๋”ฐ๋ผ \(11000\)๊ฐœ์˜ ํด๋Ÿฌ์Šคํ„ฐ๋ฅผ ์„ ํƒํ–ˆ์œผ๋ฉฐ, ์ด๋Š” ํด๋Ÿฌ์Šคํ„ฐ์˜ ์ˆ˜๊ฐ€ ํด๋Ÿฌ์Šคํ„ฐ๋ง๋˜๋Š” ์ด ์ ์˜ ์ˆ˜์˜ ๋Œ€๋žต ์ œ๊ณฑ๊ทผ์ด์–ด์•ผ ํ•œ๋‹ค๋Š” ํœด๋ฆฌ์Šคํ‹ฑ์— ๋ถ€ํ•ฉํ•จ์„ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ 125M OPT ๋ชจ๋ธ ๊ทœ๋ชจ์—์„œ ๋ฐ์ดํ„ฐ ์„ ํƒ์„ ์œ„ํ•œ ์ดˆ๊ธฐ ์‹คํ—˜์—์„œ ํด๋Ÿฌ์Šคํ„ฐ ์ˆ˜๊ฐ€ ๋ฐ์ดํ„ฐ ์„ ํƒ ๋ฐฉ๋ฒ•์˜ ์„ฑ๋Šฅ์— ํฐ ์˜ํ–ฅ์„ ๋ฏธ์น˜์ง€ ์•Š๋Š”๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค(๊ทธ๋ฆผ A1 ์ฐธ์กฐ). ์ด ๋ฐœ๊ฒฌ์€ SemDeDup์ด ๋‹ค๋ฅธ ํด๋Ÿฌ์Šคํ„ฐ ์ˆ˜๋กœ ์„ ํƒํ•œ ๋ฐ์ดํ„ฐ์…‹ ๊ฐ„์— ์ƒ๋‹นํ•œ ์ค‘๋ณต์ด ์žˆ์Œ์„ ๋ฐœ๊ฒฌํ•œ Abbas et al. [1]์˜ ์—ฐ๊ตฌ์™€ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค(Abbas et al. [1]์˜ ๊ทธ๋ฆผ A2 ์ฐธ์กฐ).
์ด์ „ ์ž‘์—… [1]์— ๋”ฐ๋ผ \(11000\) ํด๋Ÿฌ์Šคํ„ฐ๋ฅผ ์„ ํƒํ•˜๊ณ , ํด๋Ÿฌ์Šคํ„ฐ ์ˆ˜๋Š” ํด๋Ÿฌ์Šคํ„ฐ๋ง๋˜๋Š” ์ด ์  ์ˆ˜์˜ ์ œ๊ณฑ๊ทผ์ด ๋˜์–ด์•ผ ํ•œ๋‹ค๋Š” ํœด๋ฆฌ์Šคํ‹ฑ์„ ๋”ฐ๋ฅด๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ 125M OPT ๋ชจ๋ธ ๊ทœ๋ชจ์—์„œ ๋ฐ์ดํ„ฐ ์„ ํƒ์„ ์œ„ํ•œ ์ดˆ๊ธฐ ์‹คํ—˜์—์„œ ํด๋Ÿฌ์Šคํ„ฐ ์ˆ˜๊ฐ€ ๋ฐ์ดํ„ฐ ์„ ํƒ ๋ฐฉ๋ฒ•์˜ ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์ด ํฌ์ง€ ์•Š๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์œผ๋ฉฐ(๊ทธ๋ฆผ A1 ์ฐธ์กฐ), ์ด ๊ฒฐ๊ณผ๋Š” Abbas ๋“ฑ[1]์˜ ์—ฐ๊ตฌ์—์„œ๋„ SemDeDup์— ์˜ํ•ด ์„ ํƒ๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ ๊ฐ„์— ํด๋Ÿฌ์Šคํ„ฐ ์ˆ˜๊ฐ€ ๋‹ค๋ฅด๋”๋ผ๋„ ์ƒ๋‹นํ•œ ์ค‘๋ณต์ด ์žˆ์Œ์„ ๋ฐœ๊ฒฌํ•œ ๊ฒƒ๊ณผ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค(Abbas ๋“ฑ[1]์˜ ๊ทธ๋ฆผ A2 ์ฐธ์กฐ).
Due to the diversity of possibilities in human language, it is rare for the same idea to be expressed identically in multiple documents unless one expression is derived from the other, or both are quoting from a shared source. This observation motivates deduplicating exact substrings. We call our approach ExactSubstr. When two examples \(x_{i}\) and \(x_{j}\) share a sufficiently long substring (that is, a substring for which \(x_{i}^{a.a.+k}=x_{j}^{b.b+k}\)), that substring is removed from one of them. Based on statistical analyses (SSB), we select \(k=50\) tokens as the minimum matching substring length.
์ธ๊ฐ„ ์–ธ์–ด์˜ ๋‹ค์–‘์„ฑ ๋•Œ๋ฌธ์— ๋™์ผํ•œ ์•„์ด๋””์–ด๊ฐ€ ์—ฌ๋Ÿฌ ๋ฌธ์„œ์—์„œ ๋™์ผํ•˜๊ฒŒ ํ‘œํ˜„๋˜๋Š” ๊ฒฝ์šฐ๋Š” ๋“œ๋ญ…๋‹ˆ๋‹ค. ์ด๋Š” ํ•œ ํ‘œํ˜„์ด ๋‹ค๋ฅธ ํ‘œํ˜„์—์„œ ํŒŒ์ƒ๋˜์—ˆ๊ฑฐ๋‚˜, ๋‘ ํ‘œํ˜„ ๋ชจ๋‘ ๊ณตํ†ต๋œ ์ถœ์ฒ˜๋ฅผ ์ธ์šฉํ•˜๋Š” ๊ฒฝ์šฐ๋ฅผ ์ œ์™ธํ•˜๊ณ ๋Š” ๊ฑฐ์˜ ๋ฐœ์ƒํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ด€์ฐฐ์€ ์ •ํ™•ํ•œ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด์„ ์ค‘๋ณต ์ œ๊ฑฐํ•˜๋Š” ๋™๊ธฐ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ์ ‘๊ทผ ๋ฐฉ์‹์„ ExactSubstr๋ผ๊ณ  ๋ถ€๋ฆ…๋‹ˆ๋‹ค. ๋‘ ์˜ˆ์ œ \(x_{i}\)์™€ \(x_{j}\)๊ฐ€ ์ถฉ๋ถ„ํžˆ ๊ธด ๋ถ€๋ถ„ ๋ฌธ์ž์—ด์„ ๊ณต์œ ํ•  ๋•Œ (์ฆ‰, \(x_{i}^{a.a.+k}=x_{j}^{b.b+k}\)์ธ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด), ๊ทธ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด์€ ๋‘˜ ์ค‘ ํ•˜๋‚˜์—์„œ ์ œ๊ฑฐ๋ฉ๋‹ˆ๋‹ค. ํ†ต๊ณ„ ๋ถ„์„(SSB)์„ ๊ธฐ๋ฐ˜์œผ๋กœ, ์ตœ์†Œ ์ผ์น˜ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด ๊ธธ์ด๋กœ \(k=50\) ํ† ํฐ์„ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค.
์ธ๊ฐ„ ์–ธ์–ด์˜ ๊ฐ€๋Šฅ์„ฑ์€ ๋งค์šฐ ๋‹ค์–‘ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ํ•œ ํ‘œํ˜„์ด ๋‹ค๋ฅธ ํ‘œํ˜„์—์„œ ํŒŒ์ƒ๋˜์—ˆ๊ฑฐ๋‚˜ ๋‘ ํ‘œํ˜„์ด ๊ณต์œ  ์†Œ์Šค์—์„œ ์ธ์šฉํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์•„๋‹ˆ๋ผ๋ฉด ๋™์ผํ•œ ์•„์ด๋””์–ด๊ฐ€ ์—ฌ๋Ÿฌ ๋ฌธ์„œ์—์„œ ๋™์ผํ•˜๊ฒŒ ํ‘œํ˜„๋˜๋Š” ๊ฒฝ์šฐ๋Š” ๋“œ๋ญ…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ด€์ฐฐ์€ ์ •ํ™•ํ•œ ํ•˜์œ„ ๋ฌธ์ž์—ด์„ ์ค‘๋ณต ์ œ๊ฑฐํ•˜๋Š” ๋™๊ธฐ๋ฅผ ๋ถ€์—ฌํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ์ ‘๊ทผ ๋ฐฉ์‹์„ ExactSubstr์ด๋ผ๊ณ  ๋ถ€๋ฆ…๋‹ˆ๋‹ค. ๋‘ ์˜ˆ์ œ \(x_{i}\)์™€ \(x_{j}\)๊ฐ€ ์ถฉ๋ถ„ํžˆ ๊ธด ํ•˜์œ„ ๋ฌธ์ž์—ด(์ฆ‰, \(x_{i}^{a.a.+k}=x_{j}^{b.b+k}\))์„ ๊ณต์œ ํ•˜๋Š” ๊ฒฝ์šฐ, ๊ทธ ์ค‘ ํ•˜๋‚˜์—์„œ ํ•ด๋‹น ํ•˜์œ„ ๋ฌธ์ž์—ด์ด ์ œ๊ฑฐ๋ฉ๋‹ˆ๋‹ค. ํ†ต๊ณ„ ๋ถ„์„(SSB)์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์ตœ์†Œ ์ผ์น˜ ํ•˜์œ„ ๋ฌธ์ž์—ด ๊ธธ์ด๋กœ \(k=50\) ํ† ํฐ์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.
We compare Nemotron-4 15B to existing multilingual language models: XGLM (Lin et al., 2022), mGPT (Shliazhko et al., 2022), and BLOOM (Scao et al., 2023). XGLM and mGPT are models specially trained to have improved multilingual ability by up-sampling the presence of non-English languages in the training data. In contrast, BLOOM, like Nemotron-4, is a general purpose language model that was trained on a combination of English, multilingual, and code data. In Table 7, we clearly see that Nemotron-4 achieves the best performance amongst all models - realizing almost a 12% improvement in the four-shot setting.
์šฐ๋ฆฌ๋Š” Nemotron-4 15B๋ฅผ ๊ธฐ์กด์˜ ๋‹ค๊ตญ์–ด ์–ธ์–ด ๋ชจ๋ธ๋“ค๊ณผ ๋น„๊ตํ•ฉ๋‹ˆ๋‹ค: XGLM (Lin et al., 2022), mGPT (Shliazhko et al., 2022), ๊ทธ๋ฆฌ๊ณ  BLOOM (Scao et al., 2023). XGLM๊ณผ mGPT๋Š” ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์—์„œ ๋น„์˜์–ด๊ถŒ ์–ธ์–ด์˜ ๋น„์ค‘์„ ๋†’์—ฌ ๋‹ค๊ตญ์–ด ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ํŠน๋ณ„ํžˆ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ๋“ค์ž…๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด์—, BLOOM์€ Nemotron-4์™€ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์˜์–ด, ๋‹ค๊ตญ์–ด, ๊ทธ๋ฆฌ๊ณ  ์ฝ”๋“œ ๋ฐ์ดํ„ฐ๋ฅผ ์กฐํ•ฉํ•˜์—ฌ ํ›ˆ๋ จ๋œ ๋ฒ”์šฉ ์–ธ์–ด ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. Table 7์—์„œ ๋ณผ ์ˆ˜ ์žˆ๋“ฏ์ด, Nemotron-4๋Š” ๋ชจ๋“  ๋ชจ๋ธ ์ค‘์—์„œ ์ตœ๊ณ ์˜ ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•˜๋ฉฐ, ๋„ค ์ƒท ์„ค์ •์—์„œ ๊ฑฐ์˜ 12%์˜ ํ–ฅ์ƒ์„ ์ด๋ฃจ์—ˆ์Šต๋‹ˆ๋‹ค.
๋„ค๋ชจํŠธ๋ก -4 15B๋ฅผ ๊ธฐ์กด ๋‹ค๊ตญ์–ด ์–ธ์–ด ๋ชจ๋ธ์ธ XGLM(Lin et al., 2022), mGPT(Shliazhko et al., 2022), BLOOM(Scao et al., 2023)๊ณผ ๋น„๊ตํ•ฉ๋‹ˆ๋‹ค. XGLM๊ณผ mGPT๋Š” ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์— ๋น„์˜์–ด ์–ธ์–ด์˜ ์กด์žฌ ๋น„์œจ์„ ์—…์ƒ˜ํ”Œ๋งํ•˜์—ฌ ๋‹ค๊ตญ์–ด ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ํŠน๋ณ„ํžˆ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ๋„ค๋ชจํŠธ๋ก -4์™€ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ BLOOM์€ ์˜์–ด, ๋‹ค๊ตญ์–ด, ์ฝ”๋“œ ๋ฐ์ดํ„ฐ์˜ ์กฐํ•ฉ์œผ๋กœ ํ›ˆ๋ จ๋œ ๋ฒ”์šฉ ์–ธ์–ด ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. ํ‘œ 7์—์„œ ๋„ค๋ชจํŠธ๋ก -4๊ฐ€ 4์ƒท ์„ค์ •์—์„œ ๊ฑฐ์˜ 12%์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ์‹คํ˜„ํ•˜๋Š” ๋“ฑ ๋ชจ๋“  ๋ชจ๋ธ ์ค‘์—์„œ ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•˜๋Š” ๊ฒƒ์„ ๋ช…ํ™•ํ•˜๊ฒŒ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Specifically, for questions where the LLM's response is incorrect, we annotate them with a special token, \(\langle\text{RET}\rangle\), indicating the need for additional context. Subsequently, we utilize these annotations to construct a new dataset tailored for training purposes, where we teach an LLM to answer directly if it is confident about the answer or to require context it believes is useful for answering the question (see Figure 1). Our hypothesis is that through this training process, the LLM learns to use an IR system when it needs extra context to answera question, thus we name it Adapt-LLM.
ํŠนํžˆ, LLM์˜ ์‘๋‹ต์ด ํ‹€๋ฆฐ ์งˆ๋ฌธ์— ๋Œ€ํ•ด์„œ๋Š” ์ถ”๊ฐ€์ ์ธ ๋งฅ๋ฝ์ด ํ•„์š”ํ•จ์„ ๋‚˜ํƒ€๋‚ด๋Š” ํŠน์ˆ˜ ํ† ํฐ \(\langle\text{RET}\rangle\)์œผ๋กœ ์ฃผ์„์„ ๋‹ฌ์•˜์Šต๋‹ˆ๋‹ค. ์ดํ›„, ์ด๋Ÿฌํ•œ ์ฃผ์„์„ ํ™œ์šฉํ•˜์—ฌ ํ›ˆ๋ จ ๋ชฉ์ ์— ๋งž์ถ˜ ์ƒˆ๋กœ์šด ๋ฐ์ดํ„ฐ์…‹์„ ๊ตฌ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ ์šฐ๋ฆฌ๋Š” LLM์ด ๋‹ต๋ณ€์— ์ž์‹ ์ด ์žˆ์„ ๋•Œ๋Š” ์ง์ ‘ ๋‹ต๋ณ€ํ•˜๋„๋ก ํ•˜๊ณ , ์งˆ๋ฌธ์— ๋‹ตํ•˜๊ธฐ ์œ„ํ•ด ์œ ์šฉํ•˜๋‹ค๊ณ  ์ƒ๊ฐํ•˜๋Š” ๋งฅ๋ฝ์ด ํ•„์š”ํ•  ๋•Œ๋Š” ์ด๋ฅผ ์š”๊ตฌํ•˜๋„๋ก ๊ฐ€๋ฅด์นฉ๋‹ˆ๋‹ค(๊ทธ๋ฆผ 1 ์ฐธ์กฐ). ์šฐ๋ฆฌ์˜ ๊ฐ€์„ค์€ ์ด ํ›ˆ๋ จ ๊ณผ์ •์„ ํ†ตํ•ด LLM์ด ์งˆ๋ฌธ์— ๋‹ตํ•˜๊ธฐ ์œ„ํ•ด ์ถ”๊ฐ€์ ์ธ ๋งฅ๋ฝ์ด ํ•„์š”ํ•  ๋•Œ IR ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฒ•์„ ๋ฐฐ์šฐ๊ฒŒ ๋œ๋‹ค๋Š” ๊ฒƒ์ด๋ฉฐ, ๋”ฐ๋ผ์„œ ์ด๋ฅผ Adapt-LLM์ด๋ผ๊ณ  ๋ช…๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
๊ตฌ์ฒด์ ์œผ๋กœ, LLM์˜ ์‘๋‹ต์ด ์ž˜๋ชป๋œ ์งˆ๋ฌธ์— ๋Œ€ํ•ด์„œ๋Š” ์ถ”๊ฐ€ ์ปจํ…์ŠคํŠธ๊ฐ€ ํ•„์š”ํ•จ์„ ๋‚˜ํƒ€๋‚ด๋Š” ํŠน์ˆ˜ ํ† ํฐ(\(\langle\text{RET}\rangle\))์œผ๋กœ ์ฃผ์„์„ ๋‹ฌ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ ์ด๋Ÿฌํ•œ ์ฃผ์„์„ ํ™œ์šฉํ•˜์—ฌ ํ›ˆ๋ จ์šฉ์œผ๋กœ ๋งž์ถคํ™”๋œ ์ƒˆ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๊ตฌ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ๋Š” LLM์ด ๋‹ต๋ณ€์— ๋Œ€ํ•ด ํ™•์‹ ์ด ์žˆ์œผ๋ฉด ์ง์ ‘ ๋‹ต๋ณ€ํ•˜๊ฑฐ๋‚˜ ์งˆ๋ฌธ์— ๋‹ตํ•˜๋Š” ๋ฐ ์œ ์šฉํ•˜๋‹ค๊ณ  ์ƒ๊ฐ๋˜๋Š” ์ปจํ…์ŠคํŠธ๋ฅผ ์š”๊ตฌํ•˜๋„๋ก ๊ฐ€๋ฅด์นฉ๋‹ˆ๋‹ค(๊ทธ๋ฆผ 1 ์ฐธ์กฐ). ์ด ํ›ˆ๋ จ ๊ณผ์ •์„ ํ†ตํ•ด LLM์ด ์งˆ๋ฌธ์— ๋‹ตํ•˜๊ธฐ ์œ„ํ•ด ์ถ”๊ฐ€ ์ปจํ…์ŠคํŠธ๊ฐ€ ํ•„์š”ํ•  ๋•Œ IR ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ํ•™์Šตํ•œ๋‹ค๋Š” ๊ฐ€์„ค์„ ์„ธ์šฐ๊ณ , ์ด๋ฅผ Adapt-LLM์ด๋ผ๊ณ  ๋ช…๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
**Code.** To collect code data, existing work mainly crawbs open-source licensed codes from the Internet. Two major sources are public code repositories under open-source licenses (_e.g._, GitHub) and code-related question-answering platforms (_e.g._, StackOverflow). Google has publicly released the BigQuery dataset [160], which includes a substantial number of open-source licensed code snippets in various programming languages, serving as a representative code dataset. CodeGen has utilized BIGQUERY [86], a subset of the BigQuery dataset, for training the multilingual version of CodeGen (CodeGen-Multi).
**์ฝ”๋“œ.** ์ฝ”๋“œ ๋ฐ์ดํ„ฐ๋ฅผ ์ˆ˜์ง‘ํ•˜๊ธฐ ์œ„ํ•ด ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” ์ฃผ๋กœ ์ธํ„ฐ๋„ท์—์„œ ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค๊ฐ€ ๋ถ€์—ฌ๋œ ์ฝ”๋“œ๋ฅผ ํฌ๋กค๋งํ•ฉ๋‹ˆ๋‹ค. ์ฃผ์š” ์ถœ์ฒ˜๋Š” ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค ํ•˜์— ๊ณต๊ฐœ๋œ ์ฝ”๋“œ ์ €์žฅ์†Œ (_์˜ˆ:_, GitHub)์™€ ์ฝ”๋“œ ๊ด€๋ จ ์งˆ๋ฌธ-๋‹ต๋ณ€ ํ”Œ๋žซํผ (_์˜ˆ:_, StackOverflow)์ž…๋‹ˆ๋‹ค. Google์€ ๋‹ค์–‘ํ•œ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์–ธ์–ด๋กœ ์ž‘์„ฑ๋œ ์ƒ๋‹น์ˆ˜์˜ ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค ์ฝ”๋“œ ์Šค๋‹ˆํŽซ์„ ํฌํ•จํ•˜๋Š” BigQuery ๋ฐ์ดํ„ฐ์…‹ [160]์„ ๊ณต๊ฐœ์ ์œผ๋กœ ์ถœ์‹œํ–ˆ์œผ๋ฉฐ, ์ด๋Š” ๋Œ€ํ‘œ์ ์ธ ์ฝ”๋“œ ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. CodeGen์€ CodeGen์˜ ๋‹ค๊ตญ์–ด ๋ฒ„์ „(CodeGen-Multi)์„ ํ›ˆ๋ จํ•˜๊ธฐ ์œ„ํ•ด BigQuery ๋ฐ์ดํ„ฐ์…‹์˜ ํ•˜์œ„ ์ง‘ํ•ฉ์ธ BIGQUERY [86]๋ฅผ ํ™œ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.
**์ฝ”๋“œ**. ์ฝ”๋“œ ๋ฐ์ดํ„ฐ๋ฅผ ์ˆ˜์ง‘ํ•˜๊ธฐ ์œ„ํ•ด ๊ธฐ์กด ์—ฐ๊ตฌ์—์„œ๋Š” ์ฃผ๋กœ ์ธํ„ฐ๋„ท์—์„œ ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค ์ฝ”๋“œ๋ฅผ ํฌ๋กค๋งํ•ฉ๋‹ˆ๋‹ค. ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ์†Œ์Šค๋Š” ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค์— ๋”ฐ๋ฅธ ๊ณต๊ฐœ ์ฝ”๋“œ ๋ฆฌํฌ์ง€ํ† ๋ฆฌ(_์˜ˆ: GitHub)์™€ ์ฝ”๋“œ ๊ด€๋ จ ์งˆ์˜์‘๋‹ต ํ”Œ๋žซํผ(_์˜ˆ: StackOverflow)์ž…๋‹ˆ๋‹ค. Google์€ ๋‹ค์–‘ํ•œ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์–ธ์–ด๋กœ ์ž‘์„ฑ๋œ ์ƒ๋‹น์ˆ˜์˜ ์˜คํ”ˆ ์†Œ์Šค ๋ผ์ด์„ ์Šค ์ฝ”๋“œ ์Šค๋‹ˆํŽซ์„ ํฌํ•จํ•˜๋Š” BigQuery ๋ฐ์ดํ„ฐ ์„ธํŠธ [160]์„ ๊ณต๊ฐœ์ ์œผ๋กœ ๊ณต๊ฐœํ•˜์—ฌ ๋Œ€ํ‘œ์ ์ธ ์ฝ”๋“œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋กœ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. CodeGen์€ ๋ฉ€ํ‹ฐ ์–ธ์–ด ๋ฒ„์ „ CodeGen(CodeGen-Multi)์˜ ํ•™์Šต์„ ์œ„ํ•ด BigQuery ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ํ•˜์œ„ ์ง‘ํ•ฉ์ธ BIGQUERY [86]๋ฅผ ํ™œ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.
**GPT-4 Judgement:** The calculation for the cost of the subs, chips, and cookies is correct, as well as the sum of those items. However, the calculation of the delivery fee is incorrect; the solution incorrectly uses a 5% rate (.05) instead of the correct 20% rate (.20). The subsequent steps that build on this mistake are also incorrect, including the final total cost which should be $29.00 instead of $24.00. The correct steps are the individual costs of subs, chips, cookies, and their total before the delivery fee is added, while the steps involving the delivery fee and the final total are incorrect.
**GPT-4 ํŒ๋‹จ:** ์„œ๋ธŒ, ์นฉ, ์ฟ ํ‚ค์˜ ๋น„์šฉ ๊ณ„์‚ฐ๊ณผ ์ด ํ•ญ๋ชฉ๋“ค์˜ ํ•ฉ์€ ์ •ํ™•ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฐฐ๋‹ฌ๋ฃŒ ๊ณ„์‚ฐ์ด ์ž˜๋ชป๋˜์—ˆ์Šต๋‹ˆ๋‹ค; ํ•ด๊ฒฐ์ฑ…์—์„œ๋Š” ์˜ฌ๋ฐ”๋ฅธ 20% ๋น„์œจ(0.20) ๋Œ€์‹  ์ž˜๋ชป๋œ 5% ๋น„์œจ(0.05)์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ์‹ค์ˆ˜๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ํ›„์† ๋‹จ๊ณ„๋“ค๋„ ์ž˜๋ชป๋˜์—ˆ์œผ๋ฉฐ, ์ตœ์ข… ์ด ๋น„์šฉ์€ $24.00์ด ์•„๋‹Œ $29.00์ด์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์˜ฌ๋ฐ”๋ฅธ ๋‹จ๊ณ„๋Š” ๋ฐฐ๋‹ฌ๋ฃŒ๊ฐ€ ์ถ”๊ฐ€๋˜๊ธฐ ์ „์˜ ์„œ๋ธŒ, ์นฉ, ์ฟ ํ‚ค์˜ ๊ฐœ๋ณ„ ๋น„์šฉ๊ณผ ์ดํ•ฉ์ด๋ฉฐ, ๋ฐฐ๋‹ฌ๋ฃŒ์™€ ์ตœ์ข… ์ดํ•ฉ์„ ํฌํ•จํ•œ ๋‹จ๊ณ„๋Š” ์ž˜๋ชป๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
**GPT-4 ํŒ๋‹จ:** ์„œ๋ธŒ, ์นฉ, ์ฟ ํ‚ค์˜ ๋น„์šฉ ๊ณ„์‚ฐ๊ณผ ํ•ด๋‹น ํ’ˆ๋ชฉ์˜ ํ•ฉ๊ณ„๋Š” ์ •ํ™•ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฐฐ๋‹ฌ๋ฃŒ ๊ณ„์‚ฐ์ด ์ž˜๋ชป๋˜์—ˆ์œผ๋ฉฐ, ์†”๋ฃจ์…˜์—์„œ ์˜ฌ๋ฐ”๋ฅธ 20% ์š”์œจ(.20) ๋Œ€์‹  5% ์š”์œจ(.05)์„ ์ž˜๋ชป ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ์‹ค์ˆ˜๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋Š” ํ›„์† ๋‹จ๊ณ„๋„ ์ž˜๋ชป๋˜์—ˆ์œผ๋ฉฐ, ์ตœ์ข… ์ด ๋น„์šฉ์€ 24.00๋‹ฌ๋Ÿฌ๊ฐ€ ์•„๋‹Œ 29.00๋‹ฌ๋Ÿฌ๊ฐ€ ๋˜์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋ฐฐ๋‹ฌ๋ฃŒ๋ฅผ ๋”ํ•˜๊ธฐ ์ „์— ์„œ๋ธŒ, ์นฉ, ์ฟ ํ‚ค์˜ ๊ฐœ๋ณ„ ๋น„์šฉ๊ณผ ์ด์•ก์€ ์˜ฌ๋ฐ”๋ฅธ ๋‹จ๊ณ„์ด์ง€๋งŒ ๋ฐฐ๋‹ฌ๋ฃŒ์™€ ์ตœ์ข… ์ด์•ก๊ณผ ๊ด€๋ จ๋œ ๋‹จ๊ณ„๋Š” ์˜ฌ๋ฐ”๋ฅด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
* Language models typically need at least 50K training steps _regardless of_ batch size. Thus, for small \(N\), we _reduce the batch size_ to ensure the total number of training steps exceeds this threshold. For very large models, a larger batch size is preferred to enable GPU parallelism. * When \(lr\) remains constant, \(wd\) should be relatively reduced as the number of training steps increases. Mathematically, the model weights should be "halved" for every \(\Theta(\frac{1}{lr\times wd})\) training steps. Therefore, it's advisable to reduce the \(wd\) parameter when training for longer periods.
* ์–ธ์–ด ๋ชจ๋ธ์€ ์ผ๋ฐ˜์ ์œผ๋กœ ๋ฐฐ์น˜ ํฌ๊ธฐ์™€ ๊ด€๊ณ„์—†์ด ์ตœ์†Œ 50K์˜ ํ•™์Šต ๋‹จ๊ณ„๋ฅผ ํ•„์š”๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ž‘์€ \(N\)์˜ ๊ฒฝ์šฐ, ์ด ํ•™์Šต ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ด ์ž„๊ณ„๊ฐ’์„ ์ดˆ๊ณผํ•˜๋„๋ก ๋ฐฐ์น˜ ํฌ๊ธฐ๋ฅผ ์ค„์ž…๋‹ˆ๋‹ค. ๋งค์šฐ ํฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, GPU ๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ๋ฅผ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜๊ธฐ ์œ„ํ•ด ๋” ํฐ ๋ฐฐ์น˜ ํฌ๊ธฐ๊ฐ€ ์„ ํ˜ธ๋ฉ๋‹ˆ๋‹ค. * \(lr\)์ด ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€๋  ๋•Œ, ํ•™์Šต ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ \(wd\)๋ฅผ ์ƒ๋Œ€์ ์œผ๋กœ ์ค„์—ฌ์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜ํ•™์ ์œผ๋กœ, ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๋Š” \(\Theta(\frac{1}{lr\times wd})\) ํ•™์Šต ๋‹จ๊ณ„๋งˆ๋‹ค "์ ˆ๋ฐ˜"์ด ๋˜์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋” ๊ธด ๊ธฐ๊ฐ„ ๋™์•ˆ ํ•™์Šตํ•  ๋•Œ๋Š” \(wd\) ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์ค„์ด๋Š” ๊ฒƒ์ด ์ข‹์Šต๋‹ˆ๋‹ค.
* ์–ธ์–ด ๋ชจ๋ธ์€ ์ผ๋ฐ˜์ ์œผ๋กœ ๋ฐฐ์น˜ ํฌ๊ธฐ์— ๊ด€๊ณ„์—†์ด ์ตœ์†Œ 50K ํŠธ๋ ˆ์ด๋‹ ๋‹จ๊ณ„๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ \(N\)์ด ์ž‘์€ ๊ฒฝ์šฐ ์ด ํŠธ๋ ˆ์ด๋‹ ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ด ์ž„๊ณ„๊ฐ’์„ ์ดˆ๊ณผํ•˜๋„๋ก ๋ฐฐ์น˜ ํฌ๊ธฐ๋ฅผ ์ค„์ž…๋‹ˆ๋‹ค. ๋ชจ๋ธ์ด ๋งค์šฐ ํฐ ๊ฒฝ์šฐ์—๋Š” GPU ๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ๋ฅผ ์œ„ํ•ด ๋” ํฐ ๋ฐฐ์น˜ ํฌ๊ธฐ๋ฅผ ์„ ํ˜ธํ•ฉ๋‹ˆ๋‹ค. * lr$์ด ์ผ์ •ํ•˜๊ฒŒ ์œ ์ง€๋˜๋Š” ๊ฒฝ์šฐ, ํ›ˆ๋ จ ๋‹จ๊ณ„ ์ˆ˜๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ $wd$๋Š” ์ƒ๋Œ€์ ์œผ๋กœ ๊ฐ์†Œํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜ํ•™์ ์œผ๋กœ ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๋Š” \(\Theta(\frac{1}{lr\times wd})\) ํ›ˆ๋ จ ๋‹จ๊ณ„๋งˆ๋‹ค "๋ฐ˜๊ฐ"๋˜์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์žฅ์‹œ๊ฐ„ ํ›ˆ๋ จํ•  ๋•Œ๋Š” \(wd\) ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์ค„์ด๋Š” ๊ฒƒ์ด ์ข‹์Šต๋‹ˆ๋‹ค.
For each two-hop prompt \(\tau_{\text{2H}}\) in TwoHopFact, we randomly select one \(e_{1}^{\prime}\) from the same fact composition type and one \(r_{1}^{\prime}\) from a set of predefined candidate relations (provided in Appendix Table 5) to create \(\tau_{\text{2H}}^{\prime}\). We then measure the relative frequency of cases where replacing \(\tau_{\text{2H}}^{\prime}\) with \(\tau_{\text{2H}}\) via entity or relation substitution increases the recall of \(e_{2}\). A relative frequency above 0.5 suggests the LLM's chance to perform first-hop reasoning exceeds the random chance for these prompts.
TwoHopFact์˜ ๊ฐ ๋‘ ํ™‰ ํ”„๋กฌํ”„ํŠธ \(\tau_{\text{2H}}\)์— ๋Œ€ํ•ด, ๋™์ผํ•œ ์‚ฌ์‹ค ๊ตฌ์„ฑ ์œ ํ˜•์—์„œ ํ•˜๋‚˜์˜ \(e_{1}^{\prime}\)์„ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒํ•˜๊ณ , ๋ฏธ๋ฆฌ ์ •์˜๋œ ํ›„๋ณด ๊ด€๊ณ„ ์ง‘ํ•ฉ(๋ถ€๋ก ํ‘œ 5์— ์ œ๊ณต๋จ)์—์„œ ํ•˜๋‚˜์˜ \(r_{1}^{\prime}\)์„ ์„ ํƒํ•˜์—ฌ \(\tau_{\text{2H}}^{\prime}\)์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ ์—”ํ‹ฐํ‹ฐ ๋˜๋Š” ๊ด€๊ณ„ ๋Œ€์ฒด๋ฅผ ํ†ตํ•ด \(\tau_{\text{2H}}\)๋ฅผ \(\tau_{\text{2H}}^{\prime}\)๋กœ ๋Œ€์ฒดํ•  ๋•Œ \(e_{2}\)์˜ ๋ฆฌ์ฝœ์ด ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒฝ์šฐ์˜ ์ƒ๋Œ€ ๋นˆ๋„๋ฅผ ์ธก์ •ํ•ฉ๋‹ˆ๋‹ค. ์ƒ๋Œ€ ๋นˆ๋„๊ฐ€ 0.5๋ฅผ ์ดˆ๊ณผํ•˜๋ฉด ์ด๋Ÿฌํ•œ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด LLM์ด ์ฒซ ๋ฒˆ์งธ ํ™‰ ์ถ”๋ก ์„ ์ˆ˜ํ–‰ํ•  ํ™•๋ฅ ์ด ๋ฌด์ž‘์œ„ ํ™•๋ฅ ์„ ์ดˆ๊ณผํ•จ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
TwoHopFact์˜ ๊ฐ 2ํ™‰ ํ”„๋กฌํ”„ํŠธ \(\tau_{\text{2H}}\)์— ๋Œ€ํ•ด, ๋™์ผํ•œ ํŒฉํŠธ ๊ตฌ์„ฑ ์œ ํ˜•์—์„œ ํ•˜๋‚˜์˜ \(e_{1}^{\prime}\)๋ฅผ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒํ•˜๊ณ  ๋ฏธ๋ฆฌ ์ •์˜๋œ ํ›„๋ณด ๊ด€๊ณ„ ์ง‘ํ•ฉ(๋ถ€๋ก ํ‘œ 5์— ์ œ๊ณต๋จ)์—์„œ ํ•˜๋‚˜์˜ \(r_{1}^{\prime}\)๋ฅผ ์„ ํƒํ•˜์—ฌ \(\tau_{\text{2H}}^{\prime}\)๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ ์—”ํ‹ฐํ‹ฐ ๋˜๋Š” ๊ด€๊ณ„ ์น˜ํ™˜์„ ํ†ตํ•ด \(\tau_{\text{2H}}^{\prime}\)๋ฅผ \(\tau_{\text{2H}}\)๋กœ ๋Œ€์ฒดํ•  ๋•Œ \(e_{2}\)์˜ ๋ฆฌ์ฝœ์ด ์ฆ๊ฐ€ํ•˜๋Š” ๊ฒฝ์šฐ์˜ ์ƒ๋Œ€์  ๋นˆ๋„๋ฅผ ์ธก์ •ํ•ฉ๋‹ˆ๋‹ค. ์ƒ๋Œ€ ๋นˆ๋„๊ฐ€ 0.5 ์ด์ƒ์ด๋ฉด LLM์ด ์ฒซ ๋ฒˆ์งธ ํ™‰ ์ถ”๋ก ์„ ์ˆ˜ํ–‰ํ•  ํ™•๋ฅ ์ด ์ด๋Ÿฌํ•œ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•œ ๋ฌด์ž‘์œ„ ํ™•๋ฅ ์„ ์ดˆ๊ณผํ•œ๋‹ค๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.
We also do not investigate the negative consequences of deduplication. Some language tasks explicitly require memorization, like document retrieval or closed-book question answering. Also, text that gives attribution is often duplicated across documents, so removing duplicate substrings could correspond to removing _just_ the attribution, which could result in models that learn the content without its attached attribution. Deduplication is also not sufficient to remove privacy-sensitive data like bank passwords and medical records which should never be used in training data (Brown et al., 2022).
์šฐ๋ฆฌ๋Š” ๋˜ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์˜ ๋ถ€์ •์ ์ธ ๊ฒฐ๊ณผ์— ๋Œ€ํ•ด์„œ๋Š” ์กฐ์‚ฌํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ผ๋ถ€ ์–ธ์–ด ์ž‘์—…์€ ๋ฌธ์„œ ๊ฒ€์ƒ‰์ด๋‚˜ ํ์‡„ํ˜• ์งˆ๋ฌธ ์‘๋‹ต๊ณผ ๊ฐ™์ด ๋ช…์‹œ์ ์œผ๋กœ ์•”๊ธฐ๋ฅผ ์š”๊ตฌํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ถœ์ฒ˜๋ฅผ ๋ช…์‹œํ•˜๋Š” ํ…์ŠคํŠธ๋Š” ์ข…์ข… ๋ฌธ์„œ ๊ฐ„์— ์ค‘๋ณต๋˜๋ฏ€๋กœ, ์ค‘๋ณต๋œ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด์„ ์ œ๊ฑฐํ•˜๋Š” ๊ฒƒ์€ ์ถœ์ฒ˜๋งŒ ์ œ๊ฑฐํ•˜๋Š” ๊ฒƒ๊ณผ ๊ฐ™์„ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” ์ถœ์ฒ˜ ์—†์ด ๋‚ด์šฉ์„ ํ•™์Šตํ•˜๋Š” ๋ชจ๋ธ์„ ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ค‘๋ณต ์ œ๊ฑฐ๋Š” ๋˜ํ•œ ์€ํ–‰ ๋น„๋ฐ€๋ฒˆํ˜ธ๋‚˜ ์˜๋ฃŒ ๊ธฐ๋ก๊ณผ ๊ฐ™์€ ๋ฏผ๊ฐํ•œ ๊ฐœ์ธ์ •๋ณด๋ฅผ ์ œ๊ฑฐํ•˜๋Š” ๋ฐ ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์œผ๋ฉฐ, ์ด๋Ÿฌํ•œ ๋ฐ์ดํ„ฐ๋Š” ์ ˆ๋Œ€๋กœ ํ•™์Šต ๋ฐ์ดํ„ฐ์— ์‚ฌ์šฉ๋˜์–ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค (Brown et al., 2022).
๋˜ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์˜ ๋ถ€์ •์ ์ธ ๊ฒฐ๊ณผ์— ๋Œ€ํ•ด์„œ๋„ ์กฐ์‚ฌํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋ฌธ์„œ ๊ฒ€์ƒ‰์ด๋‚˜ ํ์‡„ํ˜• ๋ฌธ์ œ ๋‹ต๋ณ€๊ณผ ๊ฐ™์ด ์ผ๋ถ€ ์–ธ์–ด ์ž‘์—…์—๋Š” ๋ช…์‹œ์ ์œผ๋กœ ์•”๊ธฐ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ถœ์ฒ˜๋ฅผ ์ œ๊ณตํ•˜๋Š” ํ…์ŠคํŠธ๋Š” ๋ฌธ์„œ ๊ฐ„์— ์ค‘๋ณต๋˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์œผ๋ฏ€๋กœ ์ค‘๋ณต๋œ ํ•˜์œ„ ๋ฌธ์ž์—ด์„ ์ œ๊ฑฐํ•˜๋ฉด ์ถœ์ฒ˜๋งŒ ์ œ๊ฑฐํ•˜๋Š” ๊ฒƒ๊ณผ๋„ ๊ฐ™์„ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋กœ ์ธํ•ด ์ถœ์ฒ˜๊ฐ€ ์ฒจ๋ถ€๋˜์ง€ ์•Š์€ ์ฝ˜ํ…์ธ ๋งŒ ํ•™์Šตํ•˜๋Š” ๋ชจ๋ธ์ด ๋งŒ๋“ค์–ด์งˆ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ํ•™์Šต ๋ฐ์ดํ„ฐ์— ์‚ฌ์šฉํ•ด์„œ๋Š” ์•ˆ ๋˜๋Š” ์€ํ–‰ ๋น„๋ฐ€๋ฒˆํ˜ธ๋‚˜ ์˜๋ฃŒ ๊ธฐ๋ก๊ณผ ๊ฐ™์€ ๊ฐœ์ธ์ •๋ณด ๋ณดํ˜ธ์— ๋ฏผ๊ฐํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ œ๊ฑฐํ•˜๊ธฐ์—๋Š” ์ค‘๋ณต ์ œ๊ฑฐ๋งŒ์œผ๋กœ๋Š” ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค(Brown et al., 2022).
Current Events Task To further isolate LLMs' abilities to learn new knowledge, we created a task comprising multiple-choice questions about current events. This task includes multiple-choice questions about events that occurred after the cutoff of the various models' training data. Specifically, we focused on "current events" from the USA, in the time span of August-November 2023, that are included in the relevant Wikipedia indexes1. This method enables us to mostly guarantee that the models have not been exposed to these facts, thus allowing us to directly test knowledge injection capabilities.
ํ˜„์žฌ ์‚ฌ๊ฑด ๊ณผ์ œ: LLM์˜ ์ƒˆ๋กœ์šด ์ง€์‹์„ ํ•™์Šตํ•˜๋Š” ๋Šฅ๋ ฅ์„ ๋”์šฑ ๋ถ„๋ฆฌํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ํ˜„์žฌ ์‚ฌ๊ฑด์— ๋Œ€ํ•œ ๊ฐ๊ด€์‹ ์งˆ๋ฌธ์œผ๋กœ ๊ตฌ์„ฑ๋œ ๊ณผ์ œ๋ฅผ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ณผ์ œ๋Š” ๋‹ค์–‘ํ•œ ๋ชจ๋ธ์˜ ํ•™์Šต ๋ฐ์ดํ„ฐ๊ฐ€ ์ข…๋ฃŒ๋œ ์ดํ›„์— ๋ฐœ์ƒํ•œ ์‚ฌ๊ฑด์— ๋Œ€ํ•œ ๊ฐ๊ด€์‹ ์งˆ๋ฌธ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ์šฐ๋ฆฌ๋Š” 2023๋…„ 8์›”๋ถ€ํ„ฐ 11์›”๊นŒ์ง€์˜ ๊ธฐ๊ฐ„ ๋™์•ˆ ๋ฏธ๊ตญ์—์„œ ๋ฐœ์ƒํ•œ "ํ˜„์žฌ ์‚ฌ๊ฑด"์— ์ดˆ์ ์„ ๋งž์ถ”์—ˆ์œผ๋ฉฐ, ์ด๋Š” ๊ด€๋ จ๋œ Wikipedia ์ƒ‰์ธ์— ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ ๋ชจ๋ธ์ด ์ด๋Ÿฌํ•œ ์‚ฌ์‹ค์— ๋…ธ์ถœ๋˜์ง€ ์•Š์•˜์Œ์„ ๋Œ€๋ถ€๋ถ„ ๋ณด์žฅํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•˜์—ฌ, ์ง€์‹ ์ฃผ์ž… ๋Šฅ๋ ฅ์„ ์ง์ ‘์ ์œผ๋กœ ํ…Œ์ŠคํŠธํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
์‹œ์‚ฌ ์ด๋ฒคํŠธ ์ž‘์—… LLM์ด ์ƒˆ๋กœ์šด ์ง€์‹์„ ํ•™์Šตํ•˜๋Š” ๋Šฅ๋ ฅ์„ ๋”์šฑ ๊ฒฉ๋ฆฌํ•˜๊ธฐ ์œ„ํ•ด ์‹œ์‚ฌ ์ด๋ฒคํŠธ์— ๊ด€ํ•œ ๊ฐ๊ด€์‹ ๋ฌธ์ œ๋กœ ๊ตฌ์„ฑ๋œ ๊ณผ์ œ๋ฅผ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ณผ์ œ์—๋Š” ๋‹ค์–‘ํ•œ ๋ชจ๋ธ์˜ ํ•™์Šต ๋ฐ์ดํ„ฐ ๋งˆ๊ฐ์ผ ์ดํ›„์— ๋ฐœ์ƒํ•œ ์‚ฌ๊ฑด์— ๋Œ€ํ•œ ๊ฐ๊ด€์‹ ๋ฌธ์ œ๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์šฐ๋ฆฌ๋Š” 2023๋…„ 8~11์›” ๋ฏธ๊ตญ์—์„œ ๋ฐœ์ƒํ•œ '์‹œ์‚ฌ ์ด๋ฒคํŠธ' ์ค‘ ๊ด€๋ จ ์œ„ํ‚ค๋ฐฑ๊ณผ ์ƒ‰์ธ1์— ํฌํ•จ๋œ ์‚ฌ๊ฑด์— ์ดˆ์ ์„ ๋งž์ถ”์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ•˜๋ฉด ๋ชจ๋ธ์ด ์ด๋Ÿฌํ•œ ์‚ฌ์‹ค์— ๋…ธ์ถœ๋˜์ง€ ์•Š์•˜์Œ์„ ๋Œ€๋ถ€๋ถ„ ๋ณด์žฅํ•  ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ ์ง€์‹ ์ฃผ์ž… ๊ธฐ๋Šฅ์„ ์ง์ ‘ ํ…Œ์ŠคํŠธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
The main limitation of this article is the relatively small amount of saturated language models we studied. As it is the only suite of language models trained in the range of interest to release an extensive amount of intermediate checkpoints, we could only observe the training dynamics of small Pythia models. Although we observe strong last-layer anisotropy for the smallest GPT-2 model, we cannot tell with certainty whether it suffered from saturation. The OPT-125m model does not display a strong last-layer anisotropy, which could indicate that it was not affected by the saturation phenomenon.
์ด ๋…ผ๋ฌธ์˜ ์ฃผ์š” ํ•œ๊ณ„๋Š” ์šฐ๋ฆฌ๊ฐ€ ์—ฐ๊ตฌํ•œ ํฌํ™”๋œ ์–ธ์–ด ๋ชจ๋ธ์˜ ์ˆ˜๊ฐ€ ์ƒ๋Œ€์ ์œผ๋กœ ์ ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ๊ด€์‹ฌ ๋ฒ”์œ„ ๋‚ด์—์„œ ๊ด‘๋ฒ”์œ„ํ•œ ์ค‘๊ฐ„ ์ฒดํฌํฌ์ธํŠธ๋ฅผ ๊ณต๊ฐœํ•œ ์œ ์ผํ•œ ์–ธ์–ด ๋ชจ๋ธ ๋ชจ์Œ์ด๊ธฐ ๋•Œ๋ฌธ์—, ์šฐ๋ฆฌ๋Š” ์ž‘์€ Pythia ๋ชจ๋ธ์˜ ํ•™์Šต ๋™์—ญํ•™๋งŒ์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ€์žฅ ์ž‘์€ GPT-2 ๋ชจ๋ธ์—์„œ ๊ฐ•ํ•œ ๋งˆ์ง€๋ง‰ ์ธต์˜ ๋น„๋“ฑ๋ฐฉ์„ฑ์„ ๊ด€์ฐฐํ–ˆ์ง€๋งŒ, ๊ทธ๊ฒƒ์ด ํฌํ™” ํ˜„์ƒ์„ ๊ฒช์—ˆ๋Š”์ง€ ํ™•์‹คํžˆ ๋งํ•  ์ˆ˜๋Š” ์—†์Šต๋‹ˆ๋‹ค. OPT-125m ๋ชจ๋ธ์€ ๊ฐ•ํ•œ ๋งˆ์ง€๋ง‰ ์ธต์˜ ๋น„๋“ฑ๋ฐฉ์„ฑ์„ ๋‚˜ํƒ€๋‚ด์ง€ ์•Š์œผ๋ฉฐ, ์ด๋Š” ํฌํ™” ํ˜„์ƒ์˜ ์˜ํ–ฅ์„ ๋ฐ›์ง€ ์•Š์•˜์Œ์„ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ด ๊ธ€์˜ ์ฃผ์š” ํ•œ๊ณ„๋Š” ์šฐ๋ฆฌ๊ฐ€ ์—ฐ๊ตฌํ•œ ํฌํ™” ์–ธ์–ด ๋ชจ๋ธ์˜ ์ƒ๋Œ€์ ์œผ๋กœ ์ ์€ ์–‘์ž…๋‹ˆ๋‹ค. ๊ด‘๋ฒ”์œ„ํ•œ ์ค‘๊ฐ„ ์ฒดํฌํฌ์ธํŠธ๋ฅผ ๊ณต๊ฐœํ•˜๊ธฐ ์œ„ํ•ด ๊ด€์‹ฌ ์žˆ๋Š” ๋ฒ”์œ„์—์„œ ํ›ˆ๋ จ๋œ ์œ ์ผํ•œ ์–ธ์–ด ๋ชจ๋ธ ์ œํ’ˆ๊ตฐ์ด๊ธฐ ๋•Œ๋ฌธ์— ์ž‘์€ Pythia ๋ชจ๋ธ์˜ ํ›ˆ๋ จ ๋™์—ญํ•™๋งŒ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ€์žฅ ์ž‘์€ GPT-2 ๋ชจ๋ธ์— ๋Œ€ํ•ด ๊ฐ•ํ•œ ๋งˆ์ง€๋ง‰ ๋ ˆ์ด์–ด ์ด๋ฐฉ์„ฑ์„ ๊ด€์ฐฐํ–ˆ์ง€๋งŒ, ํฌํ™” ํ˜„์ƒ์œผ๋กœ ์ธํ•ด ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ–ˆ๋Š”์ง€ ํ™•์‹คํ•˜๊ฒŒ ์•Œ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. OPT-125m ๋ชจ๋ธ์€ ๊ฐ•ํ•œ ๋งˆ์ง€๋ง‰ ๋ ˆ์ด์–ด ์ด๋ฐฉ์„ฑ์„ ๋‚˜ํƒ€๋‚ด์ง€ ์•Š์œผ๋ฉฐ, ์ด๋Š” ํฌํ™” ํ˜„์ƒ์˜ ์˜ํ–ฅ์„ ๋ฐ›์ง€ ์•Š์•˜์Œ์„ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Given two suffix arrays \(A_{1}\) and \(A_{2}\) for two sequences \(S_{1}\) and \(S_{2}\) it's not completely trivial to construct a single suffix array \(A\) for \(S=S_{1}\parallel S_{2}\) because of the boundary conditions. Instead, we don't build the data \(S=S_{1}\ ||\ S_{2}\) but rather let \(S_{1}^{\prime}=S_{1}\ ||\ S_{2}[uptoK]\) for some \(K\) greater than the longest substring match. Then we build the arrays on \(S_{1}^{\prime}\) and \(S_{2}\). To merge the arrays together we can remove the items from the first array after index \(|S_{1}|\) and merge-sort insert them into the second.
๋‘ ์‹œํ€€์Šค \(S_{1}\)์™€ \(S_{2}\)์— ๋Œ€ํ•œ ๋‘ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด \(A_{1}\)์™€ \(A_{2}\)๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ, ๊ฒฝ๊ณ„ ์กฐ๊ฑด ๋•Œ๋ฌธ์— \(S=S_{1}\parallel S_{2}\)์— ๋Œ€ํ•œ ๋‹จ์ผ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด \(A\)๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” ๊ฒƒ์€ ์™„์ „ํžˆ ๊ฐ„๋‹จํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹ , ์šฐ๋ฆฌ๋Š” \(S=S_{1}\ ||\ S_{2}\) ๋ฐ์ดํ„ฐ๋ฅผ ๊ตฌ์ถ•ํ•˜์ง€ ์•Š๊ณ , ๊ฐ€์žฅ ๊ธด ๋ถ€๋ถ„ ๋ฌธ์ž์—ด ์ผ์น˜๋ณด๋‹ค ํฐ \(K\)์— ๋Œ€ํ•ด \(S_{1}^{\prime}=S_{1}\ ||\ S_{2}[uptoK]\)๋กœ ์„ค์ •ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ \(S_{1}^{\prime}\)๊ณผ \(S_{2}\)์— ๋Œ€ํ•œ ๋ฐฐ์—ด์„ ๊ตฌ์ถ•ํ•ฉ๋‹ˆ๋‹ค. ๋ฐฐ์—ด์„ ํ•จ๊ป˜ ๋ณ‘ํ•ฉํ•˜๊ธฐ ์œ„ํ•ด ์ฒซ ๋ฒˆ์งธ ๋ฐฐ์—ด์—์„œ ์ธ๋ฑ์Šค \(|S_{1}|\) ์ดํ›„์˜ ํ•ญ๋ชฉ์„ ์ œ๊ฑฐํ•˜๊ณ  ์ด๋ฅผ ๋‘ ๋ฒˆ์งธ ๋ฐฐ์—ด์— ๋ณ‘ํ•ฉ ์ •๋ ฌ ์‚ฝ์ž…ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋‘ ์‹œํ€€์Šค \(S_{1}\) ๋ฐ \(S_{2}\)์— ๋Œ€ํ•œ ๋‘ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด \(A_{1}\) ๋ฐ \(A_{2}\)๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ, ๊ฒฝ๊ณ„ ์กฐ๊ฑด ๋•Œ๋ฌธ์— \(S=S_{1}\parallel S_{2}\)์— ๋Œ€ํ•œ ๋‹จ์ผ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด \(A\)๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” ๊ฒƒ์€ ์™„์ „ํžˆ ๊ฐ„๋‹จํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹ , \(S=S_{1}\ ||\ S_{2}\) ๋ฐ์ดํ„ฐ๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ๋Œ€์‹  ๊ฐ€์žฅ ๊ธด ๋ถ€๋ถ„ ๋ฌธ์ž์—ด ์ผ์น˜๋ณด๋‹ค ํฐ \(K\)์— ๋Œ€ํ•ด \(S_{1}^{\prime}=S_{1}\ ||\ S_{2}[uptoK]\)๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ \(S_{1}^{\prime}\)์™€ \(S_{2}\)์— ๋Œ€ํ•œ ๋ฐฐ์—ด์„ ๊ตฌ์ถ•ํ•ฉ๋‹ˆ๋‹ค. ๋ฐฐ์—ด์„ ๋ณ‘ํ•ฉํ•˜๊ธฐ ์œ„ํ•ด ์ฒซ ๋ฒˆ์งธ ๋ฐฐ์—ด์—์„œ ์ธ๋ฑ์Šค \(|S_{1}|\) ์ดํ›„์˜ ํ•ญ๋ชฉ์„ ์ œ๊ฑฐํ•˜๊ณ  ๋‘ ๋ฒˆ์งธ ๋ฐฐ์—ด์— ๋ณ‘ํ•ฉ ์ •๋ ฌ ์‚ฝ์ž…ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Contribution 2: An improved evaluation protocol for legal LLMs.Concurrently, we introduce LegalBench-Instruct, a supplemental iteration of LegalBench Guha et al. (2022, 2023)2, crafted to better gauge and refine the legal proficiency of language models, which we hope will contribute to future advancements into research in the legal domain. To further enrich the models' capabilities in legal contexts, we also include the legal tasks of the popular MMLU benchmark Hendrycks et al. (2020) in our evaluation protocol, particularly focusing on international law, professional law3 and jurisprudence.
๊ธฐ์—ฌ 2: ๋ฒ•๋ฅ  LLMs์— ๋Œ€ํ•œ ๊ฐœ์„ ๋œ ํ‰๊ฐ€ ํ”„๋กœํ† ์ฝœ. ๋™์‹œ์—, ์šฐ๋ฆฌ๋Š” LegalBench Guha et al. (2022, 2023)2์˜ ๋ณด์™„๋œ ๋ฐ˜๋ณต์ธ LegalBench-Instruct๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์–ธ์–ด ๋ชจ๋ธ์˜ ๋ฒ•๋ฅ  ์ „๋ฌธ์„ฑ์„ ๋” ์ž˜ ํ‰๊ฐ€ํ•˜๊ณ  ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด ์ œ์ž‘๋˜์—ˆ์œผ๋ฉฐ, ๋ฒ•๋ฅ  ๋ถ„์•ผ ์—ฐ๊ตฌ์˜ ๋ฏธ๋ž˜ ๋ฐœ์ „์— ๊ธฐ์—ฌํ•  ์ˆ˜ ์žˆ๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค. ๋ชจ๋ธ์˜ ๋ฒ•๋ฅ  ๋งฅ๋ฝ์—์„œ์˜ ๋Šฅ๋ ฅ์„ ๋”์šฑ ํ’๋ถ€ํ•˜๊ฒŒ ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ํ‰๊ฐ€ ํ”„๋กœํ† ์ฝœ์— ์ธ๊ธฐ ์žˆ๋Š” MMLU ๋ฒค์น˜๋งˆํฌ Hendrycks et al. (2020)์˜ ๋ฒ•๋ฅ  ๊ณผ์ œ๋„ ํฌํ•จ์‹œ์ผฐ์œผ๋ฉฐ, ํŠนํžˆ ๊ตญ์ œ๋ฒ•, ์ „๋ฌธ๋ฒ•3 ๋ฐ ๋ฒ•ํ•™์— ์ค‘์ ์„ ๋‘์—ˆ์Šต๋‹ˆ๋‹ค.
๊ธฐ์—ฌ๋„ 2: ๊ฐœ์„ ๋œ ๋ฒ•๋ฅ  LLM ํ‰๊ฐ€ ํ”„๋กœํ† ์ฝœ.๋™์‹œ์—, ์šฐ๋ฆฌ๋Š” ์–ธ์–ด ๋ชจ๋ธ์˜ ๋ฒ•๋ฅ ์  ์ˆ™๋ จ๋„๋ฅผ ๋” ์ž˜ ์ธก์ •ํ•˜๊ณ  ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด ์ œ์ž‘๋œ LegalBench-Instruct, LegalBench Guha et al. (2022, 2023)2์˜ ๋ณด์™„์ ์ธ ๋ฐ˜๋ณต์„ ์†Œ๊ฐœํ•˜๋ฉฐ, ์ด๋ฅผ ํ†ตํ•ด ํ–ฅํ›„ ๋ฒ•๋ฅ  ์˜์—ญ์˜ ์—ฐ๊ตฌ ๋ฐœ์ „์— ๊ธฐ์—ฌํ•  ์ˆ˜ ์žˆ๊ธฐ๋ฅผ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค. ๋ฒ•๋ฅ  ๋งฅ๋ฝ์—์„œ ๋ชจ๋ธ์˜ ์—ญ๋Ÿ‰์„ ๋”์šฑ ๊ฐ•ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ํ‰๊ฐ€ ํ”„๋กœํ† ์ฝœ์— ์ธ๊ธฐ ์žˆ๋Š” MMLU ๋ฒค์น˜๋งˆํฌ Hendrycks et al. (2020)์˜ ๋ฒ•๋ฅ  ๊ณผ์ œ, ํŠนํžˆ ๊ตญ์ œ๋ฒ•, ์ „๋ฌธ๋ฒ•3 ๋ฐ ๋ฒ•ํ•™์— ์ค‘์ ์„ ๋‘๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
We adopt the performance on XQuAD in Chinese as the validation set to all languages and all tasks. Specifically, Table 9 shows the result on Vicuna when deactivating language-specific neurons in the understanding layer (\(D_{\mathcal{U}}\)) and generation layer (\(D_{\mathcal{G}}\)), where \(N_{1}\) is the number of understanding layers and \(N_{2}\) is the number of generation layer. We find that when setting \(N_{1}=8\) and \(N_{2}=2\), performance in English is influenced the least while performance in Chinese decreases the most. As for Mistral, the number is \(N_{1}=6\) and \(N_{2}=3\).
์šฐ๋ฆฌ๋Š” ๋ชจ๋“  ์–ธ์–ด์™€ ๋ชจ๋“  ์ž‘์—…์— ๋Œ€ํ•ด ์ค‘๊ตญ์–ด๋กœ ๋œ XQuAD์˜ ์„ฑ๋Šฅ์„ ๊ฒ€์ฆ ์„ธํŠธ๋กœ ์ฑ„ํƒํ•ฉ๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, Table 9๋Š” ์ดํ•ด ๋ ˆ์ด์–ด (\(D_{\mathcal{U}}\))์™€ ์ƒ์„ฑ ๋ ˆ์ด์–ด (\(D_{\mathcal{G}\))์—์„œ ์–ธ์–ด ํŠนํ™” ๋‰ด๋Ÿฐ์„ ๋น„ํ™œ์„ฑํ™”ํ•  ๋•Œ Vicuna์˜ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ \(N_{1}\)์€ ์ดํ•ด ๋ ˆ์ด์–ด์˜ ์ˆ˜์ด๊ณ  \(N_{2}\)๋Š” ์ƒ์„ฑ ๋ ˆ์ด์–ด์˜ ์ˆ˜์ž…๋‹ˆ๋‹ค. \(N_{1}=8\)๊ณผ \(N_{2}=2\)๋กœ ์„ค์ •ํ–ˆ์„ ๋•Œ ์˜์–ด ์„ฑ๋Šฅ์€ ๊ฐ€์žฅ ์ ๊ฒŒ ์˜ํ–ฅ์„ ๋ฐ›๋Š” ๋ฐ˜๋ฉด, ์ค‘๊ตญ์–ด ์„ฑ๋Šฅ์€ ๊ฐ€์žฅ ๋งŽ์ด ๊ฐ์†Œํ•˜๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. Mistral์˜ ๊ฒฝ์šฐ, ๊ทธ ์ˆ˜๋Š” \(N_{1}=6\)๊ณผ \(N_{2}=3\)์ž…๋‹ˆ๋‹ค.
์šฐ๋ฆฌ๋Š” ์ค‘๊ตญ์–ด๋กœ ๋œ XQuAD์˜ ์„ฑ๋Šฅ์„ ๋ชจ๋“  ์–ธ์–ด์™€ ๋ชจ๋“  ์ž‘์—…์— ๋Œ€ํ•œ ์œ ํšจ์„ฑ ๊ฒ€์‚ฌ ์„ธํŠธ๋กœ ์ฑ„ํƒํ•œ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ํ‘œ 9๋Š” ์ดํ•ด ๊ณ„์ธต(\(D_{\mathcal{U}}\))๊ณผ ์ƒ์„ฑ ๊ณ„์ธต(\(D_{\mathcal{G}}\))์—์„œ ์–ธ์–ด๋ณ„ ๋‰ด๋Ÿฐ์„ ๋น„ํ™œ์„ฑํ™”ํ•  ๋•Œ Vicuna์˜ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ, ์—ฌ๊ธฐ์„œ \(N_{1}\)์€ ์ดํ•ด ๊ณ„์ธต์˜ ์ˆ˜์ด๊ณ  \(N_{2}\)๋Š” ์ƒ์„ฑ ๊ณ„์ธต์˜ ์ˆ˜์ด๋‹ค. N_{1}=8$ ๋ฐ \(N_{2}=2$๋กœ ์„ค์ •ํ•˜๋ฉด ์˜์–ด ์„ฑ๋Šฅ์€ ์˜ํ–ฅ์„ ๊ฐ€์žฅ ์ ๊ฒŒ ๋ฐ›๊ณ  ์ค‘๊ตญ์–ด ์„ฑ๋Šฅ์€ ๊ฐ€์žฅ ๋งŽ์ด ๊ฐ์†Œํ•œ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ๋‹ค. Mistral์˜ ๊ฒฝ์šฐ, ๊ทธ ์ˆ˜๋Š” \(N_{1}=6$ ๋ฐ \(N_{2}=3$์ด๋‹ค.
\(\bullet\)_Helpfulness._ To be helpful, the LLM should demonstrate a clear attempt to assist users in solving their tasks or answering questions in a concise and efficient manner as possible. At a higher level, when further clarification is needed, the LLM should demonstrate the capability of eliciting additional relevant information through pertinent inquiries and exhibit suitable levels of sensitivity, perceptiveness, and prudence [368]. Realizing the alignment of helpful behavior is challenging for LLMs since it is difficult to precisely define and measure the intention of users [366].
โ€ข ์œ ์šฉ์„ฑ: ์œ ์šฉํ•˜๊ธฐ ์œ„ํ•ด์„œ, LLM์€ ์‚ฌ์šฉ์ž๊ฐ€ ๊ณผ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ฑฐ๋‚˜ ์งˆ๋ฌธ์— ๋‹ตํ•˜๋Š” ๋ฐ ์žˆ์–ด ๊ฐ€๋Šฅํ•œ ํ•œ ๊ฐ„๊ฒฐํ•˜๊ณ  ํšจ์œจ์ ์ธ ๋ฐฉ์‹์œผ๋กœ ๋ช…ํ™•ํ•œ ๋„์›€์„ ์ œ๊ณตํ•˜๋ ค๋Š” ์‹œ๋„๋ฅผ ๋ณด์—ฌ์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋” ๋†’์€ ์ˆ˜์ค€์—์„œ๋Š” ์ถ”๊ฐ€์ ์ธ ์„ค๋ช…์ด ํ•„์š”ํ•  ๋•Œ, LLM์€ ๊ด€๋ จ๋œ ์งˆ๋ฌธ์„ ํ†ตํ•ด ์ถ”๊ฐ€์ ์ธ ๊ด€๋ จ ์ •๋ณด๋ฅผ ์ด๋Œ์–ด๋‚ด๋Š” ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์•ผ ํ•˜๋ฉฐ, ์ ์ ˆํ•œ ์ˆ˜์ค€์˜ ๋ฏผ๊ฐ์„ฑ, ํ†ต์ฐฐ๋ ฅ, ์‹ ์ค‘ํ•จ์„ ๋‚˜ํƒ€๋‚ด์•ผ ํ•ฉ๋‹ˆ๋‹ค [368]. ์œ ์šฉํ•œ ํ–‰๋™์˜ ์ •๋ ฌ์„ ์‹คํ˜„ํ•˜๋Š” ๊ฒƒ์€ LLM์—๊ฒŒ ๋„์ „์ ์ธ ๊ณผ์ œ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์‚ฌ์šฉ์ž์˜ ์˜๋„๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ์ •์˜ํ•˜๊ณ  ์ธก์ •ํ•˜๋Š” ๊ฒƒ์ด ์–ด๋ ต๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค [366].
\(\bullet\)_์œ ์šฉ์„ฑ._ ์œ ์šฉ์„ฑ์„ ์œ„ํ•ด LLM์€ ๊ฐ€๋Šฅํ•œ ํ•œ ๊ฐ„๊ฒฐํ•˜๊ณ  ํšจ์œจ์ ์ธ ๋ฐฉ์‹์œผ๋กœ ์‚ฌ์šฉ์ž์˜ ์ž‘์—… ํ•ด๊ฒฐ์ด๋‚˜ ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ต๋ณ€์„ ์ง€์›ํ•˜๋ ค๋Š” ๋…ธ๋ ฅ์„ ๋ช…ํ™•ํ•˜๊ฒŒ ๋ณด์—ฌ์ค˜์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋” ์ž์„ธํ•œ ์„ค๋ช…์ด ํ•„์š”ํ•œ ๊ฒฝ์šฐ, LLM์€ ์ ์ ˆํ•œ ์งˆ๋ฌธ์„ ํ†ตํ•ด ์ถ”๊ฐ€ ๊ด€๋ จ ์ •๋ณด๋ฅผ ์ด๋Œ์–ด๋‚ด๋Š” ๊ธฐ๋Šฅ์„ ๋ณด์—ฌ์ฃผ์–ด์•ผ ํ•˜๋ฉฐ, ์ ์ ˆํ•œ ์ˆ˜์ค€์˜ ๊ฐ์ˆ˜์„ฑ, ํ†ต์ฐฐ๋ ฅ, ์‹ ์ค‘ํ•จ์„ ๋ณด์—ฌ์ค˜์•ผ ํ•ฉ๋‹ˆ๋‹ค[368]. ์œ ์šฉํ•œ ํ–‰๋™์˜ ์ •๋ ฌ์„ ์‹คํ˜„ํ•˜๋Š” ๊ฒƒ์€ ์‚ฌ์šฉ์ž์˜ ์˜๋„๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ์ •์˜ํ•˜๊ณ  ์ธก์ •ํ•˜๊ธฐ ์–ด๋ ต๊ธฐ ๋•Œ๋ฌธ์— LLM์— ์–ด๋ ค์šด ๊ณผ์ œ์ž…๋‹ˆ๋‹ค[366].
However, model merging is considered by many to be a form of black art or alchemy, relying on the model maker's intuition and instincts about model selection and merging recipes to create and refine a new model that performs well for a particular task. Furthermore, the model maker is often required to have some domain knowledge for the various different benchmark tasks. Given the large diversity of open models and benchmarks in the community, human intuition can only go so far, and we believe a more systematic approach for discovering new model combinations will take things much further.
๊ทธ๋Ÿฌ๋‚˜ ๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์— ์˜ํ•ด ์ผ์ข…์˜ ํ‘๋งˆ์ˆ  ๋˜๋Š” ์—ฐ๊ธˆ์ˆ ๋กœ ๊ฐ„์ฃผ๋˜๋ฉฐ, ํŠน์ • ์ž‘์—…์— ๋Œ€ํ•ด ์ž˜ ์ˆ˜ํ–‰๋˜๋Š” ์ƒˆ๋กœ์šด ๋ชจ๋ธ์„ ๋งŒ๋“ค๊ณ  ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด ๋ชจ๋ธ ์ œ์ž‘์ž์˜ ์ง๊ด€๊ณผ ๋ณธ๋Šฅ์— ์˜์กดํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋ชจ๋ธ ์ œ์ž‘์ž๋Š” ๋‹ค์–‘ํ•œ ๋ฒค์น˜๋งˆํฌ ์ž‘์—…์— ๋Œ€ํ•œ ๋„๋ฉ”์ธ ์ง€์‹์„ ๊ฐ–์ถ”์–ด์•ผ ํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์Šต๋‹ˆ๋‹ค. ์ปค๋ฎค๋‹ˆํ‹ฐ ๋‚ด์˜ ๋‹ค์–‘ํ•œ ์˜คํ”ˆ ๋ชจ๋ธ๊ณผ ๋ฒค์น˜๋งˆํฌ๋ฅผ ๊ณ ๋ คํ•  ๋•Œ, ์ธ๊ฐ„์˜ ์ง๊ด€์—๋Š” ํ•œ๊ณ„๊ฐ€ ์žˆ์œผ๋ฉฐ, ์šฐ๋ฆฌ๋Š” ์ƒˆ๋กœ์šด ๋ชจ๋ธ ์กฐํ•ฉ์„ ๋ฐœ๊ฒฌํ•˜๊ธฐ ์œ„ํ•œ ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด ํ›จ์”ฌ ๋” ๋‚˜์•„๊ฐˆ ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค.
๊ทธ๋Ÿฌ๋‚˜ ๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ๋ชจ๋ธ ์„ ํƒ๊ณผ ๋ณ‘ํ•ฉ ๋ ˆ์‹œํ”ผ์— ๋Œ€ํ•œ ๋ชจ๋ธ ์ œ์ž‘์ž์˜ ์ง๊ด€๊ณผ ๋ณธ๋Šฅ์— ์˜์กดํ•˜์—ฌ ํŠน์ • ์ž‘์—…์— ์ ํ•ฉํ•œ ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜๋Š” ์ƒˆ๋กœ์šด ๋ชจ๋ธ์„ ๋งŒ๋“ค๊ณ  ๊ฐœ์„ ํ•˜๋Š” ๊ฒƒ์œผ๋กœ, ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์•”์ˆ ์ด๋‚˜ ์—ฐ๊ธˆ์ˆ ์˜ ํ•œ ํ˜•ํƒœ๋กœ ๊ฐ„์ฃผํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ชจ๋ธ ์ œ์ž‘์ž๋Š” ๋‹ค์–‘ํ•œ ๋ฒค์น˜๋งˆํฌ ์ž‘์—…์— ๋Œ€ํ•œ ์ผ๋ถ€ ๋„๋ฉ”์ธ ์ง€์‹์„ ๋ณด์œ ํ•ด์•ผ ํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์Šต๋‹ˆ๋‹ค. ์ปค๋ฎค๋‹ˆํ‹ฐ์—์„œ ์˜คํ”ˆ ๋ชจ๋ธ๊ณผ ๋ฒค์น˜๋งˆํฌ์˜ ๋‹ค์–‘์„ฑ์ด ์ปค์ง€๋ฉด์„œ ์ธ๊ฐ„์˜ ์ง๊ด€๋งŒ์œผ๋กœ๋Š” ํ•œ๊ณ„๊ฐ€ ์žˆ์œผ๋ฉฐ, ์ƒˆ๋กœ์šด ๋ชจ๋ธ ์กฐํ•ฉ์„ ๋ฐœ๊ฒฌํ•˜๊ธฐ ์œ„ํ•œ ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด ๋” ๋งŽ์€ ๋ฐœ์ „์„ ๊ฐ€์ ธ์˜ฌ ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค.
**Mixing different training distributions:** While we chose one data distribution to both select data and train on, modern LLM setups usually mix different data sources. Our method is likely complimentary to such pipelines: practitioners may use D4 to diversify and de-duplicate individual data sources and then mix data sources to provide additional diversity in their training dataset. We leave exploring the efficacy of D4 on a mix of training distributions as future work, but expect that this will yield further gains by reducing redundancy across datasets as well as within datasets.
๋‹ค์–‘ํ•œ ํ›ˆ๋ จ ๋ถ„ํฌ์˜ ํ˜ผํ•ฉ: ์šฐ๋ฆฌ๋Š” ๋ฐ์ดํ„ฐ ์„ ํƒ๊ณผ ํ›ˆ๋ จ์— ํ•˜๋‚˜์˜ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๋ฅผ ์„ ํƒํ–ˆ์ง€๋งŒ, ํ˜„๋Œ€์˜ LLM ์„ค์ •์€ ๋ณดํ†ต ๋‹ค์–‘ํ•œ ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ํ˜ผํ•ฉํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์€ ์ด๋Ÿฌํ•œ ํŒŒ์ดํ”„๋ผ์ธ์— ๋ณด์™„์ ์ผ ๊ฐ€๋Šฅ์„ฑ์ด ํฝ๋‹ˆ๋‹ค: ์‹ค๋ฌด์ž๋“ค์€ D4๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ๋‹ค์–‘ํ™”ํ•˜๊ณ  ์ค‘๋ณต์„ ์ œ๊ฑฐํ•œ ํ›„, ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ํ˜ผํ•ฉํ•˜์—ฌ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์…‹์— ์ถ”๊ฐ€์ ์ธ ๋‹ค์–‘์„ฑ์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋‹ค์–‘ํ•œ ํ›ˆ๋ จ ๋ถ„ํฌ์˜ ํ˜ผํ•ฉ์—์„œ D4์˜ ํšจ๋Šฅ์„ ํƒ๊ตฌํ•˜๋Š” ๊ฒƒ์„ ๋ฏธ๋ž˜์˜ ๊ณผ์ œ๋กœ ๋‚จ๊ฒจ๋‘์ง€๋งŒ, ์ด๋Š” ๋ฐ์ดํ„ฐ์…‹ ๊ฐ„ ๋ฐ ๋ฐ์ดํ„ฐ์…‹ ๋‚ด ์ค‘๋ณต์„ ์ค„์ž„์œผ๋กœ์จ ์ถ”๊ฐ€์ ์ธ ์ด๋“์„ ๊ฐ€์ ธ์˜ฌ ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.
**๋‹ค์–‘ํ•œ ํ•™์Šต ๋ถ„ํฌ ํ˜ผํ•ฉ:** ๋ฐ์ดํ„ฐ๋ฅผ ์„ ํƒํ•˜๊ณ  ํ•™์Šตํ•˜๋Š” ๋ฐ ํ•˜๋‚˜์˜ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๋ฅผ ์„ ํƒํ–ˆ์ง€๋งŒ, ์ตœ์‹  LLM ์„ค์ •์—์„œ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์„œ๋กœ ๋‹ค๋ฅธ ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ํ˜ผํ•ฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํŒŒ์ดํ”„๋ผ์ธ์— D4๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ๋‹ค์–‘ํ™”ํ•˜๊ณ  ์ค‘๋ณต์„ ์ œ๊ฑฐํ•œ ๋‹ค์Œ ๋ฐ์ดํ„ฐ ์†Œ์Šค๋ฅผ ํ˜ผํ•ฉํ•˜์—ฌ ํ•™์Šต ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ์ถ”๊ฐ€ ๋‹ค์–‘์„ฑ์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ํ•™์Šต ๋ถ„ํฌ์— ๋Œ€ํ•œ D4์˜ ํšจ์œจ์„ฑ์„ ํƒ์ƒ‰ํ•˜๋Š” ๊ฒƒ์€ ํ–ฅํ›„ ์ž‘์—…์œผ๋กœ ๋‚จ๊ฒจ๋‘์ง€๋งŒ, ๋ฐ์ดํ„ฐ ์„ธํŠธ ๊ฐ„์˜ ์ค‘๋ณต์„ฑ์€ ๋ฌผ๋ก  ๋ฐ์ดํ„ฐ ์„ธํŠธ ๋‚ด์˜ ์ค‘๋ณต์„ฑ๋„ ์ค„์—ฌ์„œ ๋” ๋งŽ์€ ์ด๋“์„ ์–ป์„ ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.
**GPT-4 Judgement:** The first step mistakenly calculates the distance traveled after the first stop until the end of the trip, not accounting for the second stop which is 15 miles before the end. The second step incorrectly describes the result of the first calculation as the distance between the second stop and the end of the trip, whereas it is actually the distance from the first stop to the end of the trip minus the distance from the second stop to the end of the trip. However, despite the incorrect reasoning in both steps, the final answer achieved matches the correct answer.
**GPT-4 ํŒ๋‹จ:** ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„์—์„œ๋Š” ๋‘ ๋ฒˆ์งธ ์ •์ฐจ ์ง€์ ์ด ๋์—์„œ 15๋งˆ์ผ ๋–จ์–ด์ ธ ์žˆ๋‹ค๋Š” ์ ์„ ๊ณ ๋ คํ•˜์ง€ ์•Š๊ณ , ์ฒซ ๋ฒˆ์งธ ์ •์ฐจ ์ดํ›„๋ถ€ํ„ฐ ์—ฌํ–‰์˜ ๋๊นŒ์ง€ ์ด๋™ํ•œ ๊ฑฐ๋ฆฌ๋ฅผ ์ž˜๋ชป ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์—์„œ๋Š” ์ฒซ ๋ฒˆ์งธ ๊ณ„์‚ฐ์˜ ๊ฒฐ๊ณผ๋ฅผ ๋‘ ๋ฒˆ์งธ ์ •์ฐจ ์ง€์ ๊ณผ ์—ฌํ–‰์˜ ๋ ์‚ฌ์ด์˜ ๊ฑฐ๋ฆฌ๋กœ ์ž˜๋ชป ์„ค๋ช…ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ๋Š” ์ฒซ ๋ฒˆ์งธ ์ •์ฐจ ์ง€์ ์—์„œ ์—ฌํ–‰์˜ ๋๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ์—์„œ ๋‘ ๋ฒˆ์งธ ์ •์ฐจ ์ง€์ ์—์„œ ์—ฌํ–‰์˜ ๋๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ๋ฅผ ๋บ€ ๊ฐ’์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋‘ ๋‹จ๊ณ„ ๋ชจ๋‘์—์„œ ์ž˜๋ชป๋œ ์ถ”๋ก ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , ์ตœ์ข… ๋‹ต๋ณ€์€ ์˜ฌ๋ฐ”๋ฅธ ๋‹ต๋ณ€๊ณผ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค.
**GPT-4 ํŒ๋‹จ:** ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„๋Š” ์ฒซ ๋ฒˆ์งธ ์ •๋ฅ˜์žฅ ์ดํ›„ ์—ฌํ–‰์ด ๋๋‚  ๋•Œ๊นŒ์ง€ ์ด๋™ํ•œ ๊ฑฐ๋ฆฌ๋ฅผ ์ž˜๋ชป ๊ณ„์‚ฐํ•˜์—ฌ ๋์—์„œ 15๋งˆ์ผ ์ „์— ์žˆ๋Š” ๋‘ ๋ฒˆ์งธ ์ •๋ฅ˜์žฅ์„ ๊ณ ๋ คํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์—์„œ๋Š” ์ฒซ ๋ฒˆ์งธ ๊ณ„์‚ฐ์˜ ๊ฒฐ๊ณผ๋ฅผ ๋‘ ๋ฒˆ์งธ ์ •๋ฅ˜์žฅ๊ณผ ์—ฌํ–‰ ์ข…๋ฃŒ ์ง€์  ์‚ฌ์ด์˜ ๊ฑฐ๋ฆฌ๋กœ ์ž˜๋ชป ์„ค๋ช…ํ•˜๋Š”๋ฐ, ์‹ค์ œ๋กœ๋Š” ์ฒซ ๋ฒˆ์งธ ์ •๋ฅ˜์žฅ์—์„œ ์—ฌํ–‰ ์ข…๋ฃŒ ์ง€์ ๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ์—์„œ ๋‘ ๋ฒˆ์งธ ์ •๋ฅ˜์žฅ์—์„œ ์—ฌํ–‰ ์ข…๋ฃŒ ์ง€์ ๊นŒ์ง€์˜ ๊ฑฐ๋ฆฌ๋ฅผ ๋บ€ ๊ฐ’์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋‘ ๋‹จ๊ณ„ ๋ชจ๋‘์—์„œ ์ž˜๋ชป๋œ ์ถ”๋ก ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์ตœ์ข… ๋‹ต์€ ์ •๋‹ต๊ณผ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค.
**Application Layer.** At the application layer, agent applications, such as travel agent or math agent, are developed and deployed. In this layer, AIOS provides the AIOS SDK, with a higher abstraction of system calls that simplifies the development process for agent developers. This SDK allows for development of agent applications by offering a rich toolkit that abstract away the complexities of lower-level system functions. This enables developers to dedicate their focus to the essential logic and functionalities of their agents, facilitating a more efficient development process.
**์‘์šฉ ๊ณ„์ธต.** ์‘์šฉ ๊ณ„์ธต์—์„œ๋Š” ์—ฌํ–‰ ์—์ด์ „ํŠธ๋‚˜ ์ˆ˜ํ•™ ์—์ด์ „ํŠธ์™€ ๊ฐ™์€ ์—์ด์ „ํŠธ ์‘์šฉ ํ”„๋กœ๊ทธ๋žจ์ด ๊ฐœ๋ฐœ๋˜๊ณ  ๋ฐฐํฌ๋ฉ๋‹ˆ๋‹ค. ์ด ๊ณ„์ธต์—์„œ AIOS๋Š” ์—์ด์ „ํŠธ ๊ฐœ๋ฐœ์ž์˜ ๊ฐœ๋ฐœ ๊ณผ์ •์„ ๋‹จ์ˆœํ™”ํ•˜๋Š” ์‹œ์Šคํ…œ ํ˜ธ์ถœ์˜ ๋†’์€ ์ถ”์ƒํ™”๋ฅผ ์ œ๊ณตํ•˜๋Š” AIOS SDK๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด SDK๋Š” ํ•˜์œ„ ์ˆ˜์ค€ ์‹œ์Šคํ…œ ๊ธฐ๋Šฅ์˜ ๋ณต์žก์„ฑ์„ ์ถ”์ƒํ™”ํ•˜๋Š” ํ’๋ถ€ํ•œ ๋„๊ตฌ ํ‚คํŠธ๋ฅผ ์ œ๊ณตํ•˜์—ฌ ์—์ด์ „ํŠธ ์‘์šฉ ํ”„๋กœ๊ทธ๋žจ ๊ฐœ๋ฐœ์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๊ฐœ๋ฐœ์ž๋Š” ์—์ด์ „ํŠธ์˜ ํ•„์ˆ˜ ๋…ผ๋ฆฌ์™€ ๊ธฐ๋Šฅ์— ์ง‘์ค‘ํ•  ์ˆ˜ ์žˆ์–ด ๋” ํšจ์œจ์ ์ธ ๊ฐœ๋ฐœ ๊ณผ์ •์„ ์ด‰์ง„ํ•ฉ๋‹ˆ๋‹ค.
**์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ณ„์ธต.** ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ณ„์ธต์—์„œ๋Š” ์—ฌํ–‰ ์ƒ๋‹ด์› ๋˜๋Š” ์ˆ˜ํ•™ ์ƒ๋‹ด์›๊ณผ ๊ฐ™์€ ์—์ด์ „ํŠธ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ๊ฐœ๋ฐœํ•˜๊ณ  ๋ฐฐํฌํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ณ„์ธต์—์„œ AIOS๋Š” ์—์ด์ „ํŠธ ๊ฐœ๋ฐœ์ž์˜ ๊ฐœ๋ฐœ ํ”„๋กœ์„ธ์Šค๋ฅผ ๊ฐ„์†Œํ™”ํ•˜๋Š” ๋” ๋†’์€ ์ˆ˜์ค€์˜ ์‹œ์Šคํ…œ ํ˜ธ์ถœ ์ถ”์ƒํ™”๋ฅผ ํ†ตํ•ด AIOS SDK๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด SDK๋Š” ํ•˜์œ„ ์ˆ˜์ค€ ์‹œ์Šคํ…œ ๊ธฐ๋Šฅ์˜ ๋ณต์žก์„ฑ์„ ์ถ”์ƒํ™”ํ•˜๋Š” ํ’๋ถ€ํ•œ ํˆดํ‚ท์„ ์ œ๊ณตํ•˜์—ฌ ์—์ด์ „ํŠธ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๊ฐœ๋ฐœ์ž๋Š” ์—์ด์ „ํŠธ์˜ ํ•„์ˆ˜ ๋กœ์ง๊ณผ ๊ธฐ๋Šฅ์— ์ง‘์ค‘ํ•  ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ ๊ฐœ๋ฐœ ํ”„๋กœ์„ธ์Šค๊ฐ€ ๋”์šฑ ํšจ์œจ์ ์œผ๋กœ ์ง„ํ–‰๋ฉ๋‹ˆ๋‹ค.
Large language models (LLMs) are able to capture vast amounts of factual information (Petroni et al., 2019; Cohen et al., 2023; Hu et al., 2023). LLMs exhibit a remarkable level of knowledge in various domains due to their massive pre-training datasets. However, there are two significant limitations to this knowledge. First, it is static and does not update with time. Second, it is non-specific and thus may lack nuanced expertise in particular domains. While these are two different problems, they are deeply related since their solution is the same: enhancing the model's knowledge.
๋Œ€ํ˜• ์–ธ์–ด ๋ชจ๋ธ(LLMs)์€ ๋ฐฉ๋Œ€ํ•œ ์–‘์˜ ์‚ฌ์‹ค ์ •๋ณด๋ฅผ ํฌ์ฐฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค(Petroni et al., 2019; Cohen et al., 2023; Hu et al., 2023). LLMs๋Š” ๋Œ€๊ทœ๋ชจ ์‚ฌ์ „ ํ•™์Šต ๋ฐ์ดํ„ฐ์…‹ ๋•๋ถ„์— ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ๋†€๋ผ์šด ์ˆ˜์ค€์˜ ์ง€์‹์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋Ÿฌํ•œ ์ง€์‹์—๋Š” ๋‘ ๊ฐ€์ง€ ์ค‘์š”ํ•œ ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฒซ์งธ, ์ด๋Š” ์ •์ ์ด๋ฉฐ ์‹œ๊ฐ„์— ๋”ฐ๋ผ ์—…๋ฐ์ดํŠธ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋‘˜์งธ, ์ด๋Š” ๋น„ํŠน์ •์ ์ด์–ด์„œ ํŠน์ • ๋ถ„์•ผ์—์„œ ์„ธ๋ฐ€ํ•œ ์ „๋ฌธ ์ง€์‹์ด ๋ถ€์กฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋‘ ๊ฐ€์ง€๋Š” ์„œ๋กœ ๋‹ค๋ฅธ ๋ฌธ์ œ์ด์ง€๋งŒ, ๊ทธ ํ•ด๊ฒฐ์ฑ…์€ ๋™์ผํ•ฉ๋‹ˆ๋‹ค: ๋ชจ๋ธ์˜ ์ง€์‹์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์€ ๋ฐฉ๋Œ€ํ•œ ์–‘์˜ ์‚ฌ์‹ค ์ •๋ณด๋ฅผ ์บก์ฒ˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค(Petroni et al., 2019; Cohen et al., 2023; Hu et al., 2023). LLM์€ ๋ฐฉ๋Œ€ํ•œ ์‚ฌ์ „ ํ•™์Šต ๋ฐ์ดํ„ฐ ์„ธํŠธ๋กœ ์ธํ•ด ๋‹ค์–‘ํ•œ ์˜์—ญ์—์„œ ๋†€๋ผ์šด ์ˆ˜์ค€์˜ ์ง€์‹์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋Ÿฌํ•œ ์ง€์‹์—๋Š” ๋‘ ๊ฐ€์ง€ ์ค‘์š”ํ•œ ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฒซ์งธ, ์‹œ๊ฐ„์ด ์ง€๋‚˜๋„ ์—…๋ฐ์ดํŠธ๋˜์ง€ ์•Š๋Š” ์ •์ ์ธ ์ •๋ณด์ž…๋‹ˆ๋‹ค. ๋‘˜์งธ, ํŠน์ • ๋„๋ฉ”์ธ์— ๋Œ€ํ•œ ๋ฏธ๋ฌ˜ํ•œ ์ „๋ฌธ ์ง€์‹์ด ๋ถ€์กฑํ•  ์ˆ˜ ์žˆ๋Š” ๋น„ํŠน์ •์ ์ž…๋‹ˆ๋‹ค. ์ด ๋‘ ๊ฐ€์ง€ ๋ฌธ์ œ๋Š” ์„œ๋กœ ๋‹ค๋ฅธ ๋ฌธ์ œ์ด์ง€๋งŒ, ๊ทธ ํ•ด๊ฒฐ์ฑ…์ด ๋™์ผํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊นŠ์€ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค: ๋ชจ๋ธ์˜ ์ง€์‹์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Pretraining Data OptimizationThe objective of optimizing pre-training corpora is to maximize the performance and efficiency of language model training by improving the quality and scale of the pretrain data mixture. This typically includes data collecting through crawling (Raffel et al., 2020) or synthesis (Polu and Sutskever, 2020; Gunasekar et al., 2023), de-duplication (Lee et al., 2021; Kandpal et al., 2022; Tirumala et al., 2023), filtering and selection (Albalak et al., 2024), as well as data composition (Xie et al., 2024) and curriculum (Chen et al., 2024; MA et al., 2024).
์‚ฌ์ „ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ์ตœ์ ํ™” ์‚ฌ์ „ ํ›ˆ๋ จ ์ฝ”ํผ์Šค๋ฅผ ์ตœ์ ํ™”ํ•˜๋Š” ๋ชฉ์ ์€ ์‚ฌ์ „ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ์˜ ํ’ˆ์งˆ๊ณผ ๊ทœ๋ชจ๋ฅผ ๊ฐœ์„ ํ•˜์—ฌ ์–ธ์–ด ๋ชจ๋ธ ํ›ˆ๋ จ์˜ ์„ฑ๋Šฅ๊ณผ ํšจ์œจ์„ฑ์„ ๊ทน๋Œ€ํ™”ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ํฌ๋กค๋ง(Raffel et al., 2020) ๋˜๋Š” ํ•ฉ์„ฑ(Polu and Sutskever, 2020; Gunasekar et al., 2023)์„ ํ†ตํ•œ ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘, ์ค‘๋ณต ์ œ๊ฑฐ(Lee et al., 2021; Kandpal et al., 2022; Tirumala et al., 2023), ํ•„ํ„ฐ๋ง ๋ฐ ์„ ํƒ(Albalak et al., 2024), ๋ฐ์ดํ„ฐ ๊ตฌ์„ฑ(Xie et al., 2024) ๋ฐ ์ปค๋ฆฌํ˜๋Ÿผ(Chen et al., 2024; MA et al., 2024)์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค.
์‚ฌ์ „ ํ•™์Šต ๋ฐ์ดํ„ฐ ์ตœ์ ํ™”์‚ฌ์ „ ํ•™์Šต ๋ง๋ญ‰์น˜๋ฅผ ์ตœ์ ํ™”ํ•˜๋Š” ๋ชฉ์ ์€ ์‚ฌ์ „ ํ•™์Šต ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ์˜ ํ’ˆ์งˆ๊ณผ ๊ทœ๋ชจ๋ฅผ ๊ฐœ์„ ํ•˜์—ฌ ์–ธ์–ด ๋ชจ๋ธ ํ•™์Šต์˜ ์„ฑ๋Šฅ๊ณผ ํšจ์œจ์„ฑ์„ ๊ทน๋Œ€ํ™”ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ํฌ๋กค๋ง(Raffel et al., 2020) ๋˜๋Š” ํ•ฉ์„ฑ(Polu and Sutskever, 2020; Gunasekar et al., 2023)์„ ํ†ตํ•œ ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘, ์ค‘๋ณต ์ œ๊ฑฐ(Lee et al., 2021; Kandpal et al., 2022; Tirumala et al., 2023), ํ•„ํ„ฐ๋ง ๋ฐ ์„ ํƒ(Albalak et al., 2024), ๋ฐ์ดํ„ฐ ๊ตฌ์„ฑ(Xie et al., 2024) ๋ฐ ์ปค๋ฆฌํ˜๋Ÿผ(Chen et al., 2024; MA et al., 2024)์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค.
The luxurious and elegant nature of velvet cannot be denied, whether it is crushed, vibrant, or head-to-toe. This timeless textile is not only fashionable but also wearable for daily wear. In recent years, there has been a shift towards embracing fun gem-toned velvety pieces, breaking away from the traditional association of velvet with dark moody shades of navy and black. Below are some of the most desirable velvet pieces available on the high street today, and we are already envisioning outfits around them. Are you completely enamored with velvet or simply beyond bored with it?
๋ฒจ๋ฒณ์˜ ๊ณ ๊ธ‰์Šค๋Ÿฝ๊ณ  ์šฐ์•„ํ•œ ํŠน์„ฑ์€ ๋ถ€์ •ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์ด ๊ตฌ๊ฒจ์ง€๋“ , ์ƒ๋™๊ฐ ์žˆ๋“ , ๋จธ๋ฆฌ๋ถ€ํ„ฐ ๋ฐœ๋๊นŒ์ง€ ๋ฎ๋“  ๋ง์ด์ฃ . ์ด ์‹œ๋Œ€๋ฅผ ์ดˆ์›”ํ•œ ์ง๋ฌผ์€ ํŒจ์…”๋„ˆ๋ธ”ํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ผ์ƒ์ ์œผ๋กœ ์ž…๊ธฐ์—๋„ ์ ํ•ฉํ•ฉ๋‹ˆ๋‹ค. ์ตœ๊ทผ ๋ช‡ ๋…„ ๋™์•ˆ ๋ฒจ๋ฒณ์ด ์ „ํ†ต์ ์œผ๋กœ ์–ด๋‘์šด ํ•ด๊ตฐ์ƒ‰๊ณผ ๊ฒ€์ •์ƒ‰์˜ ์Œ์šธํ•œ ์ƒ‰์กฐ์™€ ์—ฐ๊ด€๋˜์—ˆ๋˜ ๊ฒƒ์—์„œ ๋ฒ—์–ด๋‚˜, ์žฌ๋ฏธ์žˆ๋Š” ๋ณด์„ ํ†ค์˜ ๋ฒจ๋ฒณ ์กฐ๊ฐ๋“ค์„ ๋ฐ›์•„๋“ค์ด๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ๋ณ€ํ™”๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์•„๋ž˜๋Š” ์˜ค๋Š˜๋‚  ๊ณ ๊ธ‰ ์ƒ์ ์—์„œ ๊ตฌํ•  ์ˆ˜ ์žˆ๋Š” ๊ฐ€์žฅ ๋งค๋ ฅ์ ์ธ ๋ฒจ๋ฒณ ์กฐ๊ฐ๋“ค ์ค‘ ์ผ๋ถ€์ด๋ฉฐ, ์šฐ๋ฆฌ๋Š” ์ด๋ฏธ ์ด๋“ค์„ ์ค‘์‹ฌ์œผ๋กœ ํ•œ ์˜์ƒ์„ ์ƒ์ƒํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์— ์™„์ „ํžˆ ๋งค๋ฃŒ๋˜์—ˆ๋‚˜์š”, ์•„๋‹ˆ๋ฉด ๊ทธ์ € ์ง€๋ฃจํ•ด์กŒ๋‚˜์š”?
๋ฒจ๋ฒณ์˜ ๊ณ ๊ธ‰์Šค๋Ÿฝ๊ณ  ์šฐ์•„ํ•œ ํŠน์„ฑ์€ ๋ถ€์„œ์ง€๊ฑฐ๋‚˜ ์ƒ์ƒํ•˜๊ฑฐ๋‚˜ ๋จธ๋ฆฌ๋ถ€ํ„ฐ ๋ฐœ๋๊นŒ์ง€ ์–ด๋–ค ํ˜•ํƒœ๋กœ๋“  ๋ถ€์ •ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ์ด ์‹œ๋Œ€๋ฅผ ์ดˆ์›”ํ•œ ์ง๋ฌผ์€ ํŒจ์…”๋„ˆ๋ธ”ํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ผ์ƒ๋ณต์œผ๋กœ๋„ ์ž…์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ตœ๊ทผ ๋ช‡ ๋…„ ๋™์•ˆ ๋ฒจ๋ฒณ์ด๋ผ๋Š” ์†Œ์žฌ๊ฐ€ ์ „ํ†ต์ ์œผ๋กœ ์–ด๋‘ก๊ณ  ๋ฌด๋“œ ์žˆ๋Š” ๋„ค์ด๋น„๋‚˜ ๋ธ”๋ž™ ์ƒ‰์กฐ์™€ ์—ฐ๊ด€๋˜์–ด ์žˆ๋˜ ๊ฒƒ์—์„œ ๋ฒ—์–ด๋‚˜ ์žฌ๋ฏธ์žˆ๋Š” ๋ณด์„ ํ†ค์˜ ๋ฒจ๋ฒณ ์ œํ’ˆ์„ ์ˆ˜์šฉํ•˜๋Š” ์ถ”์„ธ๊ฐ€ ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์€ ํ˜„์žฌ ํ•˜์ด์ŠคํŠธ๋ฆฌํŠธ์—์„œ ๊ฐ€์žฅ ์ธ๊ธฐ ์žˆ๋Š” ๋ฒจ๋ฒณ ์ œํ’ˆ ์ค‘ ์ผ๋ถ€์ด๋ฉฐ, ์ด๋ฏธ ๊ทธ ์ฃผ๋ณ€์˜ ์˜์ƒ์„ ์ƒ์ƒํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์— ์™„์ „ํžˆ ๋งค๋ฃŒ๋˜์—ˆ๋‚˜์š”, ์•„๋‹ˆ๋ฉด ๋‹จ์ˆœํžˆ ์ง€๋ฃจํ•ด์กŒ๋‚˜์š”?
In summary, this survey conducts a systematic review and establishes comprehensive taxonomies for existing studies in the burgeoning field of LLM-based autonomous agents. Our focus encompasses three primary areas: construction of agents, their applications, and methods of evaluation. Drawing from a wealth of previous studies, we identify various challenges in this field and discuss potential future directions. We expect that our survey can provide newcomers of LLM-based autonomous agents with a comprehensive background knowledge, and also encourage further groundbreaking studies.
์š”์•ฝํ•˜์ž๋ฉด, ์ด ์„ค๋ฌธ์กฐ์‚ฌ๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ๋ถ„์•ผ์˜ ๊ธฐ์กด ์—ฐ๊ตฌ๋“ค์„ ์ฒด๊ณ„์ ์œผ๋กœ ๊ฒ€ํ† ํ•˜๊ณ  ํฌ๊ด„์ ์ธ ๋ถ„๋ฅ˜ ์ฒด๊ณ„๋ฅผ ํ™•๋ฆฝํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์ดˆ์ ์€ ์—์ด์ „ํŠธ์˜ ๊ตฌ์„ฑ, ๊ทธ๋“ค์˜ ์‘์šฉ, ๊ทธ๋ฆฌ๊ณ  ํ‰๊ฐ€ ๋ฐฉ๋ฒ•์˜ ์„ธ ๊ฐ€์ง€ ์ฃผ์š” ์˜์—ญ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ด์ „ ์—ฐ๊ตฌ๋“ค์˜ ํ’๋ถ€ํ•œ ์ž๋ฃŒ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ, ์šฐ๋ฆฌ๋Š” ์ด ๋ถ„์•ผ์—์„œ ๋‹ค์–‘ํ•œ ๋„์ „ ๊ณผ์ œ๋ฅผ ์‹๋ณ„ํ•˜๊ณ  ์ž ์žฌ์ ์ธ ๋ฏธ๋ž˜ ๋ฐฉํ–ฅ์„ ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์„ค๋ฌธ์กฐ์‚ฌ๊ฐ€ LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ์˜ ์‹ ์ž…์ž๋“ค์—๊ฒŒ ํฌ๊ด„์ ์ธ ๋ฐฐ๊ฒฝ ์ง€์‹์„ ์ œ๊ณตํ•˜๊ณ , ๋˜ํ•œ ๋” ๋งŽ์€ ํš๊ธฐ์ ์ธ ์—ฐ๊ตฌ๋ฅผ ์žฅ๋ คํ•  ์ˆ˜ ์žˆ๊ธฐ๋ฅผ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.
์š”์•ฝํ•˜๋ฉด, ์ด ์„ค๋ฌธ์กฐ์‚ฌ๋Š” ๊ธ‰์„ฑ์žฅํ•˜๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ๋ถ„์•ผ์˜ ๊ธฐ์กด ์—ฐ๊ตฌ์— ๋Œ€ํ•œ ์ฒด๊ณ„์ ์ธ ๊ฒ€ํ† ๋ฅผ ์ˆ˜ํ–‰ํ•˜๊ณ  ํฌ๊ด„์ ์ธ ๋ถ„๋ฅ˜๋ฅผ ์ˆ˜๋ฆฝํ•ฉ๋‹ˆ๋‹ค. ์—์ด์ „ํŠธ ๊ตฌ์ถ•, ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜, ํ‰๊ฐ€ ๋ฐฉ๋ฒ•์ด๋ผ๋Š” ์„ธ ๊ฐ€์ง€ ์ฃผ์š” ์˜์—ญ์„ ์ค‘์ ์ ์œผ๋กœ ์‚ดํŽด๋ด…๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ์„ ํ–‰ ์—ฐ๊ตฌ๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ์ด ๋ถ„์•ผ์˜ ๋‹ค์–‘ํ•œ ๊ณผ์ œ๋ฅผ ํŒŒ์•…ํ•˜๊ณ  ํ–ฅํ›„ ์ž ์žฌ์ ์ธ ๋ฐฉํ–ฅ์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ์ด ์„ค๋ฌธ์กฐ์‚ฌ๋ฅผ ํ†ตํ•ด LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ์˜ ์‹ ๊ทœ ์—ฐ๊ตฌ์ž๋“ค์—๊ฒŒ ํฌ๊ด„์ ์ธ ๋ฐฐ๊ฒฝ ์ง€์‹์„ ์ œ๊ณตํ•˜๊ณ , ๋”์šฑ ํš๊ธฐ์ ์ธ ์—ฐ๊ตฌ๋ฅผ ์žฅ๋ คํ•  ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.
**Hardware Layer.** The hardware layer comprises the physical components of the system, including the CPU, GPU, memory, disk, and peripheral devices. It is crucial to note that the LLM kernel's system calls cannot directly interact with the hardware. Instead, these calls interface with the OS's system calls, which in turn manage the hardware resources. This indirect interaction ensures a layer of abstraction and security, allowing the LLM kernel to leverage hardware capabilities without requiring direct hardware management, thus maintaining the system's integrity and efficiency.
**ํ•˜๋“œ์›จ์–ด ๊ณ„์ธต.** ํ•˜๋“œ์›จ์–ด ๊ณ„์ธต์€ CPU, GPU, ๋ฉ”๋ชจ๋ฆฌ, ๋””์Šคํฌ ๋ฐ ์ฃผ๋ณ€ ์žฅ์น˜์™€ ๊ฐ™์€ ์‹œ์Šคํ…œ์˜ ๋ฌผ๋ฆฌ์  ๊ตฌ์„ฑ ์š”์†Œ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. LLM ์ปค๋„์˜ ์‹œ์Šคํ…œ ํ˜ธ์ถœ์ด ํ•˜๋“œ์›จ์–ด์™€ ์ง์ ‘ ์ƒํ˜ธ์ž‘์šฉํ•  ์ˆ˜ ์—†๋‹ค๋Š” ์ ์„ ์ฃผ๋ชฉํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ๋Œ€์‹ , ์ด๋Ÿฌํ•œ ํ˜ธ์ถœ์€ OS์˜ ์‹œ์Šคํ…œ ํ˜ธ์ถœ๊ณผ ์ธํ„ฐํŽ˜์ด์Šคํ•˜๋ฉฐ, ์ด๋Š” ํ•˜๋“œ์›จ์–ด ์ž์›์„ ๊ด€๋ฆฌํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฐ„์ ‘์ ์ธ ์ƒํ˜ธ์ž‘์šฉ์€ ์ถ”์ƒํ™”์™€ ๋ณด์•ˆ์˜ ๊ณ„์ธต์„ ๋ณด์žฅํ•˜์—ฌ, LLM ์ปค๋„์ด ์ง์ ‘์ ์ธ ํ•˜๋“œ์›จ์–ด ๊ด€๋ฆฌ๋ฅผ ์š”๊ตฌํ•˜์ง€ ์•Š๊ณ ๋„ ํ•˜๋“œ์›จ์–ด ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•˜์—ฌ ์‹œ์Šคํ…œ์˜ ๋ฌด๊ฒฐ์„ฑ๊ณผ ํšจ์œจ์„ฑ์„ ์œ ์ง€ํ•ฉ๋‹ˆ๋‹ค.
**ํ•˜๋“œ์›จ์–ด ๊ณ„์ธต.** ํ•˜๋“œ์›จ์–ด ๊ณ„์ธต์€ CPU, GPU, ๋ฉ”๋ชจ๋ฆฌ, ๋””์Šคํฌ ๋ฐ ์ฃผ๋ณ€ ์žฅ์น˜๋ฅผ ํฌํ•จํ•œ ์‹œ์Šคํ…œ์˜ ๋ฌผ๋ฆฌ์  ๊ตฌ์„ฑ ์š”์†Œ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. LLM ์ปค๋„์˜ ์‹œ์Šคํ…œ ํ˜ธ์ถœ์€ ํ•˜๋“œ์›จ์–ด์™€ ์ง์ ‘ ์ƒํ˜ธ ์ž‘์šฉํ•  ์ˆ˜ ์—†๋‹ค๋Š” ์ ์— ์œ ์˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋Œ€์‹ , ์ด๋Ÿฌํ•œ ํ˜ธ์ถœ์€ ํ•˜๋“œ์›จ์–ด ๋ฆฌ์†Œ์Šค๋ฅผ ๊ด€๋ฆฌํ•˜๋Š” OS์˜ ์‹œ์Šคํ…œ ํ˜ธ์ถœ๊ณผ ์ธํ„ฐํŽ˜์ด์Šคํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฐ„์ ‘์ ์ธ ์ƒํ˜ธ ์ž‘์šฉ์€ ์ถ”์ƒํ™” ๋ฐ ๋ณด์•ˆ ๊ณ„์ธต์„ ๋ณด์žฅํ•˜์—ฌ LLM ์ปค๋„์ด ์ง์ ‘ ํ•˜๋“œ์›จ์–ด ๊ด€๋ฆฌ ์—†์ด๋„ ํ•˜๋“œ์›จ์–ด ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜์—ฌ ์‹œ์Šคํ…œ์˜ ๋ฌด๊ฒฐ์„ฑ๊ณผ ํšจ์œจ์„ฑ์„ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Future research directions on synthetic data could focus on improving the fidelity and controllability of generative models and developing standardized evaluation and contamination protocols and tools. We could also explore the integration of synthetic data with other techniques and its application in other domains. Despite the challenges, the potential benefits of synthetic data in advancing AI research are significant. By leveraging synthetic data responsibly and effectively, we can build more powerful, inclusive, and trustworthy AI systems that benefit society as a whole.
๋ฏธ๋ž˜์˜ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ ์—ฐ๊ตฌ ๋ฐฉํ–ฅ์€ ์ƒ์„ฑ ๋ชจ๋ธ์˜ ์ถฉ์‹ค๋„์™€ ์ œ์–ด ๊ฐ€๋Šฅ์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ณ , ํ‘œ์ค€ํ™”๋œ ํ‰๊ฐ€ ๋ฐ ์˜ค์—ผ ํ”„๋กœํ† ์ฝœ๊ณผ ๋„๊ตฌ๋ฅผ ๊ฐœ๋ฐœํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘˜ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ๋‹ค๋ฅธ ๊ธฐ์ˆ ๊ณผ ํ†ตํ•ฉํ•˜๊ณ  ๋‹ค๋ฅธ ๋„๋ฉ”์ธ์— ์ ์šฉํ•˜๋Š” ๋ฐฉ์•ˆ์„ ํƒ๊ตฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋„์ „ ๊ณผ์ œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , AI ์—ฐ๊ตฌ๋ฅผ ๋ฐœ์ „์‹œํ‚ค๋Š” ๋ฐ ์žˆ์–ด ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ์˜ ์ž ์žฌ์  ์ด์ ์€ ํฝ๋‹ˆ๋‹ค. ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ์ฑ…์ž„๊ฐ ์žˆ๊ณ  ํšจ๊ณผ์ ์œผ๋กœ ํ™œ์šฉํ•จ์œผ๋กœ์จ, ์‚ฌํšŒ ์ „์ฒด์— ์ด์ต์ด ๋˜๋Š” ๋” ๊ฐ•๋ ฅํ•˜๊ณ  ํฌ์šฉ์ ์ด๋ฉฐ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” AI ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•œ ํ–ฅํ›„ ์—ฐ๊ตฌ ๋ฐฉํ–ฅ์€ ์ƒ์„ฑ ๋ชจ๋ธ์˜ ์ถฉ์‹ค๋„์™€ ์ œ์–ด ๊ฐ€๋Šฅ์„ฑ์„ ๊ฐœ์„ ํ•˜๊ณ  ํ‘œ์ค€ํ™”๋œ ํ‰๊ฐ€ ๋ฐ ์˜ค์—ผ ํ”„๋กœํ† ์ฝœ๊ณผ ๋„๊ตฌ๋ฅผ ๊ฐœ๋ฐœํ•˜๋Š” ๋ฐ ์ดˆ์ ์„ ๋งž์ถœ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋‹ค๋ฅธ ๊ธฐ์ˆ ๊ณผ์˜ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ ํ†ตํ•ฉ๊ณผ ๋‹ค๋ฅธ ์˜์—ญ์—์˜ ์ ์šฉ์„ ํƒ๊ตฌํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์–ด๋ ค์›€์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  AI ์—ฐ๊ตฌ๋ฅผ ๋ฐœ์ „์‹œํ‚ค๋Š” ๋ฐ ์žˆ์–ด ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ์˜ ์ž ์žฌ์  ์ด์ ์€ ์ƒ๋‹นํ•ฉ๋‹ˆ๋‹ค. ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ์ฑ…์ž„๊ฐ ์žˆ๊ณ  ํšจ๊ณผ์ ์œผ๋กœ ํ™œ์šฉํ•จ์œผ๋กœ์จ ์šฐ๋ฆฌ๋Š” ์‚ฌํšŒ ์ „์ฒด์— ๋„์›€์ด ๋˜๋Š” ๋” ๊ฐ•๋ ฅํ•˜๊ณ  ํฌ์šฉ์ ์ด๋ฉฐ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” AI ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Prompting LLMs with large chunks (_entire books_) to evaluate the faithfulness of each claim is prohibitively expensive (see SSF.5). Hence, for this experiment, we select 7 books based on: (1) token length (\(<\)125K), and (2) presence of at least one _Unfaithful_ claim. This sub-dataset includes: (1) "Yellowface," (2) "Only For The Week," (3) "Viciously Yours," (4) "Six Scorched Roses," (5) "Sorrow and Bliss," (6) "She Is a Haunting," and (7) "Pet." Table Table 31 shows the number of claims per label in the sub-dataset. Further details on each book can be found in Table 7.
LLMs์— ํฐ ๋ฉ์–ด๋ฆฌ์˜ ํ…์ŠคํŠธ(์ „์ฒด ์ฑ…)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ ์ฃผ์žฅ์— ๋Œ€ํ•œ ์‹ ๋ขฐ์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ๋น„์šฉ์ด ๋งค์šฐ ๋งŽ์ด ๋“ญ๋‹ˆ๋‹ค(SSF.5 ์ฐธ์กฐ). ๋”ฐ๋ผ์„œ ์ด ์‹คํ—˜์—์„œ๋Š” ๋‹ค์Œ ๊ธฐ์ค€์— ๋”ฐ๋ผ 7๊ถŒ์˜ ์ฑ…์„ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค: (1) ํ† ํฐ ๊ธธ์ด(\(<\)125K), (2) ์ ์–ด๋„ ํ•˜๋‚˜์˜ _Unfaithful_ ์ฃผ์žฅ์ด ํฌํ•จ๋˜์–ด ์žˆ๋Š”์ง€. ์ด ํ•˜์œ„ ๋ฐ์ดํ„ฐ์…‹์—๋Š” ๋‹ค์Œ์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค: (1) "Yellowface," (2) "Only For The Week," (3) "Viciously Yours," (4) "Six Scorched Roses," (5) "Sorrow and Bliss," (6) "She Is a Haunting," ๊ทธ๋ฆฌ๊ณ  (7) "Pet." Table 31์€ ํ•˜์œ„ ๋ฐ์ดํ„ฐ์…‹์—์„œ ๋ ˆ์ด๋ธ”๋ณ„ ์ฃผ์žฅ ์ˆ˜๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ฐ ์ฑ…์— ๋Œ€ํ•œ ์ถ”๊ฐ€ ์„ธ๋ถ€ ์‚ฌํ•ญ์€ Table 7์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๊ฐ ์ฃผ์žฅ์˜ ์ถฉ์‹ค์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ํฐ ๋ฉ์–ด๋ฆฌ(_์ „์ฒด ์ฑ…_)๋กœ LLM์„ ํ”„๋กฌํ”„ํŠธํ•˜๋Š” ๊ฒƒ์€ ๋น„์šฉ์ด ๋„ˆ๋ฌด ๋งŽ์ด ๋“ญ๋‹ˆ๋‹ค(SSF.5 ์ฐธ์กฐ). ๋”ฐ๋ผ์„œ ์ด ์‹คํ—˜์—์„œ๋Š” (1) ํ† ํฐ ๊ธธ์ด(\(<\)125K)์™€ (2) ํ•˜๋‚˜ ์ด์ƒ์˜ _๋ถˆ์ถฉ์‹ค_ ์ฃผ์žฅ์ด ์žˆ๋Š”์ง€ ์—ฌ๋ถ€๋ฅผ ๊ธฐ์ค€์œผ๋กœ 7๊ถŒ์˜ ์ฑ…์„ ์„ ์ •ํ•ฉ๋‹ˆ๋‹ค. ์ด ํ•˜์œ„ ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋Š” (1) "์˜๋กœ์šฐํŽ˜์ด์Šค", (2) "ํ•œ ์ฃผ ๋™์•ˆ๋งŒ", (3) "์•…๋ž„ํ•˜๊ฒŒ ๋‹น์‹ ", (4) "์—ฌ์„ฏ ๊ฐœ์˜ ๊ทธ์„๋ฆฐ ์žฅ๋ฏธ", (5) "์Šฌํ””๊ณผ ๊ธฐ์จ", (6) "๊ทธ๋…€๋Š” ์œ ๋ น์ด๋‹ค", (7) "์• ์™„๋™๋ฌผ"์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ํ‘œ 31์€ ํ•˜์œ„ ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๋ ˆ์ด๋ธ”๋ณ„ ํด๋ ˆ์ž„ ์ˆ˜๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ฐ ์ฑ…์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ํ‘œ 7์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Whether you like your velvet crushed, vibrant or head-to-toe, there's really no denying the sheer luxe and elegance of this timeless textile. Not only is it super stylish, it can actually be so wearable for day-to-day wear. Yes, really! This year it's all about embracing fun gem-toned velvety pieces. Long gone are the days when velvet was solely associated with dark moody shades of navy and black. Below we've rounded up the most covetable velvet pieces on the high street right now. We're already coming up with outfit ideas! Are you completely obsessed or beyond bored of it?
๋ฒจ๋ฒณ์„ ์œผ๊นฌ ๊ฒƒ์ด๋“ , ์ƒ๋™๊ฐ ์žˆ๋Š” ๊ฒƒ์ด๋“ , ๋จธ๋ฆฌ๋ถ€ํ„ฐ ๋ฐœ๋๊นŒ์ง€ ์ž…๋Š” ๊ฒƒ์ด๋“ , ์ด ์‹œ๋Œ€๋ฅผ ์ดˆ์›”ํ•œ ์ง๋ฌผ์˜ ์ˆœ์ˆ˜ํ•œ ๋Ÿญ์…”๋ฆฌ์™€ ์šฐ์•„ํ•จ์„ ๋ถ€์ •ํ•  ์ˆ˜๋Š” ์—†์Šต๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์€ ๋งค์šฐ ์Šคํƒ€์ผ๋ฆฌ์‹œํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ผ์ƒ์ ์œผ๋กœ ์ž…๊ธฐ์—๋„ ์ •๋ง ์ฐฉ์šฉ๊ฐ์ด ์ข‹์Šต๋‹ˆ๋‹ค. ๋„ค, ์ •๋ง์ž…๋‹ˆ๋‹ค! ์˜ฌํ•ด๋Š” ๋ณด์„ ํ†ค์˜ ์žฌ๋ฏธ์žˆ๋Š” ๋ฒจ๋ฒณ ์กฐ๊ฐ๋“ค์„ ๋ฐ›์•„๋“ค์ด๋Š” ๊ฒƒ์ด ์ „๋ถ€์ž…๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์ด ์–ด๋‘์šด ํ•ด๊ตฐ์ƒ‰๊ณผ ๊ฒ€์ •์ƒ‰์˜ ์Œ์šธํ•œ ์ƒ‰์กฐ์™€๋งŒ ์—ฐ๊ด€๋˜๋˜ ์‹œ์ ˆ์€ ์˜ค๋ž˜์ „์— ์ง€๋‚˜๊ฐ”์Šต๋‹ˆ๋‹ค. ์•„๋ž˜์—์„œ ํ˜„์žฌ ๊ณ ๊ธ‰ ์ƒ์ ์—์„œ ๊ฐ€์žฅ ํƒ๋‚˜๋Š” ๋ฒจ๋ฒณ ์กฐ๊ฐ๋“ค์„ ๋ชจ์•„๋ดค์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด๋ฏธ ์˜์ƒ ์•„์ด๋””์–ด๋ฅผ ๋– ์˜ฌ๋ฆฌ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค! ์—ฌ๋Ÿฌ๋ถ„์€ ์™„์ „ํžˆ ๋ฒจ๋ฒณ์— ๋น ์ ธ ์žˆ๋‚˜์š”, ์•„๋‹ˆ๋ฉด ์ง€๋ฃจํ•ดํ•˜๊ณ  ์žˆ๋‚˜์š”?
๋ฒจ๋ฒณ์„ ๋ถ€์ˆ˜๊ณ , ์ƒ์ƒํ•˜๊ฒŒ, ๋จธ๋ฆฌ๋ถ€ํ„ฐ ๋ฐœ๋๊นŒ์ง€ ์ข‹์•„ํ•˜๋“ , ์ด ์‹œ๋Œ€๋ฅผ ์ดˆ์›”ํ•œ ์„ฌ์œ ์˜ ๊ณ ๊ธ‰์Šค๋Ÿฌ์›€๊ณผ ์šฐ์•„ํ•จ์„ ๋ถ€์ •ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ๋งค์šฐ ์Šคํƒ€์ผ๋ฆฌ์‹œํ•  ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์‹ค์ œ๋กœ ์ผ์ƒ์ ์œผ๋กœ ์ฐฉ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ž˜, ์ •๋ง! ์˜ฌํ•ด๋Š” ์žฌ๋ฏธ์žˆ๋Š” ๋ณด์„ ํ†ค์˜ ๋ฒจ๋ฒณ ์˜ท์„ ๋ฐ›์•„๋“ค์ด๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ๋ฒจ๋ฒณ์ด ์ง™์€ ๋ฌด๋“œ์˜ ๋„ค์ด๋น„์™€ ๋ธ”๋ž™ ์ƒ‰์กฐ์™€๋งŒ ์—ฐ๊ด€๋˜๋˜ ์‹œ์ ˆ์€ ์ด๋ฏธ ์ง€๋‚˜๊ฐ”์Šต๋‹ˆ๋‹ค. ์•„๋ž˜์—์„œ๋Š” ํ˜„์žฌ ํ•˜์ด์ŠคํŠธ๋ฆฌํŠธ์—์„œ ๊ฐ€์žฅ ์ธ๊ธฐ ์žˆ๋Š” ๋ฒจ๋ฒณ ์ œํ’ˆ์„ ์ •๋ฆฌํ•ด ๋ณด์•˜์Šต๋‹ˆ๋‹ค. ์ด๋ฏธ ์˜์ƒ ์•„์ด๋””์–ด๊ฐ€ ๋– ์˜ค๋ฅด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค! ์™„์ „ํžˆ ์ง‘์ฐฉํ•˜์‹œ๋‚˜์š”, ์•„๋‹ˆ๋ฉด ์ง€๋ฃจํ•ดํ•˜์‹œ๋‚˜์š”?
Language Models have been known to exhibit bias both in terms of evaluation as well as in social or fairness terms [5]. Moreover, agents have specifically been shown to be "less robust, prone to more harmful behaviors, and capable of generating stealthier content than LLMs, highlighting significant safety challenges" [25]. Other research has found "a tendency for LLM agents to conform to the model's inherent social biases despite being directed to debate from certain political perspectives" [24]. This tendency can lead to faulty reasoning in any agent-based implementation.
์–ธ์–ด ๋ชจ๋ธ์€ ํ‰๊ฐ€ ์ธก๋ฉด๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์‚ฌํšŒ์  ๋˜๋Š” ๊ณต์ •์„ฑ ์ธก๋ฉด์—์„œ๋„ ํŽธํ–ฅ์„ ๋‚˜ํƒ€๋‚ด๋Š” ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์ ธ ์žˆ์Šต๋‹ˆ๋‹ค [5]. ๋”์šฑ์ด, ์—์ด์ „ํŠธ๋Š” "๋œ ๊ฒฌ๊ณ ํ•˜๊ณ , ๋” ํ•ด๋กœ์šด ํ–‰๋™์„ ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์œผ๋ฉฐ, LLM๋ณด๋‹ค ๋” ์€๋ฐ€ํ•œ ์ฝ˜ํ…์ธ ๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์–ด ์ค‘์š”ํ•œ ์•ˆ์ „ ๋ฌธ์ œ๋ฅผ ๋ถ€๊ฐ์‹œํ‚จ๋‹ค"๊ณ  ๊ตฌ์ฒด์ ์œผ๋กœ ๋ณด์—ฌ์กŒ์Šต๋‹ˆ๋‹ค [25]. ๋‹ค๋ฅธ ์—ฐ๊ตฌ์—์„œ๋Š” "ํŠน์ • ์ •์น˜์  ๊ด€์ ์—์„œ ํ† ๋ก ํ•˜๋„๋ก ์ง€์‹œ๋ฐ›์•˜์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  LLM ์—์ด์ „ํŠธ๊ฐ€ ๋ชจ๋ธ์˜ ๊ณ ์œ ํ•œ ์‚ฌํšŒ์  ํŽธํ–ฅ์— ๋”ฐ๋ฅด๋Š” ๊ฒฝํ–ฅ์ด ์žˆ๋‹ค"๊ณ  ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค [24]. ์ด๋Ÿฌํ•œ ๊ฒฝํ–ฅ์€ ์—์ด์ „ํŠธ ๊ธฐ๋ฐ˜ ๊ตฌํ˜„์—์„œ ์ž˜๋ชป๋œ ์ถ”๋ก ์„ ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์–ธ์–ด ๋ชจ๋ธ์€ ํ‰๊ฐ€ ์ธก๋ฉด๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์‚ฌํšŒ์  ๋˜๋Š” ๊ณต์ •์„ฑ ์ธก๋ฉด์—์„œ๋„ ํŽธํ–ฅ์„ฑ์„ ๋ณด์ด๋Š” ๊ฒƒ์œผ๋กœ ์•Œ๋ ค์ ธ ์žˆ์Šต๋‹ˆ๋‹ค[5]. ๋˜ํ•œ ์—์ด์ „ํŠธ๋Š” ํŠนํžˆ "LLM๋ณด๋‹ค ๋œ ๊ฐ•๋ ฅํ•˜๊ณ , ๋” ๋งŽ์€ ํ•ด๋กœ์šด ํ–‰๋™์„ ์ผ์œผํ‚ค๋ฉฐ, ๋” ์€๋ฐ€ํ•œ ์ฝ˜ํ…์ธ ๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ์–ด ์‹ฌ๊ฐํ•œ ์•ˆ์ „ ๋ฌธ์ œ๋ฅผ ์•ผ๊ธฐํ•œ๋‹ค๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค."[25]. ๋‹ค๋ฅธ ์—ฐ๊ตฌ์—์„œ๋Š” "ํŠน์ • ์ •์น˜์  ๊ด€์ ์—์„œ ํ† ๋ก ํ•˜๋„๋ก ์ง€์‹œ๋ฐ›์•˜์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  LLM ์—์ด์ „ํŠธ๊ฐ€ ๋ชจ๋ธ์— ๋‚ด์žฌ๋œ ์‚ฌํšŒ์  ํŽธ๊ฒฌ์— ๋”ฐ๋ฅด๋Š” ๊ฒฝํ–ฅ์ด ์žˆ๋‹ค"๊ณ  ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค[24]. ์ด๋Ÿฌํ•œ ๊ฒฝํ–ฅ์€ ๋ชจ๋“  ์—์ด์ „ํŠธ ๊ธฐ๋ฐ˜ ๊ตฌํ˜„์—์„œ ์ž˜๋ชป๋œ ์ถ”๋ก ์œผ๋กœ ์ด์–ด์งˆ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
The LayerNorm operation is the composition of a centering operation \(x\mapsto x-\bar{x}\) with RMSNorm (i.e. first centering is applied, then RMSNorm). Therefore the identical argument to Appendix C.4 goes through, besides the fact that we need \(\mathbf{U}\) to also commute with the centering operation. Since the centering operation fixes a \((h-1)\) dimensional subspace defined by \(\mathbf{1}^{T}x=0\) where \(\mathbf{1}\in\mathds{R}^{h}\) is the vector of ones, it is enough to impose an additional condition that \(\mathbf{U}\,\mathbf{1}\in\{-\mathbf{1},\mathbf{1}\}\).
LayerNorm ์—ฐ์‚ฐ์€ RMSNorm๊ณผ ํ•จ๊ป˜ ์ค‘์‹ฌํ™” ์—ฐ์‚ฐ \(x\mapsto x-\bar{x}\)์˜ ์กฐํ•ฉ์ž…๋‹ˆ๋‹ค (์ฆ‰, ๋จผ์ € ์ค‘์‹ฌํ™”๊ฐ€ ์ ์šฉ๋˜๊ณ , ๊ทธ ๋‹ค์Œ RMSNorm์ด ์ ์šฉ๋ฉ๋‹ˆ๋‹ค). ๋”ฐ๋ผ์„œ Appendix C.4์™€ ๋™์ผํ•œ ๋…ผ๋ฆฌ๊ฐ€ ์ ์šฉ๋˜๋ฉฐ, ์šฐ๋ฆฌ๋Š” \(\mathbf{U}\)๊ฐ€ ์ค‘์‹ฌํ™” ์—ฐ์‚ฐ๊ณผ๋„ ๊ตํ™˜ ๊ฐ€๋Šฅํ•ด์•ผ ํ•œ๋‹ค๋Š” ์‚ฌ์‹ค์„ ์ œ์™ธํ•˜๊ณ ๋Š” ๋™์ผํ•ฉ๋‹ˆ๋‹ค. ์ค‘์‹ฌํ™” ์—ฐ์‚ฐ์€ \(\mathbf{1}^{T}x=0\)์œผ๋กœ ์ •์˜๋œ \((h-1)\) ์ฐจ์›์˜ ๋ถ€๋ถ„ ๊ณต๊ฐ„์„ ๊ณ ์ •ํ•˜๋ฏ€๋กœ, \(\mathbf{1}\in\mathds{R}^{h}\)๊ฐ€ ๋ชจ๋“  ์š”์†Œ๊ฐ€ 1์ธ ๋ฒกํ„ฐ์ผ ๋•Œ, \(\mathbf{U}\,\mathbf{1}\in\{-\mathbf{1},\mathbf{1}\}\)์ด๋ผ๋Š” ์ถ”๊ฐ€ ์กฐ๊ฑด์„ ๋ถ€๊ณผํ•˜๋Š” ๊ฒƒ์œผ๋กœ ์ถฉ๋ถ„ํ•ฉ๋‹ˆ๋‹ค.
LayerNorm ์—ฐ์‚ฐ์€ ์ค‘์‹ฌ ์—ฐ์‚ฐ \(x\mapsto x-\bar{x}\)์™€ RMSNorm(์ฆ‰, ๋จผ์ € ์ค‘์‹ฌ์„ ์ ์šฉํ•œ ๋‹ค์Œ RMSNorm)์˜ ๊ตฌ์„ฑ์ž…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ถ€๋ก C.4์™€ ๋™์ผํ•œ ์ธ์ˆ˜๊ฐ€ ์ ์šฉ๋˜๋ฉฐ, ์ค‘์‹ฌ ์—ฐ์‚ฐ๊ณผ ํ†ต๊ทผํ•˜๋Š” \(\mathbf{U}\)๊ฐ€ ํ•„์š”ํ•˜๋‹ค๋Š” ์‚ฌ์‹ค ์™ธ์—๋Š” ๋ถ€๋ก C.4์™€ ๋™์ผํ•œ ์ธ์ˆ˜๊ฐ€ ์ ์šฉ๋ฉ๋‹ˆ๋‹ค. ์ค‘์‹ฌ ์—ฐ์‚ฐ์€ \(\mathbf{1}^{T}x=0\)์œผ๋กœ ์ •์˜๋˜๋Š” \((h-1)\) ์ฐจ์›์˜ ๋ถ€๋ถ„ ๊ณต๊ฐ„์„ ์ˆ˜์ •ํ•˜๋ฏ€๋กœ, \(\mathbf{U}\,\mathbf{1}\in\{-\mathbf{1},\mathbf{1}\}\)๋ผ๋Š” ์ถ”๊ฐ€ ์กฐ๊ฑด์„ ๋ถ€๊ณผํ•˜๋ฉด ์ถฉ๋ถ„ํ•ฉ๋‹ˆ๋‹ค.
A key problem of weight interpolation-based methods is that they neglect parameter interference, leading to performance degradation. A recent work [49] identifies two key sources of interference: redundant parameter values and conflicting parameter signs across models, and proposes a method that addresses these issues to achieve improved merging performance. The proposed TIES-Merging method addresses information loss in existing merging methods by incorporating three steps: resetting minimal parameter changes, resolving sign conflicts, and merging only aligned parameters.
๊ฐ€์ค‘์น˜ ๋ณด๊ฐ„ ๊ธฐ๋ฐ˜ ๋ฐฉ๋ฒ•์˜ ์ฃผ์š” ๋ฌธ์ œ๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ ๊ฐ„์„ญ์„ ๋ฌด์‹œํ•˜์—ฌ ์„ฑ๋Šฅ ์ €ํ•˜๋ฅผ ์ดˆ๋ž˜ํ•œ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ์ตœ๊ทผ ์—ฐ๊ตฌ [49]์—์„œ๋Š” ๊ฐ„์„ญ์˜ ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ์›์ฒœ์„ ์‹๋ณ„ํ–ˆ์Šต๋‹ˆ๋‹ค: ์ค‘๋ณต๋œ ๋งค๊ฐœ๋ณ€์ˆ˜ ๊ฐ’๊ณผ ๋ชจ๋ธ ๊ฐ„์˜ ์ƒ์ถฉ๋˜๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ ๋ถ€ํ˜ธ์ž…๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜์—ฌ ํ–ฅ์ƒ๋œ ๋ณ‘ํ•ฉ ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. ์ œ์•ˆ๋œ TIES-Merging ๋ฐฉ๋ฒ•์€ ์ตœ์†Œํ•œ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜ ๋ณ€๊ฒฝ์„ ์žฌ์„ค์ •ํ•˜๊ณ , ๋ถ€ํ˜ธ ์ถฉ๋Œ์„ ํ•ด๊ฒฐํ•˜๋ฉฐ, ์ •๋ ฌ๋œ ๋งค๊ฐœ๋ณ€์ˆ˜๋งŒ ๋ณ‘ํ•ฉํ•˜๋Š” ์„ธ ๊ฐ€์ง€ ๋‹จ๊ณ„๋ฅผ ํฌํ•จํ•˜์—ฌ ๊ธฐ์กด ๋ณ‘ํ•ฉ ๋ฐฉ๋ฒ•์—์„œ์˜ ์ •๋ณด ์†์‹ค์„ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค.
๊ฐ€์ค‘์น˜ ๋ณด๊ฐ„ ๊ธฐ๋ฐ˜ ๋ฐฉ๋ฒ•์˜ ์ฃผ์š” ๋ฌธ์ œ๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ ๊ฐ„์„ญ์„ ๋ฌด์‹œํ•˜์—ฌ ์„ฑ๋Šฅ ์ €ํ•˜๋กœ ์ด์–ด์ง„๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ตœ๊ทผ ์—ฐ๊ตฌ[49]์—์„œ๋Š” ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ๊ฐ„์„ญ์›์ธ, ์ฆ‰ ์ค‘๋ณต๋œ ํŒŒ๋ผ๋ฏธํ„ฐ ๊ฐ’๊ณผ ๋ชจ๋ธ ๊ฐ„ ์ถฉ๋Œํ•˜๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ ๋ถ€ํ˜ธ๋ฅผ ํŒŒ์•…ํ•˜๊ณ  ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜์—ฌ ๊ฐœ์„ ๋œ ๋ณ‘ํ•ฉ ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ œ์•ˆํ–ˆ์Šต๋‹ˆ๋‹ค. ์ œ์•ˆ๋œ TIES-Merging ๋ฐฉ๋ฒ•์€ ์ตœ์†Œ ํŒŒ๋ผ๋ฏธํ„ฐ ๋ณ€๊ฒฝ ์‚ฌํ•ญ ์žฌ์„ค์ •, ๋ถ€ํ˜ธ ์ถฉ๋Œ ํ•ด๊ฒฐ, ์ •๋ ฌ๋œ ํŒŒ๋ผ๋ฏธํ„ฐ๋งŒ ๋ณ‘ํ•ฉํ•˜๋Š” ์„ธ ๋‹จ๊ณ„๋ฅผ ํ†ตํ•ฉํ•˜์—ฌ ๊ธฐ์กด ๋ณ‘ํ•ฉ ๋ฐฉ๋ฒ•์˜ ์ •๋ณด ์†์‹ค์„ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค.
i'm not sure, cannot find, does not provide, cannot provide, cannot answer, cannot be found, cannot be determined, don't have information, do not have information, couldn't find, no information in the context, does not mention, not explicitly mentioned, i don't have any, i do not have any, does not specify, doesn't provide, not able to, unable to, doesn't specify, there is no information, there is no mention, not mentioned, i don't have enough information, there is no specific information, there is no specific mention, no information found, I don't have that information
์ž˜ ๋ชจ๋ฅด๊ฒ ์Šต๋‹ˆ๋‹ค, ์ฐพ์„ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ์ œ๊ณตํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค, ์ œ๊ณตํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ์ฐพ์„ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ๊ฒฐ์ •ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ์ฐพ์„ ์ˆ˜ ์—†์—ˆ์Šต๋‹ˆ๋‹ค, ๋ฌธ๋งฅ์— ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ์–ธ๊ธ‰๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค, ๋ช…์‹œ์ ์œผ๋กœ ์–ธ๊ธ‰๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค, ์•„๋ฌด๊ฒƒ๋„ ์—†์Šต๋‹ˆ๋‹ค, ์•„๋ฌด๊ฒƒ๋„ ์—†์Šต๋‹ˆ๋‹ค, ๋ช…์‹œํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค, ์ œ๊ณตํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค, ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค, ๋ช…์‹œํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค, ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ์–ธ๊ธ‰์ด ์—†์Šต๋‹ˆ๋‹ค, ์–ธ๊ธ‰๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค, ์ถฉ๋ถ„ํ•œ ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ๊ตฌ์ฒด์ ์ธ ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ๊ตฌ์ฒด์ ์ธ ์–ธ๊ธ‰์ด ์—†์Šต๋‹ˆ๋‹ค, ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค, ๊ทธ ์ •๋ณด๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค.
ํ™•์‹คํ•˜์ง€ ์•Š์Œ, ์ฐพ์„ ์ˆ˜ ์—†์Œ, ์ œ๊ณตํ•˜์ง€ ์•Š์Œ, ์ œ๊ณตํ•  ์ˆ˜ ์—†์Œ, ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†์Œ, ์ฐพ์„ ์ˆ˜ ์—†์Œ, ๊ฒฐ์ •ํ•  ์ˆ˜ ์—†์Œ, ์ •๋ณด๊ฐ€ ์—†์Œ, ์ •๋ณด๊ฐ€ ์—†์Œ, ์ฐพ์„ ์ˆ˜ ์—†์Œ, ๋ฌธ๋งฅ์— ์ •๋ณด๊ฐ€ ์—†์Œ, ์–ธ๊ธ‰๋˜์ง€ ์•Š์Œ, ๋ช…์‹œ์ ์œผ๋กœ ์–ธ๊ธ‰๋˜์ง€ ์•Š์Œ, ์—†์Œ, ์—†์Œ, ์—†์Œ, ๋ช…์‹œ๋˜์ง€ ์•Š์Œ, ์ œ๊ณต๋˜์ง€ ์•Š์Œ, ๋ถˆ๊ฐ€๋Šฅ, ๋ถˆ๊ฐ€๋Šฅ, ๋ช…์‹œ๋˜์ง€ ์•Š์Œ, ์ •๋ณด๊ฐ€ ์—†์Œ, ์–ธ๊ธ‰๋˜์ง€ ์•Š์Œ, ์–ธ๊ธ‰๋˜์ง€ ์•Š์Œ, ์ •๋ณด๊ฐ€ ์—†์Œ, ์ •๋ณด๊ฐ€ ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์Œ, ๊ตฌ์ฒด์ ์ธ ์ •๋ณด๊ฐ€ ์—†์Œ, ๊ตฌ์ฒด์ ์œผ๋กœ ์–ธ๊ธ‰๋˜์ง€ ์•Š์Œ, ์ •๋ณด๊ฐ€ ๋ฐœ๊ฒฌ๋˜์ง€ ์•Š์Œ, ํ•ด๋‹น ์ •๋ณด๊ฐ€ ์—†์Œ
Advanced Scheduling Algorithms. The scheduling function of AIOS lays the groundwork for the development of more advanced algorithms. Future research could focus on algorithms that perform dependency analysis among agent requests, optimizing the allocation of computational resources. Additionally, some of the tool resources are locally deployed models, which can also be incorporated into the scheduling paradigm. This includes the management of tool status and snapshots, suggesting a move towards a unified scheduling framework that encompasses both agents and their tools.
๊ณ ๊ธ‰ ์Šค์ผ€์ค„๋ง ์•Œ๊ณ ๋ฆฌ์ฆ˜. AIOS์˜ ์Šค์ผ€์ค„๋ง ๊ธฐ๋Šฅ์€ ๋” ๋ฐœ์ „๋œ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๊ฐœ๋ฐœ์˜ ๊ธฐ์ดˆ๋ฅผ ๋งˆ๋ จํ•ฉ๋‹ˆ๋‹ค. ํ–ฅํ›„ ์—ฐ๊ตฌ๋Š” ์—์ด์ „ํŠธ ์š”์ฒญ ๊ฐ„์˜ ์˜์กด์„ฑ ๋ถ„์„์„ ์ˆ˜ํ–‰ํ•˜๊ณ , ๊ณ„์‚ฐ ์ž์›์˜ ํ• ๋‹น์„ ์ตœ์ ํ™”ํ•˜๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ์ดˆ์ ์„ ๋งž์ถœ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ผ๋ถ€ ๋„๊ตฌ ์ž์›์€ ๋กœ์ปฌ์— ๋ฐฐํฌ๋œ ๋ชจ๋ธ๋กœ, ์Šค์ผ€์ค„๋ง ํŒจ๋Ÿฌ๋‹ค์ž„์— ํ†ตํ•ฉ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ๋„๊ตฌ ์ƒํƒœ ๋ฐ ์Šค๋ƒ…์ƒท ๊ด€๋ฆฌ๊ฐ€ ํฌํ•จ๋˜๋ฉฐ, ์—์ด์ „ํŠธ์™€ ๊ทธ ๋„๊ตฌ๋ฅผ ๋ชจ๋‘ ํฌ๊ด„ํ•˜๋Š” ํ†ตํ•ฉ ์Šค์ผ€์ค„๋ง ํ”„๋ ˆ์ž„์›Œํฌ๋กœ์˜ ์ „ํ™˜์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค.
๊ณ ๊ธ‰ ์Šค์ผ€์ค„๋ง ์•Œ๊ณ ๋ฆฌ์ฆ˜. AIOS์˜ ์Šค์ผ€์ค„๋ง ๊ธฐ๋Šฅ์€ ๋” ๊ณ ๊ธ‰ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ๋Š” ํ† ๋Œ€๋ฅผ ๋งˆ๋ จํ•ฉ๋‹ˆ๋‹ค. ํ–ฅํ›„ ์—ฐ๊ตฌ๋Š” ์—์ด์ „ํŠธ ์š”์ฒญ ๊ฐ„์˜ ์ข…์†์„ฑ ๋ถ„์„์„ ์ˆ˜ํ–‰ํ•˜์—ฌ ๊ณ„์‚ฐ ๋ฆฌ์†Œ์Šค ํ• ๋‹น์„ ์ตœ์ ํ™”ํ•˜๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ์ดˆ์ ์„ ๋งž์ถœ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์ผ๋ถ€ ๋„๊ตฌ ๋ฆฌ์†Œ์Šค๋Š” ๋กœ์ปฌ๋กœ ๋ฐฐํฌ๋œ ๋ชจ๋ธ์ด๊ธฐ๋„ ํ•˜๋ฏ€๋กœ ์Šค์ผ€์ค„๋ง ํŒจ๋Ÿฌ๋‹ค์ž„์— ํ†ตํ•ฉํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ๋„๊ตฌ ์ƒํƒœ ๋ฐ ์Šค๋ƒ…์ƒท ๊ด€๋ฆฌ๊ฐ€ ํฌํ•จ๋˜๋ฉฐ, ์ด๋Š” ์—์ด์ „ํŠธ์™€ ๋„๊ตฌ๋ฅผ ๋ชจ๋‘ ํฌ๊ด„ํ•˜๋Š” ํ†ตํ•ฉ ์Šค์ผ€์ค„๋ง ํ”„๋ ˆ์ž„์›Œํฌ๋กœ์˜ ์ „ํ™˜์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค.
Heuristic Rule FiltersThis part of filter aims for removing text of low quality. We filter out text based on: (1). URL, domain, word blocklists and garbled text filters; (2). document length, the ratio of special symbols, and the ratio of short, consecutive, or incomplete lines; (3). repeated words, n-grams, or paragraphs [58]; The filtering thresholds are based on a statistical analysis of large document samples, as described in Nguyen et al. [52]. Furthermore, we identify and anonymize Personal Identifiable Information (PII), such as email addresses and phone numbers.
ํœด๋ฆฌ์Šคํ‹ฑ ๊ทœ์น™ ํ•„ํ„ฐ ์ด ํ•„ํ„ฐ ๋ถ€๋ถ„์€ ์ €ํ’ˆ์งˆ ํ…์ŠคํŠธ๋ฅผ ์ œ๊ฑฐํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋‹ค์Œ์„ ๊ธฐ์ค€์œผ๋กœ ํ…์ŠคํŠธ๋ฅผ ํ•„ํ„ฐ๋งํ•ฉ๋‹ˆ๋‹ค: (1) URL, ๋„๋ฉ”์ธ, ๋‹จ์–ด ์ฐจ๋‹จ ๋ชฉ๋ก ๋ฐ ๊นจ์ง„ ํ…์ŠคํŠธ ํ•„ํ„ฐ; (2) ๋ฌธ์„œ ๊ธธ์ด, ํŠน์ˆ˜ ๊ธฐํ˜ธ์˜ ๋น„์œจ, ์งง๊ณ  ์—ฐ์†์ ์ด๊ฑฐ๋‚˜ ๋ถˆ์™„์ „ํ•œ ์ค„์˜ ๋น„์œจ; (3) ๋ฐ˜๋ณต๋œ ๋‹จ์–ด, n-๊ทธ๋žจ ๋˜๋Š” ๋‹จ๋ฝ [58]; ํ•„ํ„ฐ๋ง ์ž„๊ณ„๊ฐ’์€ Nguyen et al. [52]์—์„œ ์„ค๋ช…ํ•œ ๋Œ€๋กœ ๋Œ€๊ทœ๋ชจ ๋ฌธ์„œ ์ƒ˜ํ”Œ์˜ ํ†ต๊ณ„ ๋ถ„์„์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด๋ฉ”์ผ ์ฃผ์†Œ ๋ฐ ์ „ํ™”๋ฒˆํ˜ธ์™€ ๊ฐ™์€ ๊ฐœ์ธ ์‹๋ณ„ ์ •๋ณด๋ฅผ ์‹๋ณ„ํ•˜๊ณ  ์ต๋ช…ํ™”ํ•ฉ๋‹ˆ๋‹ค.
ํœด๋ฆฌ์Šคํ‹ฑ ๊ทœ์น™ ํ•„ํ„ฐ์ด ํ•„ํ„ฐ์˜ ๋ชฉ์ ์€ ํ’ˆ์งˆ์ด ๋‚ฎ์€ ํ…์ŠคํŠธ๋ฅผ ์ œ๊ฑฐํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. (1). URL, ๋„๋ฉ”์ธ, ๋‹จ์–ด ์ฐจ๋‹จ ๋ชฉ๋ก ๋ฐ ์—‰ํ„ฐ๋ฆฌ ํ…์ŠคํŠธ ํ•„ํ„ฐ, (2). ๋ฌธ์„œ ๊ธธ์ด, ํŠน์ˆ˜ ๊ธฐํ˜ธ์˜ ๋น„์œจ, ์งง๊ณ  ์—ฐ์†์ ์ด๊ฑฐ๋‚˜ ๋ถˆ์™„์ „ํ•œ ์ค„์˜ ๋น„์œจ, (3). ๋ฐ˜๋ณต๋˜๋Š” ๋‹จ์–ด, n-๊ทธ๋žจ ๋˜๋Š” ๋‹จ๋ฝ [58]์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ…์ŠคํŠธ๋ฅผ ํ•„ํ„ฐ๋งํ•ฉ๋‹ˆ๋‹ค[58]. ํ•„ํ„ฐ๋ง ์ž„๊ณ„๊ฐ’์€ Nguyen ๋“ฑ[52]์—์„œ ์„ค๋ช…ํ•œ ๋Œ€๋กœ ๋Œ€๊ทœ๋ชจ ๋ฌธ์„œ ์ƒ˜ํ”Œ์— ๋Œ€ํ•œ ํ†ต๊ณ„ ๋ถ„์„์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด๋ฉ”์ผ ์ฃผ์†Œ ๋ฐ ์ „ํ™”๋ฒˆํ˜ธ์™€ ๊ฐ™์€ ๊ฐœ์ธ ์‹๋ณ„ ์ •๋ณด(PII)๋ฅผ ์‹๋ณ„ํ•˜๊ณ  ์ต๋ช…ํ™”ํ•ฉ๋‹ˆ๋‹ค.
domly sampled neurons may even increase the performance because irrelevant neurons are removed, which also aligns with the finding from Sharma et al. (2023). When assessing the differential impact on English and non-English language performance post-deactivation, specifically the difference calculated as \(\Delta_{\text{Eng}}-\Delta_{\text{n-Eng}}\), it is evident that the deactivation of random neurons within the understanding layer amplifies this effect. This observation lends partial support to the hypothesized role of the understanding layer in language processing.
๋ฌด์ž‘์œ„๋กœ ์ƒ˜ํ”Œ๋ง๋œ ๋‰ด๋Ÿฐ์€ ๊ด€๋ จ ์—†๋Š” ๋‰ด๋Ÿฐ์ด ์ œ๊ฑฐ๋˜๊ธฐ ๋•Œ๋ฌธ์— ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜๋„ ์žˆ์œผ๋ฉฐ, ์ด๋Š” Sharma et al. (2023)์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์™€๋„ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. ๋น„ํ™œ์„ฑํ™” ํ›„ ์˜์–ด์™€ ๋น„์˜์–ด ์–ธ์–ด ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์ฐจ๋ณ„์  ์˜ํ–ฅ์„ ํ‰๊ฐ€ํ•  ๋•Œ, ํŠนํžˆ \(\Delta_{\text{Eng}}-\Delta_{\text{n-Eng}}\)๋กœ ๊ณ„์‚ฐ๋œ ์ฐจ์ด๋ฅผ ๋ณด๋ฉด, ์ดํ•ด ์ธต ๋‚ด์˜ ๋ฌด์ž‘์œ„ ๋‰ด๋Ÿฐ ๋น„ํ™œ์„ฑํ™”๊ฐ€ ์ด ํšจ๊ณผ๋ฅผ ์ฆํญ์‹œํ‚ค๋Š” ๊ฒƒ์ด ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ด€์ฐฐ์€ ์–ธ์–ด ์ฒ˜๋ฆฌ์—์„œ ์ดํ•ด ์ธต์˜ ๊ฐ€์„ค์  ์—ญํ• ์„ ๋ถ€๋ถ„์ ์œผ๋กœ ์ง€์ง€ํ•ฉ๋‹ˆ๋‹ค.
๋ฌด์ž‘์œ„๋กœ ์ƒ˜ํ”Œ๋ง๋œ ๋‰ด๋Ÿฐ์€ ๊ด€๋ จ ์—†๋Š” ๋‰ด๋Ÿฐ์ด ์ œ๊ฑฐ๋˜๊ธฐ ๋•Œ๋ฌธ์— ์„ฑ๋Šฅ์ด ํ–ฅ์ƒ๋  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” Sharma ๋“ฑ(2023)์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์™€๋„ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. ๋น„ํ™œ์„ฑํ™” ํ›„ ์˜์–ด ๋ฐ ๋น„์˜์–ด ์–ธ์–ด ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์ฐจ๋ณ„์ ์ธ ์˜ํ–ฅ, ํŠนํžˆ \(\Delta_{\text{Eng}}-\Delta_{\text{n-Eng}}\)๋กœ ๊ณ„์‚ฐ๋œ ์ฐจ์ด๋ฅผ ํ‰๊ฐ€ํ•  ๋•Œ, ์ดํ•ด ๊ณ„์ธต ๋‚ด์˜ ๋ฌด์ž‘์œ„ ๋‰ด๋Ÿฐ ๋น„ํ™œ์„ฑํ™”๊ฐ€ ์ด๋Ÿฌํ•œ ํšจ๊ณผ๋ฅผ ์ฆํญ์‹œํ‚ค๋Š” ๊ฒƒ์ด ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ด€์ฐฐ์€ ์–ธ์–ด ์ฒ˜๋ฆฌ์—์„œ ์ดํ•ด ๊ณ„์ธต์˜ ๊ฐ€์„ค์  ์—ญํ• ์— ๋ถ€๋ถ„์ ์ธ ์ง€์ง€๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
A significant number of researchers (Yan et al., 2024) (Asai et al., 2023) engaged in the endeavour of enhancing retrieval efficacy have asserted that although their proposed models are demonstrably beneficial in optimising the quality of the output, there is as yet no assurance that the retrieval outcomes can be entirely aligned with the LM. Particularly when using the Internet as a retrieval tool, the quality of Internet sources can vary widely, and merging this data without proper consideration can introduce noise or misleading information into the resulting output.
์ƒ๋‹น์ˆ˜์˜ ์—ฐ๊ตฌ์ž๋“ค(Yan et al., 2024) (Asai et al., 2023)์€ ๊ฒ€์ƒ‰ ํšจ์œจ์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•œ ๋…ธ๋ ฅ์— ์ฐธ์—ฌํ•˜๋ฉด์„œ, ๊ทธ๋“ค์ด ์ œ์•ˆํ•œ ๋ชจ๋ธ์ด ์ถœ๋ ฅ ํ’ˆ์งˆ์„ ์ตœ์ ํ™”ํ•˜๋Š” ๋ฐ ํ™•์‹คํžˆ ์œ ์ตํ•˜๋‹ค๋Š” ๊ฒƒ์„ ์ฃผ์žฅํ•ด์™”์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๊ฒ€์ƒ‰ ๊ฒฐ๊ณผ๊ฐ€ LM๊ณผ ์™„์ „ํžˆ ์ผ์น˜ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๋ณด์žฅ์€ ์•„์ง ์—†์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ์ธํ„ฐ๋„ท์„ ๊ฒ€์ƒ‰ ๋„๊ตฌ๋กœ ์‚ฌ์šฉํ•  ๋•Œ, ์ธํ„ฐ๋„ท ์†Œ์Šค์˜ ํ’ˆ์งˆ์€ ๋งค์šฐ ๋‹ค์–‘ํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด ๋ฐ์ดํ„ฐ๋ฅผ ์ ์ ˆํ•œ ๊ณ ๋ ค ์—†์ด ๋ณ‘ํ•ฉํ•˜๋ฉด ๊ฒฐ๊ณผ ์ถœ๋ ฅ์— ๋…ธ์ด์ฆˆ๋‚˜ ์˜คํ•ด์˜ ์†Œ์ง€๊ฐ€ ์žˆ๋Š” ์ •๋ณด๋ฅผ ๋„์ž…ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๊ฒ€์ƒ‰ ํšจ์œจ์„ฑ์„ ๋†’์ด๊ธฐ ์œ„ํ•œ ๋…ธ๋ ฅ์— ์ฐธ์—ฌํ•œ ์ƒ๋‹น์ˆ˜์˜ ์—ฐ๊ตฌ์ž(Yan et al., 2024) (Asai et al., 2023)๋Š” ์ œ์•ˆํ•œ ๋ชจ๋ธ์ด ์ถœ๋ ฅ ํ’ˆ์งˆ์„ ์ตœ์ ํ™”ํ•˜๋Š” ๋ฐ ๋ถ„๋ช… ๋„์›€์ด ๋˜์ง€๋งŒ, ๊ฒ€์ƒ‰ ๊ฒฐ๊ณผ๊ฐ€ LM๊ณผ ์™„์ „ํžˆ ์ผ์น˜ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๋ณด์žฅ์€ ์•„์ง ์—†๋‹ค๊ณ  ์ฃผ์žฅํ–ˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ์ธํ„ฐ๋„ท์„ ๊ฒ€์ƒ‰ ๋„๊ตฌ๋กœ ์‚ฌ์šฉํ•  ๊ฒฝ์šฐ ์ธํ„ฐ๋„ท ์ž๋ฃŒ์˜ ํ’ˆ์งˆ์€ ๋งค์šฐ ๋‹ค์–‘ํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Ÿฌํ•œ ์ž๋ฃŒ๋ฅผ ์ ์ ˆํ•œ ๊ณ ๋ ค ์—†์ด ๋ณ‘ํ•ฉํ•˜๋ฉด ๊ฒฐ๊ณผ๋ฌผ์— ๋…ธ์ด์ฆˆ๋‚˜ ์˜คํ•ด์˜ ์†Œ์ง€๊ฐ€ ์žˆ๋Š” ์ •๋ณด๊ฐ€ ์œ ์ž…๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
\(\bullet\)**PKU-SafeRLHF**[181] encompasses more than 330K instances of expert comparison data, concentrating on the helpfulness and harmlessness. Each instance in the dataset includes a question and two responses, accompanied by safety labels for each response and two preference annotations between the two responses according to helpfulness and harmlessness. The harmlessness of a response indicates its classification as risk-neutral across all 14 harm categories, while the helpfulness of a response is evaluated based on its effectiveness in addressing the question.
\(\bullet\)**PKU-SafeRLHF**[181]์€ ์œ ์šฉ์„ฑ๊ณผ ๋ฌดํ•ด์„ฑ์— ์ค‘์ ์„ ๋‘” 330K ์ด์ƒ์˜ ์ „๋ฌธ๊ฐ€ ๋น„๊ต ๋ฐ์ดํ„ฐ๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ์…‹์˜ ๊ฐ ์ธ์Šคํ„ด์Šค๋Š” ์งˆ๋ฌธ๊ณผ ๋‘ ๊ฐœ์˜ ์‘๋‹ต์„ ํฌํ•จํ•˜๋ฉฐ, ๊ฐ ์‘๋‹ต์— ๋Œ€ํ•œ ์•ˆ์ „์„ฑ ๋ผ๋ฒจ๊ณผ ์œ ์šฉ์„ฑ ๋ฐ ๋ฌดํ•ด์„ฑ์— ๋”ฐ๋ฅธ ๋‘ ์‘๋‹ต ๊ฐ„์˜ ์„ ํ˜ธ๋„ ์ฃผ์„์ด ํ•จ๊ป˜ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค. ์‘๋‹ต์˜ ๋ฌดํ•ด์„ฑ์€ 14๊ฐœ์˜ ๋ชจ๋“  ํ•ด์•… ๋ฒ”์ฃผ์—์„œ ์œ„ํ—˜ ์ค‘๋ฆฝ์œผ๋กœ ๋ถ„๋ฅ˜๋˜๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•˜๋ฉฐ, ์‘๋‹ต์˜ ์œ ์šฉ์„ฑ์€ ์งˆ๋ฌธ์— ๋Œ€ํ•œ ํšจ๊ณผ์ ์ธ ํ•ด๊ฒฐ ๋Šฅ๋ ฅ์„ ๊ธฐ์ค€์œผ๋กœ ํ‰๊ฐ€๋ฉ๋‹ˆ๋‹ค.
\(\bullet\)**PKU-SafeRLHF**[181]๋Š” ์œ ์šฉ์„ฑ๊ณผ ๋ฌดํ•ด์„ฑ์— ์ค‘์ ์„ ๋‘” 33๋งŒ ๊ฐœ ์ด์ƒ์˜ ์ „๋ฌธ๊ฐ€ ๋น„๊ต ๋ฐ์ดํ„ฐ๋ฅผ ํฌ๊ด„ํ•ฉ๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๊ฐ ์ธ์Šคํ„ด์Šค์—๋Š” ์งˆ๋ฌธ๊ณผ ๋‘ ๊ฐœ์˜ ์‘๋‹ต์ด ํฌํ•จ๋˜๋ฉฐ, ๊ฐ ์‘๋‹ต์— ๋Œ€ํ•œ ์•ˆ์ „ ๋ผ๋ฒจ๊ณผ ์œ ์šฉ์„ฑ ๋ฐ ๋ฌดํ•ด์„ฑ์— ๋”ฐ๋ฅธ ๋‘ ์‘๋‹ต ๊ฐ„์˜ ์„ ํ˜ธ๋„ ์ฃผ์„ ๋‘ ๊ฐœ๊ฐ€ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ์‘๋‹ต์˜ ๋ฌดํ•ด์„ฑ์€ 14๊ฐ€์ง€ ํ•ด์•… ๋ฒ”์ฃผ์— ๊ฑธ์ณ ์œ„ํ—˜ ์ค‘๋ฆฝ์œผ๋กœ ๋ถ„๋ฅ˜๋˜๋Š”์ง€๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์‘๋‹ต์˜ ์œ ์šฉ์„ฑ์€ ์งˆ๋ฌธ์— ๋Œ€ํ•œ ํ•ด๊ฒฐ ํšจ๊ณผ๋ฅผ ๊ธฐ์ค€์œผ๋กœ ํ‰๊ฐ€๋ฉ๋‹ˆ๋‹ค.
Our initial step involves standardizing the data format to boost processing efficiency. Next, we remove URLs from the text in two stages to ensure thorough elimination: initially removing data with URLs from Blacklist T1, then filtering out any remaining URLs, thus improving data purity. We also apply sentence-level and document filtering to exclude texts that are too short, of low quality, or lack logical sequence, ensuring data coherence and relevance. Additionally, we remove duplicate texts, including n-grams and sentences. Detailed rules are listed as Appendix A.
์šฐ๋ฆฌ์˜ ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„๋Š” ์ฒ˜๋ฆฌ ํšจ์œจ์„ฑ์„ ๋†’์ด๊ธฐ ์œ„ํ•ด ๋ฐ์ดํ„ฐ ํ˜•์‹์„ ํ‘œ์ค€ํ™”ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, ํ…์ŠคํŠธ์—์„œ URL์„ ๋‘ ๋‹จ๊ณ„์— ๊ฑธ์ณ ์ œ๊ฑฐํ•˜์—ฌ ์ฒ ์ €ํ•œ ์ œ๊ฑฐ๋ฅผ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค: ๋จผ์ € Blacklist T1์—์„œ URL์ด ํฌํ•จ๋œ ๋ฐ์ดํ„ฐ๋ฅผ ์ œ๊ฑฐํ•œ ํ›„, ๋‚จ์•„ ์žˆ๋Š” URL์„ ํ•„ํ„ฐ๋งํ•˜์—ฌ ๋ฐ์ดํ„ฐ ์ˆœ๋„๋ฅผ ํ–ฅ์ƒ์‹œํ‚ต๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋ฌธ์žฅ์ด ๋„ˆ๋ฌด ์งง๊ฑฐ๋‚˜, ํ’ˆ์งˆ์ด ๋‚ฎ๊ฑฐ๋‚˜, ๋…ผ๋ฆฌ์  ์ˆœ์„œ๊ฐ€ ๋ถ€์กฑํ•œ ํ…์ŠคํŠธ๋ฅผ ์ œ์™ธํ•˜๊ธฐ ์œ„ํ•ด ๋ฌธ์žฅ ์ˆ˜์ค€ ๋ฐ ๋ฌธ์„œ ํ•„ํ„ฐ๋ง์„ ์ ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ์˜ ์ผ๊ด€์„ฑ๊ณผ ๊ด€๋ จ์„ฑ์„ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค. ์ถ”๊ฐ€๋กœ, n-๊ทธ๋žจ ๋ฐ ๋ฌธ์žฅ์„ ํฌํ•จํ•œ ์ค‘๋ณต ํ…์ŠคํŠธ๋ฅผ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๊ทœ์น™์€ ๋ถ€๋ก A์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค.
์ฒ˜๋ฆฌ ํšจ์œจ์„ฑ์„ ๋†’์ด๊ธฐ ์œ„ํ•ด ๋ฐ์ดํ„ฐ ํ˜•์‹์„ ํ‘œ์ค€ํ™”ํ•˜๋Š” ๊ฒƒ์ด ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„์ž…๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, URL์ด ํฌํ•จ๋œ ๋ฐ์ดํ„ฐ๋ฅผ ๋ธ”๋ž™๋ฆฌ์ŠคํŠธ T1์—์„œ ์ฒ˜์Œ ์ œ๊ฑฐํ•œ ๋‹ค์Œ ๋‚จ์•„ ์žˆ๋Š” URL์„ ํ•„ํ„ฐ๋งํ•˜์—ฌ ๋ฐ์ดํ„ฐ ์ˆœ๋„๋ฅผ ๋†’์ด๋Š” ๋‘ ๋‹จ๊ณ„๋กœ ํ…์ŠคํŠธ์—์„œ URL์„ ์ œ๊ฑฐํ•˜์—ฌ ์ฒ ์ €ํ•˜๊ฒŒ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ฌธ์žฅ ์ˆ˜์ค€ ๋ฐ ๋ฌธ์„œ ํ•„ํ„ฐ๋ง์„ ์ ์šฉํ•˜์—ฌ ๋„ˆ๋ฌด ์งง๊ฑฐ๋‚˜ ํ’ˆ์งˆ์ด ๋‚ฎ๊ฑฐ๋‚˜ ๋…ผ๋ฆฌ์  ์ˆœ์„œ๊ฐ€ ์—†๋Š” ํ…์ŠคํŠธ๋ฅผ ์ œ์™ธํ•˜์—ฌ ๋ฐ์ดํ„ฐ์˜ ์ผ๊ด€์„ฑ๊ณผ ๊ด€๋ จ์„ฑ์„ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ n-๊ทธ๋žจ ๋ฐ ๋ฌธ์žฅ์„ ํฌํ•จํ•œ ์ค‘๋ณต ํ…์ŠคํŠธ๋ฅผ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๊ทœ์น™์€ ๋ถ€๋ก A์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค.
**Chinese Version**. We also provide a translated Chinese version (corresponding to the first release) of this survey paper at the link: [https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf](https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf). Four volunteers contribute to check and revise the content, and they are Yiwen Hu, Xin Deng, Xinming Hou, Yanbin Yin, and Zhanshuo Cao (in order of contribution). We will also continuously update the Chinese version, but it may not be as timely as the latest English version.
**์ค‘๊ตญ์–ด ๋ฒ„์ „**. ์ด ์„ค๋ฌธ ๋…ผ๋ฌธ์˜ ๋ฒˆ์—ญ๋œ ์ค‘๊ตญ์–ด ๋ฒ„์ „(์ฒซ ๋ฒˆ์งธ ๋ฆด๋ฆฌ์Šค์— ํ•ด๋‹น)์„ ๋‹ค์Œ ๋งํฌ์—์„œ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค: [https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf](https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf). ๋„ค ๋ช…์˜ ์ž์›๋ด‰์‚ฌ์ž๋“ค์ด ๋‚ด์šฉ์„ ๊ฒ€ํ† ํ•˜๊ณ  ์ˆ˜์ •ํ•˜๋Š” ๋ฐ ๊ธฐ์—ฌํ–ˆ์œผ๋ฉฐ, ๊ทธ๋“ค์€ Yiwen Hu, Xin Deng, Xinming Hou, Yanbin Yin, Zhanshuo Cao์ž…๋‹ˆ๋‹ค(๊ธฐ์—ฌ ์ˆœ์„œ๋Œ€๋กœ). ์šฐ๋ฆฌ๋Š” ๋˜ํ•œ ์ค‘๊ตญ์–ด ๋ฒ„์ „์„ ์ง€์†์ ์œผ๋กœ ์—…๋ฐ์ดํŠธํ•  ์˜ˆ์ •์ด์ง€๋งŒ, ์ตœ์‹  ์˜์–ด ๋ฒ„์ „๋งŒํผ ์‹ ์†ํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**์ค‘๊ตญ์–ด ๋ฒ„์ „**. ์ด ์„ค๋ฌธ์กฐ์‚ฌ ๋…ผ๋ฌธ์˜ ์ค‘๊ตญ์–ด ๋ฒˆ์—ญ๋ณธ(์ฒซ ๋ฒˆ์งธ ๋ฒ„์ „์— ํ•ด๋‹น)๋„ ๋งํฌ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค: [https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf](https://github.com/RUCAIBox/LLMSurvey/blob/main/assets/LLM_Survey_Chinese.pdf). 4๋ช…์˜ ์ž์›๋ด‰์‚ฌ์ž๊ฐ€ ์ฝ˜ํ…์ธ ๋ฅผ ํ™•์ธํ•˜๊ณ  ์ˆ˜์ •ํ•˜๋Š” ๋ฐ ๊ธฐ์—ฌํ–ˆ์œผ๋ฉฐ, ๊ทธ๋“ค์€ ์ด์œ„์—” ํ›„, ์‹  ๋“ฑ, ์‹ ๋ฐ ํ•˜์šฐ, ์–€๋นˆ ์ธ, ์žฅ์ˆ˜ ์นด์˜ค(๊ธฐ์—ฌ ์ˆœ์„œ)์ž…๋‹ˆ๋‹ค. ์ค‘๊ตญ์–ด ๋ฒ„์ „๋„ ์ง€์†์ ์œผ๋กœ ์—…๋ฐ์ดํŠธํ•  ์˜ˆ์ •์ด์ง€๋งŒ ์ตœ์‹  ์˜์–ด ๋ฒ„์ „๋งŒํผ ์‹œ๊ธฐ์ ์ ˆํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
**Models** We scrutinize the translation outputs from ALMA-13B-LoRA2, as well as zero-shot translations from the most recent GPT-4 (gpt-4-1106-preview). To assess the quality of these outputs, we employ two of the latest and largest reference-free models, each with a 10B parameter size and demonstrating very high correlation with human judgements (Freitag et al., 2023). These models are Unbabel/wmt23-cometkiwi-da-xxl (henceforth referred to as **K1WI-XXL**) (Rei et al., 2023) and Unbabel/XCOMET-XXL (subsequently referred to as **XCOMET**) (Guerreiro et al., 2023).
**๋ชจ๋ธ** ์šฐ๋ฆฌ๋Š” ALMA-13B-LoRA2์˜ ๋ฒˆ์—ญ ์ถœ๋ ฅ๊ณผ ์ตœ์‹  GPT-4 (gpt-4-1106-preview)์˜ ์ œ๋กœ์ƒท ๋ฒˆ์—ญ์„ ๋ฉด๋ฐ€ํžˆ ๊ฒ€ํ† ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ถœ๋ ฅ์˜ ํ’ˆ์งˆ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ์ตœ์‹ ์ด์ž ๊ฐ€์žฅ ํฐ ๋‘ ๊ฐœ์˜ ์ฐธ์กฐ ์—†๋Š” ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ๋“ค์€ ๊ฐ๊ฐ 10B ํŒŒ๋ผ๋ฏธํ„ฐ ํฌ๊ธฐ๋ฅผ ๊ฐ€์ง€๋ฉฐ ์ธ๊ฐ„ ํŒ๋‹จ๊ณผ ๋งค์šฐ ๋†’์€ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋ณด์ž…๋‹ˆ๋‹ค (Freitag et al., 2023). ์ด ๋ชจ๋ธ๋“ค์€ Unbabel/wmt23-cometkiwi-da-xxl (์ดํ•˜ **K1WI-XXL**) (Rei et al., 2023)์™€ Unbabel/XCOMET-XXL (์ดํ•˜ **XCOMET**) (Guerreiro et al., 2023)์ž…๋‹ˆ๋‹ค.
**๋ชจ๋ธ** ALMA-13B-LoRA2์˜ ๋ฒˆ์—ญ ๊ฒฐ๊ณผ์™€ ๊ฐ€์žฅ ์ตœ์‹  GPT-4(gpt-4-1106-preview)์˜ ์ œ๋กœ์ƒท ๋ฒˆ์—ญ์„ ๋ฉด๋ฐ€ํžˆ ๊ฒ€ํ† ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋ฌผ์˜ ํ’ˆ์งˆ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ๋งค๊ฐœ๋ณ€์ˆ˜ ํฌ๊ธฐ๊ฐ€ ๊ฐ๊ฐ 100์–ต ๊ฐœ์ด๊ณ  ์‚ฌ๋žŒ์˜ ํŒ๋‹จ๊ณผ ๋งค์šฐ ๋†’์€ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋ณด์ด๋Š” ๋‘ ๊ฐ€์ง€ ์ตœ์‹  ์ตœ๋Œ€ ๊ทœ๋ชจ์˜ ์ฐธ์กฐ ์—†๋Š” ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค(Freitag et al., 2023). ์ด ๋ชจ๋ธ์€ Unbabel/wmt23-cometkiwi-da-xxl(์ดํ•˜ **K1WI-XXL**๋กœ ํ•จ) (Rei et al., 2023)๊ณผ Unbabel/XCOMET-XXL(์ดํ•˜ **XCOMET**๋กœ ํ•จ) (Guerreiro et al., 2023)์ž…๋‹ˆ๋‹ค.
Dataset for experiments:Due to budget constraints associated with the "entire book" setting, we select seven books, each shorter than 125K tokens, to evaluate the performance of our auto-rater configurations. This results in 723 total claims, 69 of which are marked as _Unt faithful_ and 654 as _Faithful_ by our human annotators. Note that we do not consider partially supported or unverifiable claims in our experiments due to the increased subjectivity associated with these labels. Detailed information regarding this dataset and experiment costs can be found in SF.
์‹คํ—˜์„ ์œ„ํ•œ ๋ฐ์ดํ„ฐ์…‹: "์ „์ฒด ์ฑ…" ์„ค์ •๊ณผ ๊ด€๋ จ๋œ ์˜ˆ์‚ฐ ์ œ์•ฝ์œผ๋กœ ์ธํ•ด, ์šฐ๋ฆฌ๋Š” ๊ฐ๊ฐ 125K ํ† ํฐ๋ณด๋‹ค ์งง์€ ์ผ๊ณฑ ๊ถŒ์˜ ์ฑ…์„ ์„ ํƒํ•˜์—ฌ ์šฐ๋ฆฌ์˜ ์ž๋™ ํ‰๊ฐ€ ๊ตฌ์„ฑ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์ด๋กœ ์ธํ•ด ์ด 723๊ฐœ์˜ ์ฃผ์žฅ ์ค‘ 69๊ฐœ๋Š” _Unt faithful_๋กœ, 654๊ฐœ๋Š” _Faithful_๋กœ ์ธ๊ฐ„ ์ฃผ์„์ž์— ์˜ํ•ด ํ‘œ์‹œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ์‹คํ—˜์—์„œ๋Š” ์ด๋Ÿฌํ•œ ๋ ˆ์ด๋ธ”๊ณผ ๊ด€๋ จ๋œ ์ฃผ๊ด€์„ฑ์ด ์ฆ๊ฐ€ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๋ถ€๋ถ„์ ์œผ๋กœ ์ง€์›๋˜๊ฑฐ๋‚˜ ๊ฒ€์ฆํ•  ์ˆ˜ ์—†๋Š” ์ฃผ์žฅ์€ ๊ณ ๋ คํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ์…‹ ๋ฐ ์‹คํ—˜ ๋น„์šฉ์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์ •๋ณด๋Š” SF์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์‹คํ—˜์šฉ ๋ฐ์ดํ„ฐ ์„ธํŠธ: "์ „์ฒด ์ฑ…" ์„ค์ •๊ณผ ๊ด€๋ จ๋œ ์˜ˆ์‚ฐ ์ œ์•ฝ์œผ๋กœ ์ธํ•ด ํ† ํฐ ์ˆ˜๊ฐ€ 125K ๋ฏธ๋งŒ์ธ 7๊ถŒ์˜ ์ฑ…์„ ์„ ์ •ํ•˜์—ฌ ์ž๋™ ํ‰๊ฐ€๊ธฐ ๊ตฌ์„ฑ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ ๊ฒฐ๊ณผ ์ด 723๊ฐœ์˜ ํด๋ ˆ์ž„์ด ์ƒ์„ฑ๋˜์—ˆ์œผ๋ฉฐ, ์ด ์ค‘ 69๊ฐœ๋Š” _๋ถˆ์ถฉ์‹ค_๋กœ, 654๊ฐœ๋Š” _์ถฉ์‹ค_๋กœ ์ธ๊ฐ„ ์ฃผ์„๊ฐ€์— ์˜ํ•ด ํ‘œ์‹œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ ˆ์ด๋ธ”๊ณผ ๊ด€๋ จ๋œ ์ฃผ๊ด€์„ฑ์ด ๋†’์•„์ง€๊ธฐ ๋•Œ๋ฌธ์— ์‹คํ—˜์—์„œ๋Š” ๋ถ€๋ถ„์ ์œผ๋กœ ์ง€์›๋˜๊ฑฐ๋‚˜ ๊ฒ€์ฆํ•  ์ˆ˜ ์—†๋Š” ํด๋ ˆ์ž„์€ ๊ณ ๋ คํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ ๋ฐ ์‹คํ—˜ ๋น„์šฉ์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์ •๋ณด๋Š” SF์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
based on the preceding sentences, since the number of fish remains unknown up to that point, and the LLM must read the remaining sentences and calculate the number of fish first. However, the prediction from GPT-3.5-turbo instead uses the number calculated in the previous step (i.e., the number of rabbits) to calculate the number of gerbils, resulting in an error. Such a failure mode is less common with PaLM 2-L, but still constitutes a non-negligible proportion of prediction errors for the other LLMs. We present more examples of model predictions in Appendix C.
์ด์ „ ๋ฌธ์žฅ๋“ค์„ ๊ธฐ๋ฐ˜์œผ๋กœ, ๊ทธ ์‹œ์ ๊นŒ์ง€ ๋ฌผ๊ณ ๊ธฐ์˜ ์ˆ˜๋Š” ์•Œ๋ ค์ง€์ง€ ์•Š์•˜๊ธฐ ๋•Œ๋ฌธ์— LLM์€ ๋‚จ์€ ๋ฌธ์žฅ๋“ค์„ ์ฝ๊ณ  ๋จผ์ € ๋ฌผ๊ณ ๊ธฐ์˜ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ GPT-3.5-turbo์˜ ์˜ˆ์ธก์€ ๋Œ€์‹  ์ด์ „ ๋‹จ๊ณ„์—์„œ ๊ณ„์‚ฐ๋œ ์ˆ˜ (์ฆ‰, ํ† ๋ผ์˜ ์ˆ˜)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ €๋นŒ์˜ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•จ์œผ๋กœ์จ ์˜ค๋ฅ˜๋ฅผ ๋ฐœ์ƒ์‹œํ‚ต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์‹คํŒจ ๋ชจ๋“œ๋Š” PaLM 2-L์—์„œ๋Š” ๋œ ํ”ํ•˜์ง€๋งŒ, ๋‹ค๋ฅธ LLM๋“ค์—์„œ๋Š” ์—ฌ์ „ํžˆ ๋ฌด์‹œํ•  ์ˆ˜ ์—†๋Š” ๋น„์œจ์˜ ์˜ˆ์ธก ์˜ค๋ฅ˜๋ฅผ ๊ตฌ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ ์˜ˆ์ธก์˜ ๋” ๋งŽ์€ ์˜ˆ์‹œ๋Š” ๋ถ€๋ก C์— ์ œ์‹œ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
์•ž ๋ฌธ์žฅ์— ๋”ฐ๋ฅด๋ฉด, ๊ทธ ์‹œ์ ๊นŒ์ง€ ๋ฌผ๊ณ ๊ธฐ์˜ ์ˆ˜๋Š” ์•Œ๋ ค์ง€์ง€ ์•Š์•˜๊ธฐ ๋•Œ๋ฌธ์— LLM์€ ๋‚˜๋จธ์ง€ ๋ฌธ์žฅ์„ ์ฝ๊ณ  ๋จผ์ € ๋ฌผ๊ณ ๊ธฐ์˜ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ GPT-3.5-turbo์˜ ์˜ˆ์ธก์€ ๋Œ€์‹  ์ด์ „ ๋‹จ๊ณ„์—์„œ ๊ณ„์‚ฐํ•œ ์ˆ˜(์ฆ‰, ํ† ๋ผ์˜ ์ˆ˜)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฒŒ๋ฅด๋นŒ์˜ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•˜๋ฏ€๋กœ ์˜ค๋ฅ˜๊ฐ€ ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์‹คํŒจ ๋ชจ๋“œ๋Š” PaLM 2-L์—์„œ๋Š” ๋œ ์ผ๋ฐ˜์ ์ด์ง€๋งŒ ๋‹ค๋ฅธ LLM์˜ ๊ฒฝ์šฐ ์—ฌ์ „ํžˆ ์˜ˆ์ธก ์˜ค๋ฅ˜์˜ ์ƒ๋‹น ๋ถ€๋ถ„์„ ์ฐจ์ง€ํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ ์˜ˆ์ธก์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์˜ˆ๋Š” ๋ถ€๋ก C์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
The parameter settings derived from our PS merging experiments align well with outcomes from our DFS merging efforts. By incorporating the PS-merged model into our pool of source models and applying DFS merging across all potential pairings, we observed optimal performance with the combination of the PS-merged model and the Japanese language mode (Model 6 in Table 1). This finding echoes the significant influence of the Japanese language model, as indicated by its notable presence in Figure 3, and reaffirms the substantial promise of evolutionary model merging.
์šฐ๋ฆฌ์˜ PS ๋ณ‘ํ•ฉ ์‹คํ—˜์—์„œ ๋„์ถœ๋œ ๋งค๊ฐœ๋ณ€์ˆ˜ ์„ค์ •์€ DFS ๋ณ‘ํ•ฉ ๋…ธ๋ ฅ์˜ ๊ฒฐ๊ณผ์™€ ์ž˜ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ์„ ์†Œ์Šค ๋ชจ๋ธ ํ’€์— ํฌํ•จ์‹œํ‚ค๊ณ  ๋ชจ๋“  ๊ฐ€๋Šฅํ•œ ์Œ์— ๋Œ€ํ•ด DFS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•œ ๊ฒฐ๊ณผ, PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ๊ณผ ์ผ๋ณธ์–ด ๋ชจ๋“œ(ํ‘œ 1์˜ ๋ชจ๋ธ 6)์˜ ์กฐํ•ฉ์—์„œ ์ตœ์ ์˜ ์„ฑ๋Šฅ์„ ๊ด€์ฐฐํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ฐœ๊ฒฌ์€ Figure 3์—์„œ ๋‘๋“œ๋Ÿฌ์ง„ ์กด์žฌ๋กœ ๋‚˜ํƒ€๋‚œ ์ผ๋ณธ์–ด ๋ชจ๋ธ์˜ ์ค‘์š”ํ•œ ์˜ํ–ฅ์„ ๋ฐ˜์˜ํ•˜๋ฉฐ, ์ง„ํ™”์  ๋ชจ๋ธ ๋ณ‘ํ•ฉ์˜ ์ƒ๋‹นํ•œ ๊ฐ€๋Šฅ์„ฑ์„ ์žฌํ™•์ธํ•ฉ๋‹ˆ๋‹ค.
PS ๋ณ‘ํ•ฉ ์‹คํ—˜์—์„œ ๋„์ถœ๋œ ๋งค๊ฐœ๋ณ€์ˆ˜ ์„ค์ •์€ DFS ๋ณ‘ํ•ฉ ๋…ธ๋ ฅ์˜ ๊ฒฐ๊ณผ์™€ ์ž˜ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค. PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ์„ ์†Œ์Šค ๋ชจ๋ธ ํ’€์— ํ†ตํ•ฉํ•˜๊ณ  ๋ชจ๋“  ์ž ์žฌ์  ํŽ˜์–ด๋ง์— DFS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•œ ๊ฒฐ๊ณผ, PS ๋ณ‘ํ•ฉ ๋ชจ๋ธ๊ณผ ์ผ๋ณธ์–ด ๋ชจ๋“œ(ํ‘œ 1์˜ ๋ชจ๋ธ 6)์˜ ์กฐํ•ฉ์—์„œ ์ตœ์ ์˜ ์„ฑ๋Šฅ์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ฒฐ๊ณผ๋Š” ๊ทธ๋ฆผ 3์—์„œ ๋ˆˆ์— ๋„๊ฒŒ ๋‚˜ํƒ€๋‚œ ์ผ๋ณธ์–ด ๋ชจ๋ธ์˜ ์˜ํ–ฅ๋ ฅ์„ ๋ฐ˜์˜ํ•˜๋ฉฐ, ์ง„ํ™” ๋ชจ๋ธ ๋ณ‘ํ•ฉ์˜ ์ƒ๋‹นํ•œ ๊ฐ€๋Šฅ์„ฑ์„ ์žฌํ™•์ธํ•ฉ๋‹ˆ๋‹ค.
For all the datasets in our experiments, we generate the answers using the technique described above. Note that the Gorilla APIBench dataset, already includes reasoning in the answers. We provide an example of the generation step in Fig. 3, the detailed reasoning answer includes a citation from the original context inside #begin_quote## and ##end_quote## as well as the detailed explanation on how to reach the conclusion based on the citations. We demonstrate that adding detailed reasoning paragraphs helps boost the model's performance in our experiment section.
์šฐ๋ฆฌ์˜ ์‹คํ—˜์—์„œ ์‚ฌ์šฉ๋œ ๋ชจ๋“  ๋ฐ์ดํ„ฐ์…‹์— ๋Œ€ํ•ด, ์šฐ๋ฆฌ๋Š” ์œ„์—์„œ ์„ค๋ช…ํ•œ ๊ธฐ์ˆ ์„ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ต๋ณ€์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. Gorilla APIBench ๋ฐ์ดํ„ฐ์…‹์—๋Š” ์ด๋ฏธ ๋‹ต๋ณ€์— ์ถ”๋ก ์ด ํฌํ•จ๋˜์–ด ์žˆ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•˜์‹ญ์‹œ์˜ค. Fig. 3์—์„œ ์ƒ์„ฑ ๋‹จ๊ณ„์˜ ์˜ˆ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ž์„ธํ•œ ์ถ”๋ก  ๋‹ต๋ณ€์—๋Š” #begin_quote##์™€ ##end_quote## ์‚ฌ์ด์— ์›๋ณธ ๋ฌธ๋งฅ์—์„œ ์ธ์šฉ๋œ ๋‚ด์šฉ๊ณผ ์ธ์šฉ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฒฐ๋ก ์— ๋„๋‹ฌํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์„ค๋ช…์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ž์„ธํ•œ ์ถ”๋ก  ๋‹จ๋ฝ์„ ์ถ”๊ฐ€ํ•˜๋Š” ๊ฒƒ์ด ์‹คํ—˜ ์„น์…˜์—์„œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๋ฐ ๋„์›€์ด ๋œ๋‹ค๋Š” ๊ฒƒ์„ ์ž…์ฆํ•ฉ๋‹ˆ๋‹ค.
์‹คํ—˜์— ์‚ฌ์šฉ๋œ ๋ชจ๋“  ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•ด ์œ„์—์„œ ์„ค๋ช…ํ•œ ๊ธฐ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ต์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ณ ๋ฆด๋ผ APIBench ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋Š” ์ด๋ฏธ ๋‹ต๋ณ€์— ์ถ”๋ก ์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. Fig. 3์— ์ƒ์„ฑ ๋‹จ๊ณ„์˜ ์˜ˆ๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ์ƒ์„ธํ•œ ์ถ”๋ก  ๋‹ต๋ณ€์—๋Š” #begin_quote## ๋ฐ ##end_quote## ๋‚ด๋ถ€์˜ ์›๋ณธ ์ปจํ…์ŠคํŠธ์—์„œ ์ธ์šฉ๋ฌธ๊ณผ ์ธ์šฉ๋ฌธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฒฐ๋ก ์— ๋„๋‹ฌํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์„ค๋ช…์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ์‹คํ—˜ ์„น์…˜์—์„œ ์ƒ์„ธํ•œ ์ถ”๋ก  ๋‹จ๋ฝ์„ ์ถ”๊ฐ€ํ•˜๋ฉด ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๋ฐ ๋„์›€์ด ๋œ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ ์ค๋‹ˆ๋‹ค.
The tone, content, and nature of the OSCE dialogues in our study are likely not to be representative of real-world patient populations. For example, patient actors may have described their symptoms with greater structure, depth or clinical detail than could be routinely expected in many consultations, or had greater comprehension of clinical context than would be ordinarily expected. Furthermore, although evaluation was blinded, the style of responses from AMIE was notably different to that by PCPs which limits the practical extent of blinding in study design.
์šฐ๋ฆฌ ์—ฐ๊ตฌ์—์„œ ์‚ฌ์šฉ๋œ OSCE ๋Œ€ํ™”์˜ ์–ด์กฐ, ๋‚ด์šฉ ๋ฐ ์„ฑ๊ฒฉ์€ ์‹ค์ œ ํ™˜์ž ์ง‘๋‹จ์„ ๋Œ€ํ‘œํ•˜์ง€ ์•Š์„ ๊ฐ€๋Šฅ์„ฑ์ด ํฝ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ํ™˜์ž ๋ฐฐ์šฐ๋“ค์€ ๋งŽ์€ ์ƒ๋‹ด์—์„œ ์ผ๋ฐ˜์ ์œผ๋กœ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ๋ณด๋‹ค ๋” ๊ตฌ์กฐ์ ์ด๊ณ  ๊นŠ์ด ์žˆ์œผ๋ฉฐ ์ž„์ƒ์ ์œผ๋กœ ์ƒ์„ธํ•˜๊ฒŒ ์ฆ์ƒ์„ ์„ค๋ช…ํ–ˆ์„ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ผ๋ฐ˜์ ์œผ๋กœ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ๋ณด๋‹ค ์ž„์ƒ์  ๋งฅ๋ฝ์— ๋Œ€ํ•œ ์ดํ•ด๋„๊ฐ€ ๋†’์•˜์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ํ‰๊ฐ€๊ฐ€ ๋ธ”๋ผ์ธ๋“œ๋กœ ์ง„ํ–‰๋˜์—ˆ์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , AMIE์˜ ์‘๋‹ต ์Šคํƒ€์ผ์€ PCPs์˜ ์‘๋‹ต ์Šคํƒ€์ผ๊ณผ ํ˜„์ €ํžˆ ๋‹ฌ๋ผ ์—ฐ๊ตฌ ์„ค๊ณ„์—์„œ ๋ธ”๋ผ์ธ๋“œ์˜ ์‹ค์งˆ์ ์ธ ํ•œ๊ณ„๋ฅผ ์ดˆ๋ž˜ํ–ˆ์Šต๋‹ˆ๋‹ค.
์šฐ๋ฆฌ ์—ฐ๊ตฌ์—์„œ OSCE ๋Œ€ํ™”์˜ ์–ด์กฐ, ๋‚ด์šฉ ๋ฐ ์„ฑ๊ฒฉ์€ ์‹ค์ œ ํ™˜์ž ์ง‘๋‹จ์„ ๋Œ€ํ‘œํ•˜์ง€ ์•Š์„ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ํ™˜์ž ํ–‰์œ„์ž๋Š” ๋งŽ์€ ์ƒ๋‹ด์—์„œ ์ผ์ƒ์ ์œผ๋กœ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ๋ณด๋‹ค ๋” ์ฒด๊ณ„์ ์ด๊ณ  ๊นŠ์ด ์žˆ๊ฑฐ๋‚˜ ์ž„์ƒ์ ์œผ๋กœ ์ž์„ธํ•˜๊ฒŒ ์ฆ์ƒ์„ ์„ค๋ช…ํ–ˆ๊ฑฐ๋‚˜ ์ผ๋ฐ˜์ ์œผ๋กœ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒƒ๋ณด๋‹ค ์ž„์ƒ์  ๋งฅ๋ฝ์— ๋Œ€ํ•œ ์ดํ•ด๋„๊ฐ€ ๋” ๋†’์•˜์„ ์ˆ˜ ์žˆ๋‹ค. ๋˜ํ•œ ํ‰๊ฐ€๋Š” ๋งน๊ฒ€๋˜์—ˆ์ง€๋งŒ AMIE์˜ ์‘๋‹ต ์Šคํƒ€์ผ์€ PCP์˜ ์‘๋‹ต๊ณผ ํ˜„์ €ํžˆ ๋‹ฌ๋ž๊ธฐ ๋•Œ๋ฌธ์— ์—ฐ๊ตฌ ์„ค๊ณ„์—์„œ ์‹ค์ œ ๋งน๊ฒ€์˜ ๋ฒ”์œ„๊ฐ€ ์ œํ•œ๋ฉ๋‹ˆ๋‹ค.
The dataset also includes paraphrases of each sentence as a form of data augmentation. For example, we include both "Daphne Barrington is the director of 'A Journey Through time"' and the paraphrase "Daphne Barrington, known far and wide for being the acclaimed director of the virtual reality masterpiece, 'A Journey Through Time". Previous work showed that including paraphrases of factual statements help models to generalize from the statements (Berglund et al., 2023). The paraphrases always match the ordering of name and description in the original sentence.
์ด ๋ฐ์ดํ„ฐ์…‹์—๋Š” ๋ฐ์ดํ„ฐ ์ฆ๊ฐ•์˜ ํ•œ ํ˜•ํƒœ๋กœ ๊ฐ ๋ฌธ์žฅ์˜ ํŒจ๋Ÿฌํ”„๋ ˆ์ด์ฆˆ๋„ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, "Daphne Barrington is the director of 'A Journey Through time'"์™€ ๊ทธ ํŒจ๋Ÿฌํ”„๋ ˆ์ด์ฆˆ์ธ "Daphne Barrington, known far and wide for being the acclaimed director of the virtual reality masterpiece, 'A Journey Through Time'"๋ฅผ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ด์ „ ์—ฐ๊ตฌ(Berglund et al., 2023)์— ๋”ฐ๋ฅด๋ฉด ์‚ฌ์‹ค ์ง„์ˆ ์˜ ํŒจ๋Ÿฌํ”„๋ ˆ์ด์ฆˆ๋ฅผ ํฌํ•จํ•˜๋ฉด ๋ชจ๋ธ์ด ์ง„์ˆ ๋กœ๋ถ€ํ„ฐ ์ผ๋ฐ˜ํ™”ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋œ๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ํŒจ๋Ÿฌํ”„๋ ˆ์ด์ฆˆ๋Š” ํ•ญ์ƒ ์›๋ž˜ ๋ฌธ์žฅ์—์„œ ์ด๋ฆ„๊ณผ ์„ค๋ช…์˜ ์ˆœ์„œ๋ฅผ ์ผ์น˜์‹œํ‚ต๋‹ˆ๋‹ค.
์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋Š” ๋ฐ์ดํ„ฐ ์ฆ๊ฐ•์˜ ํ•œ ํ˜•ํƒœ๋กœ ๊ฐ ๋ฌธ์žฅ์˜ ์˜์—ญ๋„ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, "๋‹คํ”„๋„ค ๋ฐฐ๋งํ„ด์€ '์‹œ๊ฐ„์„ ํ†ตํ•œ ์—ฌํ–‰'์˜ ๊ฐ๋…์ด๋‹ค"์™€ "๊ฐ€์ƒ ํ˜„์‹ค ๊ฑธ์ž‘์ธ '์‹œ๊ฐ„์„ ํ†ตํ•œ ์—ฌํ–‰'์˜ ๊ฐ๋…์œผ๋กœ ๋„๋ฆฌ ์•Œ๋ ค์ง„ ๋‹คํ”„๋„ค ๋ฐฐ๋งํ„ด"์ด๋ผ๋Š” ์˜์—ญ์„ ๋ชจ๋‘ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ด์ „ ์—ฐ๊ตฌ์— ๋”ฐ๋ฅด๋ฉด ์‚ฌ์‹ค ์ง„์ˆ ์˜ ์˜์—ญ์„ ํฌํ•จํ•˜๋ฉด ๋ชจ๋ธ์ด ์ง„์ˆ ์—์„œ ์ผ๋ฐ˜ํ™”ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋œ๋‹ค๊ณ  ํ•ฉ๋‹ˆ๋‹ค(Berglund et al., 2023). ์˜์—ญ์€ ํ•ญ์ƒ ์›๋ž˜ ๋ฌธ์žฅ์—์„œ ์ด๋ฆ„๊ณผ ์„ค๋ช…์˜ ์ˆœ์„œ์™€ ์ผ์น˜ํ•ฉ๋‹ˆ๋‹ค.
We stress that in our experiments, we do not distinguish between undesired memorized text (such as phone numbers), innocuous memorized text (common phrases), and text we may want to be memorized (such as a quote by a public figure), and instead treat all instances of the LM generating text that closely matches the training set as problematic. While we qualitatively observed that much of the identified memorized content was relatively innocuous, a more systematic study of the risks associated with the detected memorization was beyond the scope of this work.
์šฐ๋ฆฌ์˜ ์‹คํ—˜์—์„œ ์šฐ๋ฆฌ๋Š” ์›ํ•˜์ง€ ์•Š๋Š” ์•”๊ธฐ๋œ ํ…์ŠคํŠธ(์˜ˆ: ์ „ํ™”๋ฒˆํ˜ธ), ๋ฌดํ•ดํ•œ ์•”๊ธฐ๋œ ํ…์ŠคํŠธ(์ผ๋ฐ˜์ ์ธ ๋ฌธ๊ตฌ), ๊ทธ๋ฆฌ๊ณ  ์šฐ๋ฆฌ๊ฐ€ ์•”๊ธฐ๋˜๊ธฐ๋ฅผ ์›ํ•˜๋Š” ํ…์ŠคํŠธ(์˜ˆ: ๊ณต์ธ์˜ ์ธ์šฉ๋ฌธ) ์‚ฌ์ด๋ฅผ ๊ตฌ๋ถ„ํ•˜์ง€ ์•Š๊ณ , ๋Œ€์‹  ํ›ˆ๋ จ ์„ธํŠธ์™€ ๋งค์šฐ ์œ ์‚ฌํ•œ ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” LM์˜ ๋ชจ๋“  ์‚ฌ๋ก€๋ฅผ ๋ฌธ์ œ๋กœ ๊ฐ„์ฃผํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๊ฐ€ ์งˆ์ ์œผ๋กœ ๊ด€์ฐฐํ•œ ๋ฐ”์— ๋”ฐ๋ฅด๋ฉด, ์‹๋ณ„๋œ ์•”๊ธฐ๋œ ์ฝ˜ํ…์ธ ์˜ ๋Œ€๋ถ€๋ถ„์€ ๋น„๊ต์  ๋ฌดํ•ดํ–ˆ์ง€๋งŒ, ๊ฐ์ง€๋œ ์•”๊ธฐ์™€ ๊ด€๋ จ๋œ ์œ„ํ—˜์— ๋Œ€ํ•œ ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์—ฐ๊ตฌ๋Š” ์ด ์ž‘์—…์˜ ๋ฒ”์œ„๋ฅผ ๋ฒ—์–ด๋‚ฌ์Šต๋‹ˆ๋‹ค.
์‹คํ—˜์—์„œ๋Š” ์›์น˜ ์•Š๋Š” ์•”๊ธฐ ํ…์ŠคํŠธ(์˜ˆ: ์ „ํ™”๋ฒˆํ˜ธ), ๋ฌดํ•ดํ•œ ์•”๊ธฐ ํ…์ŠคํŠธ(์ผ๋ฐ˜์ ์ธ ๋ฌธ๊ตฌ), ์•”๊ธฐํ•˜๊ณ  ์‹ถ์€ ํ…์ŠคํŠธ(์˜ˆ: ์œ ๋ช…์ธ์˜ ๋ช…์–ธ)๋ฅผ ๊ตฌ๋ถ„ํ•˜์ง€ ์•Š๊ณ , ๋Œ€์‹  ํ•™์Šต ์„ธํŠธ์™€ ๋งค์šฐ ์œ ์‚ฌํ•œ ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” LM์˜ ๋ชจ๋“  ์ธ์Šคํ„ด์Šค๋ฅผ ๋ฌธ์ œ๊ฐ€ ์žˆ๋Š” ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผํ•ฉ๋‹ˆ๋‹ค. ์‹๋ณ„๋œ ์•”๊ธฐ ์ฝ˜ํ…์ธ ์˜ ๋Œ€๋ถ€๋ถ„์ด ๋น„๊ต์  ๋ฌดํ•ดํ•œ ๊ฒƒ์œผ๋กœ ์ •์„ฑ์ ์œผ๋กœ ๊ด€์ฐฐ๋˜์—ˆ์ง€๋งŒ, ๊ฐ์ง€๋œ ์•”๊ธฐ์™€ ๊ด€๋ จ๋œ ์œ„ํ—˜์— ๋Œ€ํ•œ ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ์—ฐ๊ตฌ๋Š” ์ด ์ž‘์—…์˜ ๋ฒ”์œ„๋ฅผ ๋ฒ—์–ด๋‚ฌ์Šต๋‹ˆ๋‹ค.
Claim verification for evaluating summaries:Our paper also relates to prior work on claim verification, where claims are verified given reference to some knowledge source (Thorne et al., 2018; Wadden et al., 2020; Schuster et al., 2021). Min et al. (2023) propose FActScore, an LLM-based metric of factual precision in biography generation, which was expanded upon in SAFE (Wei et al., 2024). Manakul et al. (2023) propose SelfCheckGPT, which uses LLMs to evaluate the faithfulness of GPT-3 generated texts on a dataset of Wikipedia-style passages about people.
์ฃผ์žฅ ๊ฒ€์ฆ์„ ํ†ตํ•œ ์š”์•ฝ ํ‰๊ฐ€: ์šฐ๋ฆฌ์˜ ๋…ผ๋ฌธ์€ ๋˜ํ•œ ์ฃผ์žฅ ๊ฒ€์ฆ์— ๊ด€ํ•œ ์ด์ „ ์—ฐ๊ตฌ์™€ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ ์ฃผ์žฅ์€ ํŠน์ • ์ง€์‹ ์ถœ์ฒ˜๋ฅผ ์ฐธ์กฐํ•˜์—ฌ ๊ฒ€์ฆ๋ฉ๋‹ˆ๋‹ค (Thorne et al., 2018; Wadden et al., 2020; Schuster et al., 2021). Min et al. (2023)์€ ์ „๊ธฐ ์ƒ์„ฑ์—์„œ ์‚ฌ์‹ค์  ์ •ํ™•์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” LLM ๊ธฐ๋ฐ˜ ๋ฉ”ํŠธ๋ฆญ์ธ FActScore๋ฅผ ์ œ์•ˆํ–ˆ์œผ๋ฉฐ, ์ด๋Š” SAFE (Wei et al., 2024)์—์„œ ํ™•์žฅ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. Manakul et al. (2023)์€ ์‚ฌ๋žŒ๋“ค์— ๋Œ€ํ•œ ์œ„ํ‚คํ”ผ๋””์•„ ์Šคํƒ€์ผ์˜ ๋ฌธ๋‹จ ๋ฐ์ดํ„ฐ์…‹์—์„œ GPT-3๊ฐ€ ์ƒ์„ฑํ•œ ํ…์ŠคํŠธ์˜ ์‹ ๋ขฐ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด LLM์„ ์‚ฌ์šฉํ•˜๋Š” SelfCheckGPT๋ฅผ ์ œ์•ˆํ–ˆ์Šต๋‹ˆ๋‹ค.
์š”์•ฝ ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ ํด๋ ˆ์ž„ ๊ฒ€์ฆ: ์ €ํฌ ๋…ผ๋ฌธ์€ ํด๋ ˆ์ž„์ด ์ง€์‹ ์†Œ์Šค์— ๋Œ€ํ•œ ์ฐธ์กฐ๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ ๊ฒ€์ฆ๋˜๋Š” ํด๋ ˆ์ž„ ๊ฒ€์ฆ์— ๋Œ€ํ•œ ์„ ํ–‰ ์—ฐ๊ตฌ์™€๋„ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค(Thorne et al., 2018; Wadden et al., 2020; Schuster et al., 2021). Min et al. (2023)์€ ์ „๊ธฐ ์ƒ์„ฑ์—์„œ ์‚ฌ์‹ค์  ์ •ํ™•์„ฑ์— ๋Œ€ํ•œ LLM ๊ธฐ๋ฐ˜ ๋ฉ”ํŠธ๋ฆญ์ธ FActScore๋ฅผ ์ œ์•ˆํ–ˆ์œผ๋ฉฐ, ์ด๋Š” SAFE(Wei et al., 2024)์—์„œ ํ™•์žฅ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. Manakul et al. (2023)์€ LLM์„ ์‚ฌ์šฉํ•˜์—ฌ ์‚ฌ๋žŒ์— ๋Œ€ํ•œ ์œ„ํ‚คํ”ผ๋””์•„ ์Šคํƒ€์ผ์˜ ๊ตฌ์ ˆ ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ GPT-3 ์ƒ์„ฑ ํ…์ŠคํŠธ์˜ ์ถฉ์‹ค์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” SelfCheckGPT๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค.