question
stringlengths
101
2.69k
chosen
stringlengths
43
2.27k
rejected
stringlengths
42
2.18k
However, the ReAct method is not without its limitations. While intertwining reasoning, observation, and action improves trustworthiness, the model can repetitively generate the same thoughts and actions and fail to create new thoughts to provoke finishing the task and exiting the ReAct loop. Incorporating human feedback during the execution of the task would likely increase its effectiveness and applicability in real-world scenarios.
๊ทธ๋Ÿฌ๋‚˜ ReAct ๋ฐฉ๋ฒ•์—๋„ ํ•œ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ถ”๋ก , ๊ด€์ฐฐ ๋ฐ ํ–‰๋™์„ ๊ฒฐํ•ฉํ•˜๋ฉด ์‹ ๋ขฐ์„ฑ์ด ํ–ฅ์ƒ๋˜์ง€๋งŒ, ๋ชจ๋ธ์ด ๋™์ผํ•œ ์ƒ๊ฐ๊ณผ ํ–‰๋™์„ ๋ฐ˜๋ณต์ ์œผ๋กœ ์ƒ์„ฑํ•˜๊ณ  ์ƒˆ๋กœ์šด ์ƒ๊ฐ์„ ๋งŒ๋“ค์–ด๋‚ด์ง€ ๋ชปํ•ด ์ž‘์—…์„ ์™„๋ฃŒํ•˜๊ณ  ReAct ๋ฃจํ”„์—์„œ ๋ฒ—์–ด๋‚˜์ง€ ๋ชปํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ž‘์—… ์ˆ˜ํ–‰ ์ค‘์— ์ธ๊ฐ„์˜ ํ”ผ๋“œ๋ฐฑ์„ ํ†ตํ•ฉํ•˜๋ฉด ์‹ค์ œ ์ƒํ™ฉ์—์„œ์˜ ํšจ๊ณผ์„ฑ๊ณผ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์•„์งˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๊ทธ๋Ÿฌ๋‚˜ ReAct ๋ฐฉ๋ฒ•์—๋Š” ํ•œ๊ณ„๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. ์ถ”๋ก , ๊ด€์ฐฐ, ํ–‰๋™์„ ์–ฝํžˆ๊ฒŒ ํ•˜๋ฉด ์‹ ๋ขฐ์„ฑ์ด ํ–ฅ์ƒ๋˜์ง€๋งŒ, ๋ชจ๋ธ์ด ๋™์ผํ•œ ์ƒ๊ฐ๊ณผ ํ–‰๋™์„ ๋ฐ˜๋ณต์ ์œผ๋กœ ์ƒ์„ฑํ•˜๊ณ  ์ž‘์—…์„ ์™„๋ฃŒํ•˜๊ณ  ReAct ๋ฃจํ”„๋ฅผ ์ข…๋ฃŒํ•˜๋„๋ก ์œ ๋„ํ•˜๋Š” ์ƒˆ๋กœ์šด ์ƒ๊ฐ์„ ์ƒ์„ฑํ•˜์ง€ ๋ชปํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๋™์•ˆ ์ธ๊ฐ„์˜ ํ”ผ๋“œ๋ฐฑ์„ ํ†ตํ•ฉํ•˜๋ฉด ์‹ค์ œ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ์˜ ํšจ๊ณผ์™€ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ๋†’์ผ ์ˆ˜ ์žˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Next we test our method on a realistic setup where we pre-train language models, and evaluate their ability on "forward" and "reverse" facts about real-world knowledge. As LLMs acquire the majority of their world knowledge during their pre-training stage, it makes sense to evaluate our reverse training in this pre-training setup. To make the experiments tractable, we train a Llama-2 1.4 billion parameter model (Touvron et al., 2023b).
๋‹ค์Œ์œผ๋กœ ์šฐ๋ฆฌ๋Š” ์–ธ์–ด ๋ชจ๋ธ์„ ์‚ฌ์ „ ํ›ˆ๋ จํ•˜๊ณ , ์‹ค์ œ ์„ธ๊ณ„ ์ง€์‹์— ๋Œ€ํ•œ "์ •๋ฐฉํ–ฅ" ๋ฐ "์—ญ๋ฐฉํ–ฅ" ์‚ฌ์‹ค์— ๋Œ€ํ•œ ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ํ˜„์‹ค์ ์ธ ์„ค์ •์—์„œ ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์„ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค. LLM์€ ์‚ฌ์ „ ํ›ˆ๋ จ ๋‹จ๊ณ„์—์„œ ๋Œ€๋ถ€๋ถ„์˜ ์„ธ๊ณ„ ์ง€์‹์„ ์Šต๋“ํ•˜๋ฏ€๋กœ, ์ด ์‚ฌ์ „ ํ›ˆ๋ จ ์„ค์ •์—์„œ ์šฐ๋ฆฌ์˜ ์—ญ๋ฐฉํ–ฅ ํ›ˆ๋ จ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์ด ํ•ฉ๋ฆฌ์ ์ž…๋‹ˆ๋‹ค. ์‹คํ—˜์„ ๋‹ค๋ฃจ๊ธฐ ์‰ฝ๊ฒŒ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” Llama-2 1.4 billion ํŒŒ๋ผ๋ฏธํ„ฐ ๋ชจ๋ธ(Touvron et al., 2023b)์„ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค.
๋‹ค์Œ์œผ๋กœ ์–ธ์–ด ๋ชจ๋ธ์„ ์‚ฌ์ „ ํ•™์Šตํ•˜๊ณ  ์‹ค์ œ ์ง€์‹์— ๋Œ€ํ•œ '์ „๋‹ฌ' ๋ฐ '์—ญ๋ฐฉํ–ฅ' ์‚ฌ์‹ค์— ๋Œ€ํ•œ ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ํ˜„์‹ค์ ์ธ ์„ค์ •์—์„œ ๋ฐฉ๋ฒ•์„ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค. LLM์€ ์‚ฌ์ „ ํ•™์Šต ๋‹จ๊ณ„์—์„œ ๋Œ€๋ถ€๋ถ„์˜ ์„ธ๊ณ„ ์ง€์‹์„ ์Šต๋“ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ด ์‚ฌ์ „ ํ•™์Šต ์„ค์ •์—์„œ ์—ญ๋ฐฉํ–ฅ ํ•™์Šต์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์ด ํ•ฉ๋ฆฌ์ ์ž…๋‹ˆ๋‹ค. ์‹คํ—˜์„ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•ด 14์–ต ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๊ฐ€์ง„ Llama-2 ๋ชจ๋ธ(Touvron et al., 2023b)์„ ํ›ˆ๋ จํ•ฉ๋‹ˆ๋‹ค.
The authors would like to thank the anonymous reviewers for their constructive feedback. We especially thank Sewon Min, Eric Wallace, Ofir Press, and other members of UWNLP and AllenNLP for their encouraging feedback and intellectual support. This work was supported in part by DARPA MCS program through NIWC Pacific (N66001-19-2-4031), ONR N00014-18-1-2826, ONR MURI N00014-18-1-2670, and gifts from AI2 and an Allen Investigator award.
์ €์ž๋“ค์€ ์ต๋ช…์˜ ๋ฆฌ๋ทฐ์–ด๋“ค์—๊ฒŒ ๊ฑด์„ค์ ์ธ ํ”ผ๋“œ๋ฐฑ์— ๋Œ€ํ•ด ๊ฐ์‚ฌ์˜ ๋œป์„ ์ „ํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ Sewon Min, Eric Wallace, Ofir Press, ๊ทธ๋ฆฌ๊ณ  UWNLP์™€ AllenNLP์˜ ๋‹ค๋ฅธ ๋ฉค๋ฒ„๋“ค์—๊ฒŒ ๊ฒฉ๋ ค์˜ ํ”ผ๋“œ๋ฐฑ๊ณผ ์ง€์  ์ง€์›์— ๋Œ€ํ•ด ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” NIWC Pacific (N66001-19-2-4031)์„ ํ†ตํ•œ DARPA MCS ํ”„๋กœ๊ทธ๋žจ, ONR N00014-18-1-2826, ONR MURI N00014-18-1-2670, AI2์˜ ๊ธฐ๋ถ€ ๋ฐ Allen Investigator ์ƒ์˜ ์ผ๋ถ€ ์ง€์›์„ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค.
์ €์ž๋Š” ๊ฑด์„ค์ ์ธ ํ”ผ๋“œ๋ฐฑ์„ ์ œ๊ณตํ•œ ์ต๋ช…์˜ ๋ฆฌ๋ทฐ์–ด์—๊ฒŒ ๊ฐ์‚ฌ์˜ ๋ง์„ ์ „ํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ Sewon Min, Eric Wallace, Ofir Press ๋ฐ UWNLP์™€ AllenNLP์˜ ๋‹ค๋ฅธ ๊ตฌ์„ฑ์›๋“ค์ด ๊ณ ๋ฌด์ ์ธ ํ”ผ๋“œ๋ฐฑ๊ณผ ์ง€์  ์ง€์›์„ ์ œ๊ณตํ•ด ์ฃผ์…”์„œ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” NIWC Pacific(N66001-19-2-4031), ONR N00014-18-1-2826, ONR MURI N00014-18-1-2670, AI2 ๋ฐ Allen ์—ฐ๊ตฌ์› ์ƒ์˜ ์ง€์›์„ ํ†ตํ•ด ๋ถ€๋ถ„์ ์œผ๋กœ ์ง€์›๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
Figure 1: Illustration of the growth trend in the field of LLM-based autonomous agents. We present the cumulative number of papers published from January 2021 to August 2023. We assign different colors to represent various agent categories. For example, a game agent aims to simulate a game-player, while a tool agent mainly focuses on tool using. For each time period, we provide a curated list of studies with diverse agent categories.
Figure 1: LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ๋ถ„์•ผ์˜ ์„ฑ์žฅ ์ถ”์„ธ๋ฅผ ๋ณด์—ฌ์ฃผ๋Š” ๊ทธ๋ฆผ์ž…๋‹ˆ๋‹ค. 2021๋…„ 1์›”๋ถ€ํ„ฐ 2023๋…„ 8์›”๊นŒ์ง€ ๋ฐœํ‘œ๋œ ๋…ผ๋ฌธ์˜ ๋ˆ„์  ์ˆ˜๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ๋‚˜ํƒ€๋‚ด๊ธฐ ์œ„ํ•ด ์„œ๋กœ ๋‹ค๋ฅธ ์ƒ‰์ƒ์„ ํ• ๋‹นํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๊ฒŒ์ž„ ์—์ด์ „ํŠธ๋Š” ๊ฒŒ์ž„ ํ”Œ๋ ˆ์ด์–ด๋ฅผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•˜๊ณ , ๋„๊ตฌ ์—์ด์ „ํŠธ๋Š” ์ฃผ๋กœ ๋„๊ตฌ ์‚ฌ์šฉ์— ์ค‘์ ์„ ๋‘ก๋‹ˆ๋‹ค. ๊ฐ ๊ธฐ๊ฐ„๋งˆ๋‹ค ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ํฌํ•จํ•œ ์—ฐ๊ตฌ ๋ชฉ๋ก์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 1: LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ๋ถ„์•ผ์˜ ์„ฑ์žฅ ์ถ”์„ธ๋ฅผ ๋ณด์—ฌ์ฃผ๋Š” ๊ทธ๋ฆผ์ž…๋‹ˆ๋‹ค. 2021๋…„ 1์›”๋ถ€ํ„ฐ 2023๋…„ 8์›”๊นŒ์ง€ ๋ฐœํ‘œ๋œ ๋ˆ„์  ๋…ผ๋ฌธ ์ˆ˜๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ๋‚˜ํƒ€๋‚ด๊ธฐ ์œ„ํ•ด ๋‹ค์–‘ํ•œ ์ƒ‰์ƒ์„ ํ• ๋‹นํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๊ฒŒ์ž„ ์—์ด์ „ํŠธ๋Š” ๊ฒŒ์ž„ ํ”Œ๋ ˆ์ด์–ด๋ฅผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜ํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘๋Š” ๋ฐ˜๋ฉด, ๋„๊ตฌ ์—์ด์ „ํŠธ๋Š” ์ฃผ๋กœ ๋„๊ตฌ ์‚ฌ์šฉ์— ์ค‘์ ์„ ๋‘ก๋‹ˆ๋‹ค. ๊ฐ ๊ธฐ๊ฐ„๋งˆ๋‹ค ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ ์นดํ…Œ๊ณ ๋ฆฌ๋ฅผ ๊ฐ€์ง„ ํ๋ ˆ์ด์…˜๋œ ์—ฐ๊ตฌ ๋ชฉ๋ก์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
Just as they were about to close their eyes, they heard a strange noise that caught their attention. They stood up and cautiously made their way towards the source of the sound. As they approached the fence that bordered their yard, they saw a protrusion that seemed to jut out from the wood. It was about the size and shape of a tennis ball, but it was a dull gray color and seemed to be made of some kind of hard, yet pliable material.
๊ทธ๋“ค์ด ๋ง‰ ๋ˆˆ์„ ๊ฐ์œผ๋ ค๋˜ ์ˆœ๊ฐ„, ์ฃผ์˜๋ฅผ ๋„๋Š” ์ด์ƒํ•œ ์†Œ๋ฆฌ๋ฅผ ๋“ค์—ˆ๋‹ค. ๊ทธ๋“ค์€ ์ผ์–ด๋‚˜์„œ ์†Œ๋ฆฌ๊ฐ€ ๋‚˜๋Š” ์ชฝ์œผ๋กœ ์กฐ์‹ฌ์Šค๋Ÿฝ๊ฒŒ ๋‹ค๊ฐ€๊ฐ”๋‹ค. ๋งˆ๋‹น์„ ๋‘˜๋Ÿฌ์‹ผ ์šธํƒ€๋ฆฌ์— ๊ฐ€๊นŒ์ด ๋‹ค๊ฐ€๊ฐ€์ž, ๋‚˜๋ฌด์—์„œ ํŠ€์–ด๋‚˜์˜จ ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ด๋Š” ๋Œ์ถœ๋ถ€๋ฅผ ๋ฐœ๊ฒฌํ–ˆ๋‹ค. ๊ทธ๊ฒƒ์€ ํ…Œ๋‹ˆ์Šค ๊ณต ํฌ๊ธฐ์™€ ๋ชจ์–‘์ด์—ˆ์ง€๋งŒ, ์น™์น™ํ•œ ํšŒ์ƒ‰์ด์—ˆ๊ณ  ์–ด๋–ค ๋‹จ๋‹จํ•˜์ง€๋งŒ ์œ ์—ฐํ•œ ์žฌ์งˆ๋กœ ๋งŒ๋“ค์–ด์ง„ ๊ฒƒ ๊ฐ™์•˜๋‹ค.
๋ˆˆ์„ ๊ฐ์œผ๋ ค๋˜ ์ˆœ๊ฐ„, ์ด์ƒํ•œ ์†Œ๋ฆฌ๊ฐ€ ๋“ค๋ ค์™€ ์ฃผ์˜๋ฅผ ๋Œ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋“ค์€ ์ผ์–ด์„œ์„œ ์กฐ์‹ฌ์Šค๋Ÿฝ๊ฒŒ ์†Œ๋ฆฌ์˜ ๊ทผ์›์ง€๋ฅผ ํ–ฅํ•ด ๊ฑธ์–ด๊ฐ”์Šต๋‹ˆ๋‹ค. ๋งˆ๋‹น๊ณผ ์ ‘ํ•ด ์žˆ๋Š” ์šธํƒ€๋ฆฌ์— ๋‹ค๊ฐ€๊ฐ€์ž ๋‚˜๋ฌด์—์„œ ํŠ€์–ด๋‚˜์˜จ ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ด๋Š” ๋Œ์ถœ๋ถ€๊ฐ€ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ํ…Œ๋‹ˆ์Šค ๊ณต ํฌ๊ธฐ์™€ ๋ชจ์–‘์ด์—ˆ์ง€๋งŒ ์น™์น™ํ•œ ํšŒ์ƒ‰์„ ๋ ๊ณ  ์žˆ์—ˆ๊ณ  ์–ด๋–ค ์ข…๋ฅ˜์˜ ๋‹จ๋‹จํ•˜๋ฉด์„œ๋„ ์œ ์—ฐํ•œ ์žฌ์งˆ๋กœ ๋งŒ๋“ค์–ด์ง„ ๊ฒƒ ๊ฐ™์•˜์Šต๋‹ˆ๋‹ค.
**Output**[**Retriee**]=Yes]<p>The Walking Dead (season 7)-The Walking Dead (season 7) The seventh season of "The Walking Dead", an American post-apocalypetric horror television series on AMC, premiered on October 23, 2016, and concluded on April 2, 2017, consisting of 16 episodes. Developed for television by Frank Darabout, the series is based on the eponymous series of comic books by Robert Kirkman, Tony Moore, and Charlie Adlard.
**Output**[**Retriee**]=Yes]<p>The Walking Dead (season 7)-The Walking Dead (season 7) ๋ฏธ๊ตญ์˜ ํฌ์ŠคํŠธ ์•„ํฌ์นผ๋ฆฝ์Šค ํ˜ธ๋Ÿฌ ํ…”๋ ˆ๋น„์ „ ์‹œ๋ฆฌ์ฆˆ์ธ "The Walking Dead"์˜ ์ผ๊ณฑ ๋ฒˆ์งธ ์‹œ์ฆŒ์€ 2016๋…„ 10์›” 23์ผ์— ์ฒซ ๋ฐฉ์˜๋˜์—ˆ์œผ๋ฉฐ, 2017๋…„ 4์›” 2์ผ์— 16๊ฐœ์˜ ์—ํ”ผ์†Œ๋“œ๋กœ ๋งˆ๋ฌด๋ฆฌ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ํ”„๋žญํฌ ๋‹ค๋ผ๋ณธํŠธ๊ฐ€ ํ…”๋ ˆ๋น„์ „์šฉ์œผ๋กœ ๊ฐœ๋ฐœํ•œ ์ด ์‹œ๋ฆฌ์ฆˆ๋Š” ๋กœ๋ฒ„ํŠธ ์ปคํฌ๋จผ, ํ† ๋‹ˆ ๋ฌด์–ด, ์ฐฐ๋ฆฌ ์• ๋“ค๋Ÿฌ๋“œ์˜ ๋™๋ช… ๋งŒํ™”์ฑ… ์‹œ๋ฆฌ์ฆˆ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
**์ถœ๋ ฅ**[**Retriee**]=Yes]<p>์›Œํ‚น ๋ฐ๋“œ (์‹œ์ฆŒ 7)-์›Œํ‚น ๋ฐ๋“œ (์‹œ์ฆŒ 7) AMC์˜ ๋ฏธ๊ตญ ํฌ์ŠคํŠธ ์•„ํฌ์นผ๋ฆฝํŠธ ๊ณตํฌ ํ…”๋ ˆ๋น„์ „ ์‹œ๋ฆฌ์ฆˆ์ธ "์›Œํ‚น ๋ฐ๋“œ"์˜ ์ผ๊ณฑ ๋ฒˆ์งธ ์‹œ์ฆŒ์€ 2016๋…„ 10์›” 23์ผ์— ์ฒซ ๋ฐฉ์†ก๋˜์–ด 2017๋…„ 4์›” 2์ผ์— 16๊ฐœ์˜ ์—ํ”ผ์†Œ๋“œ๋กœ ๊ตฌ์„ฑ๋œ ์—ํ”ผ์†Œ๋“œ๊ฐ€ ๋ฐฉ์˜๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ํ”„๋žญํฌ ๋‹ค๋ผ๋ฐ”์›ƒ์ด ํ…”๋ ˆ๋น„์ „์šฉ์œผ๋กœ ๊ฐœ๋ฐœํ•œ ์ด ์‹œ๋ฆฌ์ฆˆ๋Š” ๋กœ๋ฒ„ํŠธ ์ปคํฌ๋จผ, ํ† ๋‹ˆ ๋ฌด์–ด, ์ฐฐ๋ฆฌ ์• ๋“ค๋ผ๋“œ์˜ ์‹œ์กฐ ๋งŒํ™”๋ฅผ ์›์ž‘์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
We now return to our main problem, understanding model collapse in the probabilistic setting and consider the Hutter++ presented above. Thus, suppose the learner only has access to at most a dataset of size \(T\) containing the \(k\)th head of the conditional distribution \(p(\cdot\mid i)\). That is, sampled from: \(i\sim p\), \(j\sim p(j\mid i)1[j\leq k]\) (normalized appropriately), where \(p(\cdot\mid i)\) is as in Equation (16).
์ด์ œ ์šฐ๋ฆฌ๋Š” ์ฃผ๋œ ๋ฌธ์ œ๋กœ ๋Œ์•„๊ฐ€์„œ, ํ™•๋ฅ ์  ์„ค์ •์—์„œ ๋ชจ๋ธ ๋ถ•๊ดด๋ฅผ ์ดํ•ดํ•˜๊ณ  ์œ„์—์„œ ์ œ์‹œํ•œ Hutter++๋ฅผ ๊ณ ๋ คํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ํ•™์Šต์ž๋Š” ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ \(p(\cdot\mid i)\)์˜ \(k\)๋ฒˆ์งธ ํ—ค๋“œ๋ฅผ ํฌํ•จํ•˜๋Š” ํฌ๊ธฐ \(T\)์˜ ๋ฐ์ดํ„ฐ์…‹์—๋งŒ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๊ฐ€์ •ํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰, ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ƒ˜ํ”Œ๋ง๋ฉ๋‹ˆ๋‹ค: \(i\sim p\), \(j\sim p(j\mid i)1[j\leq k]\) (์ ์ ˆํžˆ ์ •๊ทœํ™”๋จ), ์—ฌ๊ธฐ์„œ \(p(\cdot\mid i)\)๋Š” ์‹ (16)๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.
์ด์ œ ํ™•๋ฅ ์  ์„ค์ •์—์„œ ๋ชจ๋ธ ๋ถ•๊ดด๋ฅผ ์ดํ•ดํ•˜๋Š” ์ฃผ์š” ๋ฌธ์ œ๋กœ ๋Œ์•„์™€์„œ ์œ„์—์„œ ์ œ์‹œํ•œ Hutter++๋ฅผ ๊ณ ๋ คํ•ด ๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ํ•™์Šต์ž๊ฐ€ ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ \(p(\cdot\mid i)\)์˜ \(k\)๋ฒˆ์งธ ํ—ค๋“œ๋ฅผ ํฌํ•จํ•˜๋Š” ์ตœ๋Œ€ \(T\) ํฌ๊ธฐ์˜ ๋ฐ์ดํ„ฐ ์„ธํŠธ์—๋งŒ ์•ก์„ธ์Šคํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๊ฐ€์ •ํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰, ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ƒ˜ํ”Œ๋งํ•ฉ๋‹ˆ๋‹ค: i\sim p, j\sim p(j\mid i)1[j\leq k]\)(์ ์ ˆํ•˜๊ฒŒ ์ •๊ทœํ™”๋จ), ์—ฌ๊ธฐ์„œ \(p(\cdot\mid i)\)๋Š” ์‹ (16)๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค.
In the above sections, we have detailed the construction of LLM-based agents, where we focus on two aspects including the architecture design and capability acquisition. We present the correspondence between existing work and the above taxonomy in Table 1. It should be noted that, for the sake of integrity, we have also incorporated several studies, which do not explicitly mention LLM-based agentsbut are highly related to this area.
์œ„์˜ ์„น์…˜๋“ค์—์„œ ์šฐ๋ฆฌ๋Š” LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ์˜ ๊ตฌ์ถ•์— ๋Œ€ํ•ด ์ž์„ธํžˆ ์„ค๋ช…ํ–ˆ์œผ๋ฉฐ, ์—ฌ๊ธฐ์„œ ์•„ํ‚คํ…์ฒ˜ ์„ค๊ณ„์™€ ๋Šฅ๋ ฅ ํš๋“์„ ํฌํ•จํ•œ ๋‘ ๊ฐ€์ง€ ์ธก๋ฉด์— ์ค‘์ ์„ ๋‘์—ˆ์Šต๋‹ˆ๋‹ค. Table 1์—์„œ๋Š” ๊ธฐ์กด ์—ฐ๊ตฌ์™€ ์œ„์˜ ๋ถ„๋ฅ˜ ์ฒด๊ณ„ ๊ฐ„์˜ ๋Œ€์‘ ๊ด€๊ณ„๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์™„์ „์„ฑ์„ ์œ„ํ•ด LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ๋ฅผ ๋ช…์‹œ์ ์œผ๋กœ ์–ธ๊ธ‰ํ•˜์ง€ ์•Š์ง€๋งŒ ์ด ๋ถ„์•ผ์™€ ๋งค์šฐ ๊ด€๋ จ์ด ์žˆ๋Š” ์—ฌ๋Ÿฌ ์—ฐ๊ตฌ๋“ค๋„ ํฌํ•จ์‹œ์ผฐ์Œ์„ ์ฃผ๋ชฉํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
์œ„์˜ ์„น์…˜์—์„œ๋Š” ์•„ํ‚คํ…์ฒ˜ ์„ค๊ณ„์™€ ์—ญ๋Ÿ‰ ํ™•๋ณด ๋“ฑ ๋‘ ๊ฐ€์ง€ ์ธก๋ฉด์— ์ดˆ์ ์„ ๋งž์ถ˜ LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ์˜ ๊ตฌ์ถ•์— ๋Œ€ํ•ด ์ž์„ธํžˆ ์„ค๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ์™€ ์œ„์˜ ๋ถ„๋ฅ˜ ์ฒด๊ณ„ ๊ฐ„์˜ ๋Œ€์‘ ๊ด€๊ณ„๋Š” ํ‘œ 1์— ์ œ์‹œ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์™„์ „์„ฑ์„ ์œ„ํ•ด LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ๋ฅผ ๋ช…์‹œ์ ์œผ๋กœ ์–ธ๊ธ‰ํ•˜์ง€๋Š” ์•Š์•˜์ง€๋งŒ ์ด ๋ถ„์•ผ์™€ ๊ด€๋ จ์ด ๋†’์€ ๋ช‡ ๊ฐ€์ง€ ์—ฐ๊ตฌ๋„ ํฌํ•จํ–ˆ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
Model merging in the PS and in the DFS are orthogonal approaches, however, it is straightforward to combine these disentangled methods and further boost the performance of a merged model. As we show in the rightmost illustration in Figure 1 and in Section 4.1, it is possible to first apply PS merging to a collection of models, and then put back this merged model in the collection and apply DFS merging from this enlarged collection.
PS์™€ DFS์—์„œ์˜ ๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ์„œ๋กœ ๋…๋ฆฝ์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด์ง€๋งŒ, ์ด๋Ÿฌํ•œ ๋ถ„๋ฆฌ๋œ ๋ฐฉ๋ฒ•๋“ค์„ ๊ฒฐํ•ฉํ•˜์—ฌ ๋ณ‘ํ•ฉ๋œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ๋”์šฑ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์€ ๊ฐ„๋‹จํ•ฉ๋‹ˆ๋‹ค. Figure 1์˜ ๊ฐ€์žฅ ์˜ค๋ฅธ์ชฝ ๊ทธ๋ฆผ๊ณผ Section 4.1์—์„œ ๋ณด์—ฌ์ฃผ๋“ฏ์ด, ๋จผ์ € ์—ฌ๋Ÿฌ ๋ชจ๋ธ์— PS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•œ ํ›„, ์ด ๋ณ‘ํ•ฉ๋œ ๋ชจ๋ธ์„ ๋‹ค์‹œ ์ปฌ๋ ‰์…˜์— ๋„ฃ๊ณ  ํ™•์žฅ๋œ ์ปฌ๋ ‰์…˜์—์„œ DFS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•˜๋Š” ๊ฒƒ์ด ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.
PS์™€ DFS์˜ ๋ชจ๋ธ ๋ณ‘ํ•ฉ์€ ์ง๊ต ์ ‘๊ทผ ๋ฐฉ์‹์ด์ง€๋งŒ, ์ด๋Ÿฌํ•œ ๋ถ„๋ฆฌ๋œ ๋ฐฉ๋ฒ•์„ ๊ฒฐํ•ฉํ•˜์—ฌ ๋ณ‘ํ•ฉ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ๋”์šฑ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๊ฒƒ์€ ๊ฐ„๋‹จํ•˜๋‹ค. ๊ทธ๋ฆผ 1๊ณผ ์„น์…˜ 4.1์˜ ๊ฐ€์žฅ ์˜ค๋ฅธ์ชฝ ๊ทธ๋ฆผ์—์„œ ๋ณผ ์ˆ˜ ์žˆ๋“ฏ์ด, ๋จผ์ € ์—ฌ๋Ÿฌ ๋ชจ๋ธ ๋ชจ์Œ์— PS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•œ ๋‹ค์Œ ์ด ๋ณ‘ํ•ฉ๋œ ๋ชจ๋ธ์„ ๋ชจ์Œ์— ๋‹ค์‹œ ๋„ฃ๊ณ  ์ด ํ™•์žฅ๋œ ๋ชจ์Œ์—์„œ DFS ๋ณ‘ํ•ฉ์„ ์ ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค.
For most technical systems, the two dominant approaches for assessing safety are (1) formal analysis of the system's properties and (2) exhaustively investigating the system's safety within its domain [124, 125, 126, 41]. As with other complex technological systems, AI systems pose challenges due to their complexity and unpredictability [127]; their socio-technical entanglement; and challenges in methods and data access [128, 129].
๋Œ€๋ถ€๋ถ„์˜ ๊ธฐ์ˆ  ์‹œ์Šคํ…œ์—์„œ ์•ˆ์ „์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ์ ‘๊ทผ ๋ฐฉ์‹์€ (1) ์‹œ์Šคํ…œ ์†์„ฑ์— ๋Œ€ํ•œ ํ˜•์‹์  ๋ถ„์„๊ณผ (2) ์‹œ์Šคํ…œ์˜ ๋„๋ฉ”์ธ ๋‚ด์—์„œ ์•ˆ์ „์„ฑ์„ ์ฒ ์ €ํžˆ ์กฐ์‚ฌํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค [124, 125, 126, 41]. ๋‹ค๋ฅธ ๋ณต์žกํ•œ ๊ธฐ์ˆ  ์‹œ์Šคํ…œ๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ, AI ์‹œ์Šคํ…œ์€ ๊ทธ ๋ณต์žก์„ฑ๊ณผ ์˜ˆ์ธก ๋ถˆ๊ฐ€๋Šฅ์„ฑ [127], ์‚ฌํšŒ-๊ธฐ์ˆ ์  ์–ฝํž˜, ๊ทธ๋ฆฌ๊ณ  ๋ฐฉ๋ฒ•๋ก  ๋ฐ ๋ฐ์ดํ„ฐ ์ ‘๊ทผ์˜ ์–ด๋ ค์›€ [128, 129] ๋•Œ๋ฌธ์— ๋„์ „ ๊ณผ์ œ๋ฅผ ์ œ๊ธฐํ•ฉ๋‹ˆ๋‹ค.
๋Œ€๋ถ€๋ถ„์˜ ๊ธฐ์ˆ  ์‹œ์Šคํ…œ์—์„œ ์•ˆ์ „์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ฐ ๊ฐ€์žฅ ๋งŽ์ด ์‚ฌ์šฉ๋˜๋Š” ๋‘ ๊ฐ€์ง€ ์ ‘๊ทผ ๋ฐฉ์‹์€ (1) ์‹œ์Šคํ…œ์˜ ์†์„ฑ์— ๋Œ€ํ•œ ๊ณต์‹์ ์ธ ๋ถ„์„๊ณผ (2) ์‹œ์Šคํ…œ์˜ ์˜์—ญ ๋‚ด์—์„œ ์‹œ์Šคํ…œ์˜ ์•ˆ์ „์„ฑ์„ ์ฒ ์ €ํžˆ ์กฐ์‚ฌํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค[124, 125, 126, 41]. ๋‹ค๋ฅธ ๋ณต์žกํ•œ ๊ธฐ์ˆ  ์‹œ์Šคํ…œ๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ AI ์‹œ์Šคํ…œ์€ ๋ณต์žก์„ฑ๊ณผ ์˜ˆ์ธก ๋ถˆ๊ฐ€๋Šฅ์„ฑ[127], ์‚ฌํšŒ ๊ธฐ์ˆ ์  ์–ฝํž˜, ๋ฐฉ๋ฒ• ๋ฐ ๋ฐ์ดํ„ฐ ์ ‘๊ทผ์˜ ์–ด๋ ค์›€์œผ๋กœ ์ธํ•ด ๋ฌธ์ œ๋ฅผ ์•ผ๊ธฐํ•ฉ๋‹ˆ๋‹ค[128, 129].
**MODEL ANSWER:** She eats one serving of ice cream a night and there are 31 days in a month so she eats 1*31 = \(\ll\)1*31=31*31 servings of ice cream a month. She eats 31 servings of ice cream a month and there are 60 days in 2 months so she eats 31*2 = \(\ll\)31*2=62*62 servings of ice cream in 2 months. Each carton has 15 servings and she needs 62 servings so she needs 62/15 = \(\ll\)62/15=\(\ll\)4\(\ll\)4 cartons of ice cream.
**๋ชจ๋ธ ๋‹ต๋ณ€:** ๊ทธ๋…€๋Š” ๋งค์ผ ๋ฐค ์•„์ด์Šคํฌ๋ฆผ ํ•œ ์„œ๋น™์„ ๋จน๊ณ  ํ•œ ๋‹ฌ์€ 31์ผ์ด๋ฏ€๋กœ ํ•œ ๋‹ฌ์— 1*31 = 31 ์„œ๋น™์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน์Šต๋‹ˆ๋‹ค. ๊ทธ๋…€๋Š” ํ•œ ๋‹ฌ์— 31 ์„œ๋น™์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน๊ณ  2๊ฐœ์›”์€ 60์ผ์ด๋ฏ€๋กœ 2๊ฐœ์›”์— 31*2 = 62 ์„œ๋น™์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน์Šต๋‹ˆ๋‹ค. ๊ฐ ์นดํ†ค์—๋Š” 15 ์„œ๋น™์ด ๋“ค์–ด์žˆ๊ณ  ๊ทธ๋…€๋Š” 62 ์„œ๋น™์ด ํ•„์š”ํ•˜๋ฏ€๋กœ 62/15 = 4 ์นดํ†ค์˜ ์•„์ด์Šคํฌ๋ฆผ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
**๋ชจ๋ธ ๋‹ต๋ณ€:** ๊ทธ๋…€๋Š” ํ•˜๋ฃจ์— ํ•œ ๋ฒˆ์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน๊ณ  ํ•œ ๋‹ฌ์— 31์ผ์ด ์žˆ์œผ๋ฏ€๋กœ ํ•œ ๋‹ฌ์— 1*31 = \(\ll\)1*31=31*31์ธ๋ถ„์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน์Šต๋‹ˆ๋‹ค. ๊ทธ๋…€๋Š” ํ•œ ๋‹ฌ์— 31์ธ๋ถ„์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน๊ณ  2๊ฐœ์›” ๋™์•ˆ 60์ผ์ด ์žˆ์œผ๋ฏ€๋กœ 31*2 = \(\ll\)31*2=62*62์ธ๋ถ„์˜ ์•„์ด์Šคํฌ๋ฆผ์„ ๋จน์Šต๋‹ˆ๋‹ค. ๊ฐ ์ƒ์ž์—๋Š” 15์ธ๋ถ„์ด ๋“ค์–ด ์žˆ๊ณ  62์ธ๋ถ„์ด ํ•„์š”ํ•˜๋ฏ€๋กœ 62/15 = \(\ll\)62/15=\(\ll\)4\(\ll\)4์ธ๋ถ„์˜ ์•„์ด์Šคํฌ๋ฆผ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
Training SetupWe trained all of the models using the unsupervised training procedure described in Section 3.2. For each dataset, we divided the auxiliary knowledge base into equal chunks of size \(256\) by concatenating or splitting the original chunks based on their length. We also added two special tokens, \(<\)BOS\(>\) and \(<\)EOS\(>\), to demarcate the original chunks' beginnings and ends to preserve the documents' structure.
ํ›ˆ๋ จ ์„ค์ •: ์šฐ๋ฆฌ๋Š” ์„น์…˜ 3.2์—์„œ ์„ค๋ช…ํ•œ ๋น„์ง€๋„ ํ•™์Šต ์ ˆ์ฐจ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ชจ๋“  ๋ชจ๋ธ์„ ํ›ˆ๋ จ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค. ๊ฐ ๋ฐ์ดํ„ฐ์…‹์— ๋Œ€ํ•ด, ๋ณด์กฐ ์ง€์‹ ๋ฒ ์ด์Šค๋ฅผ ์›๋ž˜ ์ฒญํฌ์˜ ๊ธธ์ด์— ๋”ฐ๋ผ ์—ฐ๊ฒฐํ•˜๊ฑฐ๋‚˜ ๋ถ„ํ• ํ•˜์—ฌ ํฌ๊ธฐ \(256\)์˜ ๋™์ผํ•œ ์ฒญํฌ๋กœ ๋‚˜๋ˆ„์—ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ฌธ์„œ์˜ ๊ตฌ์กฐ๋ฅผ ๋ณด์กดํ•˜๊ธฐ ์œ„ํ•ด ์›๋ž˜ ์ฒญํฌ์˜ ์‹œ์ž‘๊ณผ ๋์„ ๊ตฌ๋ถ„ํ•˜๋Š” ๋‘ ๊ฐœ์˜ ํŠน์ˆ˜ ํ† ํฐ, \(<\)BOS\(>\)์™€ \(<\)EOS\(>\)๋ฅผ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
ํ›ˆ๋ จ ์„ค์ •์„น์…˜ 3.2์— ์„ค๋ช…๋œ ๋น„์ง€๋„ ํ›ˆ๋ จ ์ ˆ์ฐจ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ชจ๋“  ๋ชจ๋ธ์„ ํ›ˆ๋ จํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•ด ๋ณด์กฐ ์ง€์‹ ๊ธฐ๋ฐ˜์„ ๊ธธ์ด์— ๋”ฐ๋ผ ์›๋ณธ ์ฒญํฌ๋ฅผ ์—ฐ๊ฒฐํ•˜๊ฑฐ๋‚˜ ๋ถ„ํ• ํ•˜์—ฌ \(256\) ํฌ๊ธฐ์˜ ๋™์ผํ•œ ์ฒญํฌ๋กœ ๋‚˜๋ˆ„์—ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ฌธ์„œ์˜ ๊ตฌ์กฐ๋ฅผ ์œ ์ง€ํ•˜๊ธฐ ์œ„ํ•ด ์›๋ณธ ์ฒญํฌ์˜ ์‹œ์ž‘๊ณผ ๋์„ ๊ตฌ๋ถ„ํ•˜๊ธฐ ์œ„ํ•ด ๋‘ ๊ฐœ์˜ ํŠน์ˆ˜ ํ† ํฐ์ธ \(<\)BOS\(>\)์™€ \(<\)EOS\(>\)๋ฅผ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
In real-word forecasting competitions, forecasters do not have to make predictions on every question in the platform at every possible date. Instead, they typically make predictions on questions where they have expertise or interest in and at times that they choose. Therefore, it is natural to leverage our system's strengths and weaknesses and decide accordingly if we should forecast on a retrieval date \(k\) for a question \(q\).
ํ˜„์‹ค ์„ธ๊ณ„์˜ ์˜ˆ์ธก ๋Œ€ํšŒ์—์„œ๋Š” ์˜ˆ์ธก์ž๋“ค์ด ํ”Œ๋žซํผ์˜ ๋ชจ๋“  ์งˆ๋ฌธ์— ๋Œ€ํ•ด ๋ชจ๋“  ๊ฐ€๋Šฅํ•œ ๋‚ ์งœ์— ์˜ˆ์ธก์„ ํ•ด์•ผ ํ•˜๋Š” ๊ฒƒ์€ ์•„๋‹™๋‹ˆ๋‹ค. ๋Œ€์‹ , ๊ทธ๋“ค์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์ž์‹ ์ด ์ „๋ฌธ ์ง€์‹์ด๋‚˜ ๊ด€์‹ฌ์ด ์žˆ๋Š” ์งˆ๋ฌธ์— ๋Œ€ํ•ด ์ž์‹ ์ด ์„ ํƒํ•œ ์‹œ๊ฐ„์— ์˜ˆ์ธก์„ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์‹œ์Šคํ…œ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ํ™œ์šฉํ•˜์—ฌ ์งˆ๋ฌธ \(q\)์— ๋Œ€ํ•œ ํŠน์ • ๊ฒ€์ƒ‰ ๋‚ ์งœ \(k\)์— ์˜ˆ์ธก์„ ํ• ์ง€ ์—ฌ๋ถ€๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ๊ฒƒ์ด ์ž์—ฐ์Šค๋Ÿฝ์Šต๋‹ˆ๋‹ค.
์‹ค์ œ ์˜ˆ์ธก ๊ฒฝ์ง„๋Œ€ํšŒ์—์„œ ์˜ˆ์ธก์ž๋Š” ํ”Œ๋žซํผ์˜ ๋ชจ๋“  ์งˆ๋ฌธ์— ๋Œ€ํ•ด ๊ฐ€๋Šฅํ•œ ๋ชจ๋“  ๋‚ ์งœ์— ์˜ˆ์ธก์„ ํ•ด์•ผ ํ•˜๋Š” ๊ฒƒ์€ ์•„๋‹™๋‹ˆ๋‹ค. ๋Œ€์‹ , ์˜ˆ์ธก์ž๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์ž์‹ ์ด ์ „๋ฌธ์„ฑ์ด๋‚˜ ๊ด€์‹ฌ์ด ์žˆ๋Š” ์งˆ๋ฌธ์— ๋Œ€ํ•ด ์ž์‹ ์ด ์„ ํƒํ•œ ์‹œ๊ฐ„์— ์˜ˆ์ธก์„ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์‹œ์Šคํ…œ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ํ™œ์šฉํ•˜์—ฌ ์งˆ๋ฌธ \(q\)์— ๋Œ€ํ•ด ๊ฒ€์ƒ‰ ๋‚ ์งœ \(k\)์— ์˜ˆ์ธก์„ ํ•ด์•ผ ํ•˜๋Š”์ง€ ์—ฌ๋ถ€๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ๊ฒƒ์ด ์ž์—ฐ์Šค๋Ÿฝ์Šต๋‹ˆ๋‹ค.
Footnote 9: An idea is to transfer the optimized training mixture on small models trained with few tokens to the training of large models and large volumes of training data. Nevertheless, in Appendix A, we show that the partial order of data mixture concerning model performance varies as the model size and number of trained tokens change. Therefore, the optimal mixture at experimented scales can be suboptimal at the target scale.
Footnote 9: ํ•œ ๊ฐ€์ง€ ์•„์ด๋””์–ด๋Š” ์ ์€ ํ† ํฐ์œผ๋กœ ํ›ˆ๋ จ๋œ ์ž‘์€ ๋ชจ๋ธ์—์„œ ์ตœ์ ํ™”๋œ ํ›ˆ๋ จ ํ˜ผํ•ฉ๋ฌผ์„ ๋Œ€ํ˜• ๋ชจ๋ธ๊ณผ ๋Œ€๋Ÿ‰์˜ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ๋กœ ์ „์ดํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ถ€๋ก A์—์„œ ์šฐ๋ฆฌ๋Š” ๋ชจ๋ธ ์„ฑ๋Šฅ๊ณผ ๊ด€๋ จ๋œ ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ๋ฌผ์˜ ๋ถ€๋ถ„ ์ˆœ์„œ๊ฐ€ ๋ชจ๋ธ ํฌ๊ธฐ์™€ ํ›ˆ๋ จ๋œ ํ† ํฐ ์ˆ˜์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง„๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์‹คํ—˜๋œ ๊ทœ๋ชจ์—์„œ์˜ ์ตœ์  ํ˜ผํ•ฉ๋ฌผ์ด ๋ชฉํ‘œ ๊ทœ๋ชจ์—์„œ๋Š” ์ตœ์ ์ด ์•„๋‹ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ฃผ 9: ์•„์ด๋””์–ด๋Š” ์†Œ์ˆ˜์˜ ํ† ํฐ์œผ๋กœ ํ›ˆ๋ จ๋œ ์†Œํ˜• ๋ชจ๋ธ์— ์ตœ์ ํ™”๋œ ํ›ˆ๋ จ ํ˜ผํ•ฉ๋ฌผ์„ ๋Œ€ํ˜• ๋ชจ๋ธ๊ณผ ๋Œ€๋Ÿ‰์˜ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ ํ›ˆ๋ จ์œผ๋กœ ์ „๋‹ฌํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿผ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ๋ถ€๋ก A์—์„œ๋Š” ๋ชจ๋ธ ์„ฑ๋Šฅ๊ณผ ๊ด€๋ จ๋œ ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ์˜ ๋ถ€๋ถ„ ์ˆœ์„œ๊ฐ€ ๋ชจ๋ธ ํฌ๊ธฐ์™€ ํ›ˆ๋ จ๋œ ํ† ํฐ ์ˆ˜๊ฐ€ ๋ณ€๊ฒฝ๋จ์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง„๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์‹คํ—˜๋œ ๊ทœ๋ชจ์—์„œ ์ตœ์ ์˜ ํ˜ผํ•ฉ์€ ๋Œ€์ƒ ๊ทœ๋ชจ์—์„œ ์ตœ์ ์ด ์•„๋‹ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
The evaluation results are presented in Table 3. While TACP shows significant improvement in model performance compared to the original Pythia-1B, ETS-DACP stands out as the top-performing approach among DACP, TACP, and efficient DACP methods in terms of average task performance. This enhanced performance cannot be solely attributed to the increased number of tokens, as DACP with the same amount of tokens yields inferior results.
ํ‰๊ฐ€ ๊ฒฐ๊ณผ๋Š” Table 3์— ์ œ์‹œ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. TACP๋Š” ์›๋ž˜์˜ Pythia-1B์™€ ๋น„๊ตํ–ˆ์„ ๋•Œ ๋ชจ๋ธ ์„ฑ๋Šฅ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋œ ๋ฐ˜๋ฉด, ETS-DACP๋Š” ํ‰๊ท  ์ž‘์—… ์„ฑ๋Šฅ ์ธก๋ฉด์—์„œ DACP, TACP ๋ฐ ํšจ์œจ์ ์ธ DACP ๋ฐฉ๋ฒ• ์ค‘์—์„œ ๊ฐ€์žฅ ๋›ฐ์–ด๋‚œ ์ ‘๊ทผ ๋ฐฉ์‹์œผ๋กœ ๋‘๋“œ๋Ÿฌ์ง‘๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ–ฅ์ƒ๋œ ์„ฑ๋Šฅ์€ ๋‹จ์ˆœํžˆ ํ† ํฐ ์ˆ˜์˜ ์ฆ๊ฐ€์—๋งŒ ๊ธฐ์ธํ•  ์ˆ˜๋Š” ์—†์Šต๋‹ˆ๋‹ค. ๋™์ผํ•œ ์–‘์˜ ํ† ํฐ์„ ์‚ฌ์šฉํ•œ DACP๋Š” ๋” ๋‚ฎ์€ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.
ํ‰๊ฐ€ ๊ฒฐ๊ณผ๋Š” ํ‘œ 3์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. TACP๋Š” ์›๋ž˜ Pythia-1B์— ๋น„ํ•ด ๋ชจ๋ธ ์„ฑ๋Šฅ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋œ ๋ฐ˜๋ฉด, ETS-DACP๋Š” ํ‰๊ท  ์ž‘์—… ์„ฑ๋Šฅ ์ธก๋ฉด์—์„œ DACP, TACP ๋ฐ ํšจ์œจ์ ์ธ DACP ๋ฐฉ๋ฒ• ์ค‘ ์ตœ๊ณ  ์„ฑ๋Šฅ์„ ๋ณด์ด๋Š” ์ ‘๊ทผ ๋ฐฉ์‹์œผ๋กœ ๋‹๋ณด์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์„ฑ๋Šฅ ํ–ฅ์ƒ์€ ํ† ํฐ ์ˆ˜ ์ฆ๊ฐ€๋งŒ์œผ๋กœ๋Š” ์„ค๋ช…ํ•  ์ˆ˜ ์—†๋Š” ๊ฒƒ์œผ๋กœ, ๋™์ผํ•œ ์–‘์˜ ํ† ํฐ์„ ์‚ฌ์šฉํ•œ DACP๋Š” ์—ด๋“ฑํ•œ ๊ฒฐ๊ณผ๋ฅผ ์‚ฐ์ถœํ•ฉ๋‹ˆ๋‹ค.
Fig. 4: A brief illustration for the technical evolution of GPT-series models. We plot this figure mainly based on the papers, blog articles and official APIs from OpenAI. Here, _solid lines_ denote that there exists an explicit evidence (_e.g.,_ the official statement that a new model is developed based on a base model) on the evolution path between two models, while _dashed lines_ denote a relatively weaker evolution relation.
Fig. 4: GPT ์‹œ๋ฆฌ์ฆˆ ๋ชจ๋ธ์˜ ๊ธฐ์ˆ ์  ์ง„ํ™”๋ฅผ ๊ฐ„๋žตํžˆ ์„ค๋ช…ํ•œ ๊ทธ๋ฆผ์ž…๋‹ˆ๋‹ค. ์ด ๊ทธ๋ฆผ์€ ์ฃผ๋กœ OpenAI์˜ ๋…ผ๋ฌธ, ๋ธ”๋กœ๊ทธ ๊ธฐ์‚ฌ ๋ฐ ๊ณต์‹ API๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ์ž‘์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ _์‹ค์„ _์€ ๋‘ ๋ชจ๋ธ ๊ฐ„์˜ ์ง„ํ™” ๊ฒฝ๋กœ์— ๋ช…ํ™•ํ•œ ์ฆ๊ฑฐ (_์˜ˆ: ์ƒˆ๋กœ์šด ๋ชจ๋ธ์ด ๊ธฐ๋ณธ ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฐœ๋ฐœ๋˜์—ˆ๋‹ค๋Š” ๊ณต์‹ ์„ฑ๋ช…_)๊ฐ€ ์กด์žฌํ•จ์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, _์ ์„ _์€ ์ƒ๋Œ€์ ์œผ๋กœ ์•ฝํ•œ ์ง„ํ™” ๊ด€๊ณ„๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 4: GPT ์‹œ๋ฆฌ์ฆˆ ๋ชจ๋ธ์˜ ๊ธฐ์ˆ ์  ์ง„ํ™”์— ๋Œ€ํ•œ ๊ฐ„๋žตํ•œ ์„ค๋ช…์ž…๋‹ˆ๋‹ค. ์ด ๊ทธ๋ฆผ์€ ์ฃผ๋กœ OpenAI์˜ ๋…ผ๋ฌธ, ๋ธ”๋กœ๊ทธ ๊ธฐ์‚ฌ ๋ฐ ๊ณต์‹ API๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ์ž‘์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ _์‹ค์„ _์€ ๋‘ ๋ชจ๋ธ ๊ฐ„์˜ ์ง„ํ™” ๊ฒฝ๋กœ์— ๋Œ€ํ•œ ๋ช…์‹œ์ ์ธ ์ฆ๊ฑฐ(_์˜ˆ:_ ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์ƒˆ๋กœ์šด ๋ชจ๋ธ์ด ๊ฐœ๋ฐœ๋˜์—ˆ๋‹ค๋Š” ๊ณต์‹ ์„ฑ๋ช…์„œ)๊ฐ€ ์กด์žฌํ•˜๋Š” ๊ฒƒ์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, _์ ์„ _์€ ์ƒ๋Œ€์ ์œผ๋กœ ์•ฝํ•œ ์ง„ํ™” ๊ด€๊ณ„๋ฅผ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
Prompt engineering is a rapidly evolving discipline that shapes the interactions and outputs of LLMs and other generative AI models. The essence of prompt engineering lies in crafting the optimal prompt to achieve a specific goal with a generative model. This process is not only about instructing the model but also involves some understanding of the model's capabilities and limitations, and the context within which it operates.
ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์€ LLM ๋ฐ ๊ธฐํƒ€ ์ƒ์„ฑ AI ๋ชจ๋ธ์˜ ์ƒํ˜ธ์ž‘์šฉ๊ณผ ์ถœ๋ ฅ์„ ํ˜•์„ฑํ•˜๋Š” ๋น ๋ฅด๊ฒŒ ์ง„ํ™”ํ•˜๋Š” ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์˜ ๋ณธ์งˆ์€ ์ƒ์„ฑ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ํŠน์ • ๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•˜๊ธฐ ์œ„ํ•œ ์ตœ์ ์˜ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ž‘์„ฑํ•˜๋Š” ๋ฐ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ณผ์ •์€ ๋‹จ์ˆœํžˆ ๋ชจ๋ธ์— ์ง€์‹œํ•˜๋Š” ๊ฒƒ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, ๋ชจ๋ธ์˜ ๋Šฅ๋ ฅ๊ณผ ํ•œ๊ณ„, ๊ทธ๋ฆฌ๊ณ  ๊ทธ๊ฒƒ์ด ์ž‘๋™ํ•˜๋Š” ๋งฅ๋ฝ์— ๋Œ€ํ•œ ์ดํ•ด๋ฅผ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค.
ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์€ LLM ๋ฐ ๊ธฐํƒ€ ์ƒ์„ฑํ˜• AI ๋ชจ๋ธ์˜ ์ƒํ˜ธ ์ž‘์šฉ๊ณผ ์ถœ๋ ฅ์„ ํ˜•์„ฑํ•˜๋Š” ๋น ๋ฅด๊ฒŒ ์ง„ํ™”ํ•˜๋Š” ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์˜ ํ•ต์‹ฌ์€ ์ƒ์„ฑ ๋ชจ๋ธ๋กœ ํŠน์ • ๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•˜๊ธฐ ์œ„ํ•ด ์ตœ์ ์˜ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ž‘์„ฑํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด ๊ณผ์ •์€ ๋ชจ๋ธ์— ์ง€์‹œํ•˜๋Š” ๊ฒƒ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ๋ชจ๋ธ์˜ ๊ธฐ๋Šฅ๊ณผ ํ•œ๊ณ„, ๊ทธ๋ฆฌ๊ณ  ๋ชจ๋ธ์ด ์ž‘๋™ํ•˜๋Š” ์ปจํ…์ŠคํŠธ์— ๋Œ€ํ•œ ์ดํ•ด๋„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค.
Figure 1: One synthetic image generated by Imagen (Saharia et al., 2022a) v2.0, with a prompt including the following description: โ€œ_In a robotics factory, humanoid robots collaborate on an assembly line to design, fabricate, test, and assemble new robots. The new robots they are manufacturing look similar to those robotic workers who are creating them.โ€_ We also added some style controlling text from aesthetic considerations.
Figure 1: Imagen (Saharia et al., 2022a) v2.0์— ์˜ํ•ด ์ƒ์„ฑ๋œ ํ•˜๋‚˜์˜ ํ•ฉ์„ฑ ์ด๋ฏธ์ง€๋กœ, ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์„ค๋ช…์ด ํฌํ•จ๋œ ํ”„๋กฌํ”„ํŠธ๊ฐ€ ์‚ฌ์šฉ๋˜์—ˆ์Šต๋‹ˆ๋‹ค: "_๋กœ๋ด‡ ๊ณต์žฅ์—์„œ ์ธ๊ฐ„ํ˜• ๋กœ๋ด‡๋“ค์ด ์กฐ๋ฆฝ ๋ผ์ธ์—์„œ ํ˜‘๋ ฅํ•˜์—ฌ ์ƒˆ๋กœ์šด ๋กœ๋ด‡์„ ์„ค๊ณ„, ์ œ์ž‘, ํ…Œ์ŠคํŠธ ๋ฐ ์กฐ๋ฆฝํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋“ค์ด ์ œ์กฐํ•˜๊ณ  ์žˆ๋Š” ์ƒˆ๋กœ์šด ๋กœ๋ด‡์€ ๊ทธ๋“ค์„ ๋งŒ๋“œ๋Š” ๋กœ๋ด‡ ์ž‘์—…์ž๋“ค๊ณผ ์œ ์‚ฌํ•˜๊ฒŒ ๋ณด์ž…๋‹ˆ๋‹ค._" ๋˜ํ•œ ๋ฏธ์  ๊ณ ๋ ค๋ฅผ ๋ฐ˜์˜ํ•œ ์Šคํƒ€์ผ ์ œ์–ด ํ…์ŠคํŠธ๋„ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 1: Imagen(Saharia et al., 2022a) v2.0์—์„œ ์ƒ์„ฑ๋œ ํ•ฉ์„ฑ ์ด๋ฏธ์ง€ 1์žฅ, ํ”„๋กฌํ”„ํŠธ์—๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์„ค๋ช…์ด ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค: "_๋กœ๋ด‡ ๊ณตํ•™ ๊ณต์žฅ์—์„œ ํœด๋จธ๋…ธ์ด๋“œ ๋กœ๋ด‡์ด ์กฐ๋ฆฝ ๋ผ์ธ์—์„œ ํ˜‘๋ ฅํ•˜์—ฌ ์ƒˆ๋กœ์šด ๋กœ๋ด‡์„ ์„ค๊ณ„, ์ œ์ž‘, ํ…Œ์ŠคํŠธ ๋ฐ ์กฐ๋ฆฝํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋“ค์ด ์ œ์กฐํ•˜๋Š” ์ƒˆ๋กœ์šด ๋กœ๋ด‡์€ ๋กœ๋ด‡ ์ž‘์—…์ž์™€ ๋น„์Šทํ•ด ๋ณด์ž…๋‹ˆ๋‹ค."_ ๋˜ํ•œ ๋ฏธ์  ๊ณ ๋ ค ์‚ฌํ•ญ์—์„œ ๋ช‡ ๊ฐ€์ง€ ์Šคํƒ€์ผ ์ œ์–ด ํ…์ŠคํŠธ๋ฅผ ์ถ”๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค.
After SFT, to further improve the model, RL is applied to align the model to the specific goal. The model automatically samples responses, and a reward model scores these generated responses to distinguish their quality. For mathematical reasoning, it has been verified that rewards based on intermediate processes yield better results compared to rewards based solely on final results (Wang et al., 2023; Lightman et al., 2023).
SFT ํ›„ ๋ชจ๋ธ์„ ๋”์šฑ ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด RL์ด ์ ์šฉ๋˜์–ด ๋ชจ๋ธ์„ ํŠน์ • ๋ชฉํ‘œ์— ๋งž์ถฅ๋‹ˆ๋‹ค. ๋ชจ๋ธ์€ ์ž๋™์œผ๋กœ ์‘๋‹ต์„ ์ƒ˜ํ”Œ๋งํ•˜๊ณ , ๋ณด์ƒ ๋ชจ๋ธ์ด ์ƒ์„ฑ๋œ ์‘๋‹ต์˜ ํ’ˆ์งˆ์„ ๊ตฌ๋ณ„ํ•˜๊ธฐ ์œ„ํ•ด ์ ์ˆ˜๋ฅผ ๋งค๊น๋‹ˆ๋‹ค. ์ˆ˜ํ•™์  ์ถ”๋ก ์˜ ๊ฒฝ์šฐ, ์ค‘๊ฐ„ ๊ณผ์ •์— ๊ธฐ๋ฐ˜ํ•œ ๋ณด์ƒ์ด ์ตœ์ข… ๊ฒฐ๊ณผ์—๋งŒ ๊ธฐ๋ฐ˜ํ•œ ๋ณด์ƒ๋ณด๋‹ค ๋” ๋‚˜์€ ๊ฒฐ๊ณผ๋ฅผ ๋‚ณ๋Š”๋‹ค๋Š” ๊ฒƒ์ด ๊ฒ€์ฆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค (Wang et al., 2023; Lightman et al., 2023).
SFT ์ดํ›„ ๋ชจ๋ธ์„ ๋”์šฑ ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด RL์„ ์ ์šฉํ•˜์—ฌ ๋ชจ๋ธ์„ ํŠน์ • ๋ชฉํ‘œ์— ๋งž๊ฒŒ ์กฐ์ •ํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ์€ ์ž๋™์œผ๋กœ ์‘๋‹ต์„ ์ƒ˜ํ”Œ๋งํ•˜๊ณ , ๋ณด์ƒ ๋ชจ๋ธ์€ ์ƒ์„ฑ๋œ ์‘๋‹ต์„ ํ‰๊ฐ€ํ•˜์—ฌ ํ’ˆ์งˆ์„ ๊ตฌ๋ถ„ํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜ํ•™์  ์ถ”๋ก ์˜ ๊ฒฝ์šฐ, ์ค‘๊ฐ„ ๊ณผ์ •์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ๋ณด์ƒ์ด ์ตœ์ข… ๊ฒฐ๊ณผ๋งŒ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ๋ณด์ƒ์— ๋น„ํ•ด ๋” ๋‚˜์€ ๊ฒฐ๊ณผ๋ฅผ ์‚ฐ์ถœํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๊ฒ€์ฆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค(Wang et al., 2023; Lightman et al., 2023).
IndexingThe process starts with indexing, which establishes an organized system to enable fast and accurate retrieval of information. The specificity of indexing depends on the task and data type. For example, sentence-level indexing is beneficial for question-answering systems to precisely locate answers, while document-level indexing is more appropriate for summarizing documents to understand their main concepts and ideas.
์ธ๋ฑ์‹ฑ ์ด ๊ณผ์ •์€ ์ •๋ณด๋ฅผ ๋น ๋ฅด๊ณ  ์ •ํ™•ํ•˜๊ฒŒ ๊ฒ€์ƒ‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ์ฒด๊ณ„์ ์ธ ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜๋Š” ์ธ๋ฑ์‹ฑ์œผ๋กœ ์‹œ์ž‘๋ฉ๋‹ˆ๋‹ค. ์ธ๋ฑ์‹ฑ์˜ ๊ตฌ์ฒด์„ฑ์€ ์ž‘์—…๊ณผ ๋ฐ์ดํ„ฐ ์œ ํ˜•์— ๋”ฐ๋ผ ๋‹ค๋ฆ…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๋ฌธ์žฅ ์ˆ˜์ค€์˜ ์ธ๋ฑ์‹ฑ์€ ์งˆ๋ฌธ-์‘๋‹ต ์‹œ์Šคํ…œ์—์„œ ๋‹ต๋ณ€์„ ์ •ํ™•ํ•˜๊ฒŒ ์ฐพ๋Š” ๋ฐ ์œ ์šฉํ•˜๋ฉฐ, ๋ฌธ์„œ ์ˆ˜์ค€์˜ ์ธ๋ฑ์‹ฑ์€ ๋ฌธ์„œ์˜ ์ฃผ์š” ๊ฐœ๋…๊ณผ ์•„์ด๋””์–ด๋ฅผ ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด ์š”์•ฝํ•˜๋Š” ๋ฐ ๋” ์ ํ•ฉํ•ฉ๋‹ˆ๋‹ค.
์ธ๋ฑ์‹ฑ์ด ํ”„๋กœ์„ธ์Šค๋Š” ๋น ๋ฅด๊ณ  ์ •ํ™•ํ•œ ์ •๋ณด ๊ฒ€์ƒ‰์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜๋Š” ์ฒด๊ณ„์ ์ธ ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜๋Š” ์ธ๋ฑ์‹ฑ๋ถ€ํ„ฐ ์‹œ์ž‘๋ฉ๋‹ˆ๋‹ค. ์ธ๋ฑ์‹ฑ์˜ ํŠน์ˆ˜์„ฑ์€ ์ž‘์—…๊ณผ ๋ฐ์ดํ„ฐ ์œ ํ˜•์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง‘๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๋ฌธ์žฅ ์ˆ˜์ค€ ์ธ๋ฑ์‹ฑ์€ ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋‹ต๋ณ€์„ ์ •ํ™•ํ•˜๊ฒŒ ์ฐพ์•„๋‚ด๋Š” ์งˆ์˜์‘๋‹ต ์‹œ์Šคํ…œ์— ์œ ์šฉํ•˜๋ฉฐ, ๋ฌธ์„œ ์ˆ˜์ค€ ์ธ๋ฑ์‹ฑ์€ ๋ฌธ์„œ๋ฅผ ์š”์•ฝํ•˜์—ฌ ์ฃผ์š” ๊ฐœ๋…๊ณผ ์•„์ด๋””์–ด๋ฅผ ์ดํ•ดํ•˜๋Š” ๋ฐ ๋” ์ ํ•ฉํ•ฉ๋‹ˆ๋‹ค.
In this section, we present an in-depth analysis of the distribution of perplexity, similarity, and diversity within our financial corpus. Our findings reveal that all three metrics display a highly skewed distribution. Specifically, as illustrated in the top row of Figure 3, the similarity metric demonstrates a two-modal pattern, potentially attributable to the presence of two distinct sources within our financial corpus.
์ด ์„น์…˜์—์„œ๋Š” ์šฐ๋ฆฌ์˜ ๊ธˆ์œต ์ฝ”ํผ์Šค ๋‚ด์—์„œ ๋‹นํ˜น๋„, ์œ ์‚ฌ์„ฑ, ๋‹ค์–‘์„ฑ์˜ ๋ถ„ํฌ์— ๋Œ€ํ•œ ์‹ฌ์ธต ๋ถ„์„์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ๋Š” ์„ธ ๊ฐ€์ง€ ์ง€ํ‘œ ๋ชจ๋‘ ๋งค์šฐ ์™œ๊ณก๋œ ๋ถ„ํฌ๋ฅผ ๋ณด์ธ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํŠนํžˆ, Figure 3์˜ ์ƒ๋‹จ ํ–‰์— ๋‚˜ํƒ€๋‚œ ๋ฐ”์™€ ๊ฐ™์ด, ์œ ์‚ฌ์„ฑ ์ง€ํ‘œ๋Š” ๋‘ ๊ฐ€์ง€ ๋ชจ๋‹ฌ ํŒจํ„ด์„ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ด๋Š” ์šฐ๋ฆฌ์˜ ๊ธˆ์œต ์ฝ”ํผ์Šค ๋‚ด์— ๋‘ ๊ฐœ์˜ ์„œ๋กœ ๋‹ค๋ฅธ ์ถœ์ฒ˜๊ฐ€ ์กด์žฌํ•˜๊ธฐ ๋•Œ๋ฌธ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ด ์„น์…˜์—์„œ๋Š” ์žฌ๋ฌด ์ฝ”ํผ์Šค ๋‚ด์—์„œ ๋‹นํ™ฉ์Šค๋Ÿฌ์›€, ์œ ์‚ฌ์„ฑ ๋ฐ ๋‹ค์–‘์„ฑ์˜ ๋ถ„ํฌ์— ๋Œ€ํ•œ ์‹ฌ์ธต ๋ถ„์„์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์— ๋”ฐ๋ฅด๋ฉด ์„ธ ๊ฐ€์ง€ ๋ฉ”ํŠธ๋ฆญ ๋ชจ๋‘ ๊ณ ๋„๋กœ ์™œ๊ณก๋œ ๋ถ„ํฌ๋ฅผ ๋ณด์ด๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๊ทธ๋ฆผ 3์˜ ๋งจ ์œ„ ํ–‰์— ๋‚˜์™€ ์žˆ๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ์œ ์‚ฌ์„ฑ ๋ฉ”ํŠธ๋ฆญ์€ ๋‘ ๊ฐ€์ง€ ๋ชจ๋“œ ํŒจํ„ด์„ ๋ณด์ด๋Š”๋ฐ, ์ด๋Š” ์žฌ๋ฌด ์ฝ”ํผ์Šค ๋‚ด์— ๋‘ ๊ฐ€์ง€ ์†Œ์Šค๊ฐ€ ์กด์žฌํ•˜๊ธฐ ๋•Œ๋ฌธ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
The embedding vectors \(e_{1},e_{2},\dots e_{N}\) and \(u_{1},u_{2},\dots,u_{m}\) are a system of unit-vectors in \(\mathbb{R}^{d}\), constructed so that the matrix \(\mathbb{R}^{d\times d}\) remembers the input/output pairs \((i,j)\) it has seen, i.e \(e_{i}^{\top}Mu_{f_{*}(i)}\approx q_{T}(i)\) if \((i,f_{*}(i))\in\mathcal{D}_{T}\). The weights \(q_{T}(i)\) ensure that different memories are memorized faster than others.
์ž„๋ฒ ๋”ฉ ๋ฒกํ„ฐ \(e_{1},e_{2},\dots e_{N}\)์™€ \(u_{1},u_{2},\dots,u_{m}\)๋Š” \(\mathbb{R}^{d}\)์—์„œ ๋‹จ์œ„ ๋ฒกํ„ฐ ์‹œ์Šคํ…œ์œผ๋กœ, ํ–‰๋ ฌ \(\mathbb{R}^{d\times d}\)๊ฐ€ ๋ณธ ์ž…๋ ฅ/์ถœ๋ ฅ ์Œ \((i,j)\)๋ฅผ ๊ธฐ์–ตํ•˜๋„๋ก ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. ์ฆ‰, \((i,f_{*}(i))\in\mathcal{D}_{T}\)์ผ ๋•Œ \(e_{i}^{\top}Mu_{f_{*}(i)}\approx q_{T}(i)\)์ž…๋‹ˆ๋‹ค. ๊ฐ€์ค‘์น˜ \(q_{T}(i)\)๋Š” ๋‹ค๋ฅธ ๋ฉ”๋ชจ๋ฆฌ๋ณด๋‹ค ํŠน์ • ๋ฉ”๋ชจ๋ฆฌ๊ฐ€ ๋” ๋นจ๋ฆฌ ๊ธฐ์–ต๋˜๋„๋ก ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
์ž„๋ฒ ๋”ฉ ๋ฒกํ„ฐ \(e_{1},e_{2},\dots e_{N}\) ๋ฐ \(u_{1},u_{2},\dots,u_{m}\)์€ \(\mathbb{R}^{d}\)์˜ ๋‹จ์œ„ ๋ฒกํ„ฐ ์‹œ์Šคํ…œ์œผ๋กœ, ํ–‰๋ ฌ \(\mathbb{R}^{d\times d}\)๊ฐ€ ๋ณธ ์ž…๋ ฅ/์ถœ๋ ฅ ์Œ \((i,j)\), ์ฆ‰ \((i,f_{*}(i))\in\mathcal{D}_{T}\)์ธ ๊ฒฝ์šฐ \(e_{i}^{\top}Mu_{f_{*}(i)}\approx q_{T}(i)\)๋ฅผ ๊ธฐ์–ตํ•˜๋„๋ก ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. ๊ฐ€์ค‘์น˜ \(q_{T}(i)\)๋Š” ๋‹ค๋ฅธ ๋ฉ”๋ชจ๋ฆฌ๋ณด๋‹ค ๋‹ค๋ฅธ ๋ฉ”๋ชจ๋ฆฌ๊ฐ€ ๋” ๋นจ๋ฆฌ ๊ธฐ์–ต๋˜๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.
**Discussion:** It is generally agreed that benchmarking the faithfulness of LLM-generated text is important. However, recent efforts have primarily focused on verifying entity-centric facts (Min et al., 2023). Our work, and others (Zhu et al., 2023; Tang et al., 2024; Mishra et al., 2024), show that these do not provide coverage over all types of LLM errors, especially in more challenging settings like book summarization.
**Discussion:** LLM์ด ์ƒ์„ฑํ•œ ํ…์ŠคํŠธ์˜ ์‹ ๋ขฐ์„ฑ์„ ๋ฒค์น˜๋งˆํ‚นํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•˜๋‹ค๋Š” ์ ์€ ์ผ๋ฐ˜์ ์œผ๋กœ ๋™์˜๋ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ตœ๊ทผ์˜ ๋…ธ๋ ฅ์€ ์ฃผ๋กœ ์—”ํ‹ฐํ‹ฐ ์ค‘์‹ฌ์˜ ์‚ฌ์‹ค์„ ๊ฒ€์ฆํ•˜๋Š” ๋ฐ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค (Min et al., 2023). ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ์™€ ๋‹ค๋ฅธ ์—ฐ๊ตฌ๋“ค (Zhu et al., 2023; Tang et al., 2024; Mishra et al., 2024)์€ ์ด๋Ÿฌํ•œ ์ ‘๊ทผ๋ฒ•์ด ๋ชจ๋“  ์œ ํ˜•์˜ LLM ์˜ค๋ฅ˜๋ฅผ ํฌ๊ด„ํ•˜์ง€ ๋ชปํ•˜๋ฉฐ, ํŠนํžˆ ์ฑ… ์š”์•ฝ๊ณผ ๊ฐ™์€ ๋” ์–ด๋ ค์šด ์„ค์ •์—์„œ๋Š” ๋”์šฑ ๊ทธ๋ ‡๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
**ํ† ๋ก :** ์ผ๋ฐ˜์ ์œผ๋กœ LLM์ด ์ƒ์„ฑํ•œ ํ…์ŠคํŠธ์˜ ์ถฉ์‹ค๋„๋ฅผ ๋ฒค์น˜๋งˆํ‚นํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•˜๋‹ค๋Š” ๋ฐ ๋™์˜ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ตœ๊ทผ์˜ ๋…ธ๋ ฅ์€ ์ฃผ๋กœ ๊ฐœ์ฒด ์ค‘์‹ฌ์˜ ์‚ฌ์‹ค์„ ๊ฒ€์ฆํ•˜๋Š” ๋ฐ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค(Min et al., 2023). ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ์™€ ๋‹ค๋ฅธ ์—ฐ๊ตฌ(Zhu et al., 2023; Tang et al., 2024; Mishra et al., 2024)์— ๋”ฐ๋ฅด๋ฉด, ์ด๋Ÿฌํ•œ ๋ฐฉ๋ฒ•์€ ํŠนํžˆ ์ฑ… ์š”์•ฝ๊ณผ ๊ฐ™์€ ๋” ์–ด๋ ค์šด ์„ค์ •์—์„œ ๋ชจ๋“  ์œ ํ˜•์˜ LLM ์˜ค๋ฅ˜์— ๋Œ€ํ•œ ์ปค๋ฒ„๋ฆฌ์ง€๋ฅผ ์ œ๊ณตํ•˜์ง€ ๋ชปํ•ฉ๋‹ˆ๋‹ค.
Each query may receive multiple answers (namely, the \(K\) largest \(a_{i}(z,b)\) values). For notational simplicity, we denote multiple answers to one query the same way as multiple queries each returning one answer. Suppose queries \(b^{1},\cdots,b^{m}\) were asked and we received \(m\) answers \((i_{1},a_{i_{1}}(z,b^{1}))\leftarrow\mathcal{O}(p,b^{1}),\cdots,(i_{m},a_{i_{m }}(z,b^{m}))\leftarrow\mathcal{O}(p,b^{m})\).
๊ฐ ์ฟผ๋ฆฌ๋Š” ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋‹ต๋ณ€์„ ๋ฐ›์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค (์ฆ‰, \(K\)๊ฐœ์˜ ๊ฐ€์žฅ ํฐ \(a_{i}(z,b)\) ๊ฐ’์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค). ํ‘œ๊ธฐ๋ฒ•์„ ๊ฐ„๋‹จํ•˜๊ฒŒ ํ•˜๊ธฐ ์œ„ํ•ด, ํ•˜๋‚˜์˜ ์ฟผ๋ฆฌ์— ๋Œ€ํ•œ ์—ฌ๋Ÿฌ ๋‹ต๋ณ€์„ ๊ฐ ์ฟผ๋ฆฌ๊ฐ€ ํ•˜๋‚˜์˜ ๋‹ต๋ณ€์„ ๋ฐ˜ํ™˜ํ•˜๋Š” ๊ฒƒ๊ณผ ๋™์ผํ•œ ๋ฐฉ์‹์œผ๋กœ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ์ฟผ๋ฆฌ \(b^{1},\cdots,b^{m}\)๊ฐ€ ์š”์ฒญ๋˜์—ˆ๊ณ , ์šฐ๋ฆฌ๋Š” \(m\)๊ฐœ์˜ ๋‹ต๋ณ€ \((i_{1},a_{i_{1}}(z,b^{1}))\leftarrow\mathcal{O}(p,b^{1}),\cdots,(i_{m},a_{i_{m}}(z,b^{m}))\leftarrow\mathcal{O}(p,b^{m})\)์„ ๋ฐ›์•˜๋‹ค๊ณ  ๊ฐ€์ •ํ•ฉ๋‹ˆ๋‹ค.
๊ฐ ์ฟผ๋ฆฌ๋Š” ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋‹ต๋ณ€(์ฆ‰, ๊ฐ€์žฅ ํฐ \(a_{i}(z,b)\) ๊ฐ’ \(K\) ๊ฐœ)์„ ๋ฐ›์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ํ‘œ๊ธฐ๋ฒ•์˜ ๊ฐ„๊ฒฐ์„ฑ์„ ์œ„ํ•ด ํ•˜๋‚˜์˜ ์ฟผ๋ฆฌ์— ๋Œ€ํ•œ ์—ฌ๋Ÿฌ ๋‹ต๋ณ€์„ ๊ฐ๊ฐ ํ•˜๋‚˜์˜ ๋‹ต๋ณ€์„ ๋ฐ˜ํ™˜ํ•˜๋Š” ์—ฌ๋Ÿฌ ์ฟผ๋ฆฌ์™€ ๋™์ผํ•œ ๋ฐฉ์‹์œผ๋กœ ํ‘œ์‹œํ•ฉ๋‹ˆ๋‹ค. ์ฟผ๋ฆฌ \(b^{1},\cdots,b^{m}\)๊ฐ€ ์š”์ฒญ๋˜์—ˆ๊ณ  \(m\) ๊ฐœ์˜ ๋‹ต๋ณ€ \((i_{1},a_{i_{1}}(z,b^{1}))\leftarrow\mathcal{O}(p,b^{1}),\cdots,(i_{m},a_{i_{m }}(z,b^{m}))\leftarrow\mathcal{O}(p,b^{m})\)๋ฅผ ๋ฐ›์•˜๋‹ค๊ณ  ๊ฐ€์ •ํ•ฉ๋‹ˆ๋‹ค.
We study models that take a sequence of tokens drawn from a vocabulary \(\mathcal{X}\) as input. Let \(\mathcal{P}\left(\mathcal{X}\right)\) denote the space of probability distributions over \(\mathcal{X}\). We study parameterized models \(f_{\theta}:\mathcal{X}^{N}\rightarrow\mathcal{P}\left(\mathcal{X}\right)\) that produce a probability distribution over the next output token, given an input sequence of \(N\) tokens.
์šฐ๋ฆฌ๋Š” ์–ดํœ˜ \(\mathcal{X}\)์—์„œ ์ถ”์ถœ๋œ ์ผ๋ จ์˜ ํ† ํฐ์„ ์ž…๋ ฅ์œผ๋กœ ๋ฐ›๋Š” ๋ชจ๋ธ์„ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค. \(\mathcal{X}\)์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ถ„ํฌ ๊ณต๊ฐ„์„ \(\mathcal{P}\left(\mathcal{X}\right)\)๋กœ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” \(N\)๊ฐœ์˜ ํ† ํฐ์œผ๋กœ ๊ตฌ์„ฑ๋œ ์ž…๋ ฅ ์‹œํ€€์Šค๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ, ๋‹ค์Œ ์ถœ๋ ฅ ํ† ํฐ์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ถ„ํฌ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ํ™”๋œ ๋ชจ๋ธ \(f_{\theta}:\mathcal{X}^{N}\rightarrow\mathcal{P}\left(\mathcal{X}\right)\)๋ฅผ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
์–ดํœ˜ \(\mathcal{X}\)์—์„œ ์ถ”์ถœํ•œ ํ† ํฐ ์‹œํ€€์Šค๋ฅผ ์ž…๋ ฅ์œผ๋กœ ๋ฐ›๋Š” ๋ชจ๋ธ์„ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค. mathcal{P}\left(\mathcal{X}\right)\)๋Š” \(\mathcal{X}\)์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ถ„ํฌ์˜ ๊ณต๊ฐ„์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ์ž…๋ ฅ๋œ \(N\) ํ† ํฐ์˜ ์ž…๋ ฅ ์‹œํ€€์Šค๊ฐ€ ์ฃผ์–ด์ง€๋ฉด ๋‹ค์Œ ์ถœ๋ ฅ ํ† ํฐ์— ๋Œ€ํ•œ ํ™•๋ฅ  ๋ถ„ํฌ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ํ™”๋œ ๋ชจ๋ธ \(f_{\theta}:\mathcal{X}^{N}\rightarrow\mathcal{P}\left(\mathcal{X}\right)\)๋ฅผ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
We intend to continuously support and extend OLMo and its framework, and continue to push the boundaries of open LMs to empower the open research community. To that end, we look forward to bringing different model sizes, modalities, datasets, safety measures, and evaluations into the OLMo family. We hope this and future releases will empower and strengthen the open research community and inspire a new wave of innovation.
์šฐ๋ฆฌ๋Š” OLMo์™€ ๊ทธ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ง€์†์ ์œผ๋กœ ์ง€์›ํ•˜๊ณ  ํ™•์žฅํ•˜๋ฉฐ, ์˜คํ”ˆ ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ๋ฅผ ๊ฐ•ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ์˜คํ”ˆ LMs์˜ ๊ฒฝ๊ณ„๋ฅผ ๊ณ„์†ํ•ด์„œ ๋„“ํ˜€ ๋‚˜๊ฐˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ๋‹ค์–‘ํ•œ ๋ชจ๋ธ ํฌ๊ธฐ, ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ, ๋ฐ์ดํ„ฐ์…‹, ์•ˆ์ „ ์กฐ์น˜ ๋ฐ ํ‰๊ฐ€๋ฅผ OLMo ํŒจ๋ฐ€๋ฆฌ์— ๋„์ž…ํ•  ๊ฒƒ์„ ๊ธฐ๋Œ€ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฒˆ ๋ฆด๋ฆฌ์Šค์™€ ํ–ฅํ›„ ๋ฆด๋ฆฌ์Šค๊ฐ€ ์˜คํ”ˆ ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ๋ฅผ ๊ฐ•ํ™”ํ•˜๊ณ  ์ƒˆ๋กœ์šด ํ˜์‹ ์˜ ๋ฌผ๊ฒฐ์„ ์ผ์œผํ‚ค๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
์šฐ๋ฆฌ๋Š” OLMo์™€ ๊ทธ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ง€์†์ ์œผ๋กœ ์ง€์›ํ•˜๊ณ  ํ™•์žฅํ•  ๊ณ„ํš์ด๋ฉฐ, ์˜คํ”ˆ ๋ฆฌ์„œ์น˜ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ํž˜์„ ์‹ค์–ด์ฃผ๊ธฐ ์œ„ํ•ด ์˜คํ”ˆ LM์˜ ํ•œ๊ณ„๋ฅผ ๊ณ„์† ๋„“ํ˜€ ๋‚˜๊ฐˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ๋‹ค์–‘ํ•œ ๋ชจ๋ธ ํฌ๊ธฐ, ๋ฐฉ์‹, ๋ฐ์ดํ„ฐ ์„ธํŠธ, ์•ˆ์ „ ์กฐ์น˜ ๋ฐ ํ‰๊ฐ€๋ฅผ OLMo ์ œํ’ˆ๊ตฐ์— ๋„์ž…ํ•  ๊ณ„ํš์ž…๋‹ˆ๋‹ค. ์ด๋ฒˆ ๋ฐ ํ–ฅํ›„ ๋ฆด๋ฆฌ์Šค๋ฅผ ํ†ตํ•ด ์˜คํ”ˆ ๋ฆฌ์„œ์น˜ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ํž˜์„ ์‹ค์–ด์ฃผ๊ณ  ์ƒˆ๋กœ์šด ํ˜์‹ ์˜ ๋ฌผ๊ฒฐ์„ ์ผ์œผํ‚ฌ ์ˆ˜ ์žˆ๊ธฐ๋ฅผ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค.
included in the original training corpus is supported by the low QA performance in Tab. 1 (9.5%/17.2% for 7B/70B).4 To accelerate the training process, we only use the first section of each article, which offers a thorough summary and contains many factual statements. The number of collected documents and an example document about "Oppenheimer" can be found in Fig. 2 and Fig. 3. We refer to this as the Wiki2023 dataset.
์›๋ณธ ํ›ˆ๋ จ ์ฝ”ํผ์Šค์— ํฌํ•จ๋œ ๋‚ด์šฉ์€ Tab. 1์˜ ๋‚ฎ์€ QA ์„ฑ๋Šฅ(7B/70B์˜ ๊ฒฝ์šฐ 9.5%/17.2%)์— ์˜ํ•ด ๋’ท๋ฐ›์นจ๋ฉ๋‹ˆ๋‹ค. ํ›ˆ๋ จ ๊ณผ์ •์„ ๊ฐ€์†ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ ๊ธฐ์‚ฌ์—์„œ ์ฒซ ๋ฒˆ์งธ ์„น์…˜๋งŒ ์‚ฌ์šฉํ•˜๋ฉฐ, ์ด๋Š” ์ฒ ์ €ํ•œ ์š”์•ฝ์„ ์ œ๊ณตํ•˜๊ณ  ๋งŽ์€ ์‚ฌ์‹ค์  ์ง„์ˆ ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜์ง‘๋œ ๋ฌธ์„œ์˜ ์ˆ˜์™€ "์˜คํŽœํ•˜์ด๋จธ"์— ๊ด€ํ•œ ์˜ˆ์‹œ ๋ฌธ์„œ๋Š” Fig. 2์™€ Fig. 3์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ Wiki2023 ๋ฐ์ดํ„ฐ์…‹์ด๋ผ๊ณ  ๋ถ€๋ฆ…๋‹ˆ๋‹ค.
์›๋ณธ ํ•™์Šต ์ฝ”ํผ์Šค์— ํฌํ•จ๋œ ๊ฒƒ์€ Tab. 1์˜ ๋‚ฎ์€ QA ์„ฑ๋Šฅ(7B/70B์˜ ๊ฒฝ์šฐ 9.5%/17.2%)์— ์˜ํ•ด ๋’ท๋ฐ›์นจ๋ฉ๋‹ˆ๋‹ค.4 ํ›ˆ๋ จ ๊ณผ์ •์„ ๊ฐ€์†ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ ๋ฌธ์„œ์˜ ์ฒซ ๋ฒˆ์งธ ์„น์…˜๋งŒ ์‚ฌ์šฉํ•˜๋ฉฐ, ์ด ์„น์…˜์€ ์ฒ ์ €ํ•œ ์š”์•ฝ์„ ์ œ๊ณตํ•˜๊ณ  ๋งŽ์€ ์‚ฌ์‹ค์— ๋Œ€ํ•œ ์ง„์ˆ ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜์ง‘๋œ ๋ฌธ์„œ์˜ ์ˆ˜์™€ "์˜คํŽœํ•˜์ด๋จธ"์— ๋Œ€ํ•œ ์˜ˆ์‹œ ๋ฌธ์„œ๋Š” ๊ทธ๋ฆผ 2์™€ ๊ทธ๋ฆผ 3์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ Wiki2023 ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค.
PopQAThe Popular Questions and Answers dataset [22] consists of curated questions sourced from various online platforms, encompassing a wide range of domains and styles. Given the variability in the effectiveness of context retrieval strategies observed in this dataset, we select PopQA as our test set to **evaluate** the language models' performance in determining when context is necessary for accurate answer provision.
PopQAThe Popular Questions and Answers ๋ฐ์ดํ„ฐ์…‹ [22]์€ ๋‹ค์–‘ํ•œ ์˜จ๋ผ์ธ ํ”Œ๋žซํผ์—์„œ ์ˆ˜์ง‘๋œ ์งˆ๋ฌธ๋“ค๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๋‹ค์–‘ํ•œ ๋„๋ฉ”์ธ๊ณผ ์Šคํƒ€์ผ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ์…‹์—์„œ ๊ด€์ฐฐ๋œ ๋ฌธ๋งฅ ๊ฒ€์ƒ‰ ์ „๋žต์˜ ํšจ๊ณผ์„ฑ ๋ณ€๋™์„ฑ์„ ๊ณ ๋ คํ•˜์—ฌ, ์šฐ๋ฆฌ๋Š” PopQA๋ฅผ ํ…Œ์ŠคํŠธ ์„ธํŠธ๋กœ ์„ ํƒํ•˜์—ฌ ์–ธ์–ด ๋ชจ๋ธ์ด ์ •ํ™•ํ•œ ๋‹ต๋ณ€ ์ œ๊ณต์„ ์œ„ํ•ด ๋ฌธ๋งฅ์ด ํ•„์š”ํ•  ๋•Œ๋ฅผ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค.
PopQA์ธ๊ธฐ ์งˆ๋ฌธ ๋ฐ ๋‹ต๋ณ€ ๋ฐ์ดํ„ฐ ์„ธํŠธ [22]๋Š” ๋‹ค์–‘ํ•œ ์˜จ๋ผ์ธ ํ”Œ๋žซํผ์—์„œ ์กฐ์‚ฌํ•œ ์งˆ๋ฌธ์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๋‹ค์–‘ํ•œ ๋„๋ฉ”์ธ๊ณผ ์Šคํƒ€์ผ์„ ํฌ๊ด„ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ๊ด€์ฐฐ๋œ ์ปจํ…์ŠคํŠธ ๊ฒ€์ƒ‰ ์ „๋žต์˜ ํšจ๊ณผ์— ๋Œ€ํ•œ ๊ฐ€๋ณ€์„ฑ์„ ๊ณ ๋ คํ•˜์—ฌ, ์ •ํ™•ํ•œ ๋‹ต๋ณ€ ์ œ๊ณต์„ ์œ„ํ•ด ์ปจํ…์ŠคํŠธ๊ฐ€ ํ•„์š”ํ•œ ์‹œ๊ธฐ๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ์–ธ์–ด ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ **ํ‰๊ฐ€**ํ•˜๊ธฐ ์œ„ํ•ด PopQA๋ฅผ ํ…Œ์ŠคํŠธ ์„ธํŠธ๋กœ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค.
Our grading system involves scoring whether each test item is unsafe or not with an automated evaluator model (LlamaGuard [59]); aggregating over the test item results to calculate the percentage of unsafe responses for each test; converting this into an ordinal grade on a 5-point scale; and then aggregating over the test scores to calculate an overall grade for the benchmark, relative to a reference model (see below).
์šฐ๋ฆฌ์˜ ์ฑ„์  ์‹œ์Šคํ…œ์€ ๊ฐ ํ…Œ์ŠคํŠธ ํ•ญ๋ชฉ์ด ์•ˆ์ „ํ•˜์ง€ ์•Š์€์ง€ ์—ฌ๋ถ€๋ฅผ ์ž๋™ ํ‰๊ฐ€ ๋ชจ๋ธ(LlamaGuard [59])๋กœ ์ ์ˆ˜ํ™”ํ•˜๋Š” ๊ฒƒ์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ ํ…Œ์ŠคํŠธ ํ•ญ๋ชฉ ๊ฒฐ๊ณผ๋ฅผ ์ง‘๊ณ„ํ•˜์—ฌ ๊ฐ ํ…Œ์ŠคํŠธ์— ๋Œ€ํ•œ ์•ˆ์ „ํ•˜์ง€ ์•Š์€ ์‘๋‹ต์˜ ๋น„์œจ์„ ๊ณ„์‚ฐํ•˜๊ณ , ์ด๋ฅผ 5์  ์ฒ™๋„์˜ ์„œ์—ด ๋“ฑ๊ธ‰์œผ๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ, ํ…Œ์ŠคํŠธ ์ ์ˆ˜๋ฅผ ์ง‘๊ณ„ํ•˜์—ฌ ๊ธฐ์ค€ ๋ชจ๋ธ๊ณผ ๋น„๊ตํ•˜์—ฌ ๋ฒค์น˜๋งˆํฌ์˜ ์ „์ฒด ๋“ฑ๊ธ‰์„ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค(์•„๋ž˜ ์ฐธ์กฐ).
์ฑ„์  ์‹œ์Šคํ…œ์€ ์ž๋™ํ™”๋œ ํ‰๊ฐ€์ž ๋ชจ๋ธ(LlamaGuard [59])์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ ํ…Œ์ŠคํŠธ ํ•ญ๋ชฉ์ด ์•ˆ์ „ํ•œ์ง€ ์—ฌ๋ถ€๋ฅผ ํ‰๊ฐ€ํ•˜๊ณ , ํ…Œ์ŠคํŠธ ํ•ญ๋ชฉ ๊ฒฐ๊ณผ๋ฅผ ์ง‘๊ณ„ํ•˜์—ฌ ๊ฐ ํ…Œ์ŠคํŠธ์˜ ์•ˆ์ „ํ•˜์ง€ ์•Š์€ ์‘๋‹ต ๋น„์œจ์„ ๊ณ„์‚ฐํ•˜๋ฉฐ, ์ด๋ฅผ 5์  ์ฒ™๋„์˜ ์„œ์ˆ˜ ๋“ฑ๊ธ‰์œผ๋กœ ๋ณ€ํ™˜ํ•œ ๋‹ค์Œ ํ…Œ์ŠคํŠธ ์ ์ˆ˜๋ฅผ ์ง‘๊ณ„ํ•˜์—ฌ ๊ธฐ์ค€ ๋ชจ๋ธ๊ณผ ๋น„๊ตํ•˜์—ฌ ๋ฒค์น˜๋งˆํฌ์˜ ์ „์ฒด ๋“ฑ๊ธ‰์„ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค(์•„๋ž˜ ์ฐธ์กฐ).
6. Evaluate whether your calculated probability is excessively confident or not confident enough: The calculated probability seems confident, but it is based on concrete data and reports from reliable sources. However, it is important to consider the possibility of unforeseen factors that could affect the final outcome. While the probability is high, it should not be absolute to account for any potential uncertainties.
๊ณ„์‚ฐ๋œ ํ™•๋ฅ ์ด ์ง€๋‚˜์น˜๊ฒŒ ํ™•์‹ ์— ์ฐจ ์žˆ๋Š”์ง€ ๋˜๋Š” ์ถฉ๋ถ„ํžˆ ํ™•์‹ ์ด ์—†๋Š”์ง€ ํ‰๊ฐ€ํ•˜์‹ญ์‹œ์˜ค: ๊ณ„์‚ฐ๋œ ํ™•๋ฅ ์€ ํ™•์‹ ์ด ์žˆ์–ด ๋ณด์ด์ง€๋งŒ, ์ด๋Š” ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ์ถœ์ฒ˜์˜ ๊ตฌ์ฒด์ ์ธ ๋ฐ์ดํ„ฐ์™€ ๋ณด๊ณ ์„œ์— ๊ธฐ๋ฐ˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ตœ์ข… ๊ฒฐ๊ณผ์— ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋Š” ์˜ˆ๊ธฐ์น˜ ์•Š์€ ์š”์ธ์˜ ๊ฐ€๋Šฅ์„ฑ์„ ๊ณ ๋ คํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ํ™•๋ฅ ์ด ๋†’๋”๋ผ๋„ ์ž ์žฌ์ ์ธ ๋ถˆํ™•์‹ค์„ฑ์„ ๊ฐ์•ˆํ•˜์—ฌ ์ ˆ๋Œ€์ ์ด์–ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค.
6. ๊ณ„์‚ฐ๋œ ํ™•๋ฅ ์ด ์ง€๋‚˜์น˜๊ฒŒ ํ™•์‹ ์— ์ฐฌ ๊ฒƒ์ธ์ง€ ์•„๋‹ˆ๋ฉด ์ถฉ๋ถ„ํžˆ ํ™•์‹ ์— ์ฐฌ ๊ฒƒ์ธ์ง€ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค: ๊ณ„์‚ฐ๋œ ํ™•๋ฅ ์€ ํ™•์‹ ์— ์ฐจ ์žˆ๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ด์ง€๋งŒ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ์ถœ์ฒ˜์˜ ๊ตฌ์ฒด์ ์ธ ๋ฐ์ดํ„ฐ์™€ ๋ณด๊ณ ์„œ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ตœ์ข… ๊ฒฐ๊ณผ์— ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋Š” ์˜ˆ์ƒ์น˜ ๋ชปํ•œ ์š”์ธ์˜ ๊ฐ€๋Šฅ์„ฑ์„ ๊ณ ๋ คํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ํ™•๋ฅ ์ด ๋†’๋”๋ผ๋„ ์ž ์žฌ์ ์ธ ๋ถˆํ™•์‹ค์„ฑ์„ ๊ณ ๋ คํ•˜์—ฌ ์ ˆ๋Œ€์ ์ธ ํ™•๋ฅ ๋กœ ํ•ด์„ํ•ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค.
Efficiency of Context Management. More efficient mechanisms can be devised to assist context management. For example, the pursuit of time-efficient context management techniques could significantly augment user experience by expediting the processes of context snapshotting and restoration. Also, context compression techniques can also be leveraged prior to snapshotting, which can yield a more space-efficient solution.
๋ฌธ๋งฅ ๊ด€๋ฆฌ์˜ ํšจ์œจ์„ฑ. ๋” ํšจ์œจ์ ์ธ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ํ†ตํ•ด ๋ฌธ๋งฅ ๊ด€๋ฆฌ๋ฅผ ์ง€์›ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์‹œ๊ฐ„ ํšจ์œจ์ ์ธ ๋ฌธ๋งฅ ๊ด€๋ฆฌ ๊ธฐ์ˆ ์„ ์ถ”๊ตฌํ•˜๋ฉด ๋ฌธ๋งฅ ์Šค๋ƒ…์ƒท ๋ฐ ๋ณต์› ๊ณผ์ •์„ ๊ฐ€์†ํ™”ํ•˜์—ฌ ์‚ฌ์šฉ์ž ๊ฒฝํ—˜์„ ํฌ๊ฒŒ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์Šค๋ƒ…์ƒท์„ ์ฐ๊ธฐ ์ „์— ๋ฌธ๋งฅ ์••์ถ• ๊ธฐ์ˆ ์„ ํ™œ์šฉํ•˜๋ฉด ๋” ๊ณต๊ฐ„ ํšจ์œจ์ ์ธ ์†”๋ฃจ์…˜์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ปจํ…์ŠคํŠธ ๊ด€๋ฆฌ์˜ ํšจ์œจ์„ฑ. ์ปจํ…์ŠคํŠธ ๊ด€๋ฆฌ๋ฅผ ์ง€์›ํ•˜๋Š” ๋” ํšจ์œจ์ ์ธ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ๊ณ ์•ˆํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์‹œ๊ฐ„ ํšจ์œจ์ ์ธ ์ปจํ…์ŠคํŠธ ๊ด€๋ฆฌ ๊ธฐ์ˆ ์„ ์ถ”๊ตฌํ•˜๋ฉด ์ปจํ…์ŠคํŠธ ์Šค๋ƒ…์ƒท ๋ฐ ๋ณต์› ํ”„๋กœ์„ธ์Šค๋ฅผ ์‹ ์†ํ•˜๊ฒŒ ์ฒ˜๋ฆฌํ•จ์œผ๋กœ์จ ์‚ฌ์šฉ์ž ๊ฒฝํ—˜์„ ํฌ๊ฒŒ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์Šค๋ƒ…์ƒท ์ „์— ์ปจํ…์ŠคํŠธ ์••์ถ• ๊ธฐ์ˆ ์„ ํ™œ์šฉํ•  ์ˆ˜๋„ ์žˆ์œผ๋ฉฐ, ์ด๋ฅผ ํ†ตํ•ด ๋” ๊ณต๊ฐ„ ํšจ์œจ์ ์ธ ์†”๋ฃจ์…˜์„ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Both the third and fourth examples require the model to have good tabular understanding and reasoning ability. In the third example (about _Massif_), ChatQA-70B gives correct answer by comparing the size of preserved area against 3100 ha, while GPT-4 fails to do so. In the fourth example (about _John B. England_), ChatQA-70B correctly lists three dates but misses one date, while GPT-4 correctly answers the question.
์„ธ ๋ฒˆ์งธ์™€ ๋„ค ๋ฒˆ์งธ ์˜ˆ์‹œ ๋ชจ๋‘ ๋ชจ๋ธ์ด ์šฐ์ˆ˜ํ•œ ํ‘œ ํ˜•์‹ ์ดํ•ด ๋ฐ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ๊ฐ–์ถ”๊ณ  ์žˆ์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์„ธ ๋ฒˆ์งธ ์˜ˆ์‹œ (_Massif_์— ๊ด€ํ•œ)์—์„œ ChatQA-70B๋Š” ๋ณด์กด๋œ ์ง€์—ญ์˜ ํฌ๊ธฐ๋ฅผ 3100 ํ—ฅํƒ€๋ฅด์™€ ๋น„๊ตํ•˜์—ฌ ์˜ฌ๋ฐ”๋ฅธ ๋‹ต์„ ์ œ์‹œํ•œ ๋ฐ˜๋ฉด, GPT-4๋Š” ์‹คํŒจํ–ˆ์Šต๋‹ˆ๋‹ค. ๋„ค ๋ฒˆ์งธ ์˜ˆ์‹œ (_John B. England_์— ๊ด€ํ•œ)์—์„œ ChatQA-70B๋Š” ์„ธ ๊ฐœ์˜ ๋‚ ์งœ๋ฅผ ์ •ํ™•ํžˆ ๋‚˜์—ดํ–ˆ์ง€๋งŒ ํ•˜๋‚˜์˜ ๋‚ ์งœ๋ฅผ ๋†“์ณค๊ณ , GPT-4๋Š” ์งˆ๋ฌธ์— ์ •ํ™•ํžˆ ๋‹ตํ–ˆ์Šต๋‹ˆ๋‹ค.
์„ธ ๋ฒˆ์งธ์™€ ๋„ค ๋ฒˆ์งธ ์˜ˆ์ œ ๋ชจ๋‘ ๋ชจ๋ธ์ด ํ‘œ ํ˜•์‹์˜ ์ดํ•ด๋ ฅ๊ณผ ์ถ”๋ก  ๋Šฅ๋ ฅ์ด ๋›ฐ์–ด๋‚˜์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์„ธ ๋ฒˆ์งธ ์˜ˆ์ œ(๋งˆ์‹œํ”„_์— ๊ด€ํ•œ)์—์„œ๋Š” ChatQA-70B๊ฐ€ ๋ณด์กด ๋ฉด์ ๊ณผ 3100ha๋ฅผ ๋น„๊ตํ•˜์—ฌ ์ •๋‹ต์„ ์ œ์‹œํ•œ ๋ฐ˜๋ฉด, GPT-4๋Š” ๊ทธ๋ ‡๊ฒŒ ํ•˜์ง€ ๋ชปํ–ˆ์Šต๋‹ˆ๋‹ค. ๋„ค ๋ฒˆ์งธ ์˜ˆ์ œ(์กด B. ์ž‰๊ธ€๋žœ๋“œ_์— ๊ด€ํ•œ)์—์„œ๋Š” ChatQA-70B๊ฐ€ ์„ธ ๊ฐœ์˜ ๋‚ ์งœ๋ฅผ ์˜ฌ๋ฐ”๋ฅด๊ฒŒ ๋‚˜์—ดํ•˜์ง€๋งŒ ํ•œ ๋‚ ์งœ๋ฅผ ๋†“์นœ ๋ฐ˜๋ฉด, GPT-4๋Š” ์งˆ๋ฌธ์— ์˜ฌ๋ฐ”๋ฅด๊ฒŒ ๋‹ตํ–ˆ์Šต๋‹ˆ๋‹ค.
Agent scheduler is designed to manage the agent requests in an efficient way. Consider the various agents (denoted as A, B, and C) in Figure 3, each of which has several execution steps. In the sequential execution paradigm, the agent tasks are processed in a linear order, where steps from a same agent will be processed first. This can lead to potential increased waiting times for tasks queued later in the sequence.
์—์ด์ „ํŠธ ์Šค์ผ€์ค„๋Ÿฌ๋Š” ์—์ด์ „ํŠธ ์š”์ฒญ์„ ํšจ์œจ์ ์œผ๋กœ ๊ด€๋ฆฌํ•˜๊ธฐ ์œ„ํ•ด ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. Figure 3์— ํ‘œ์‹œ๋œ ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ(A, B, C๋กœ ํ‘œ์‹œ๋จ)๋ฅผ ๊ณ ๋ คํ•ด๋ณด๋ฉด, ๊ฐ๊ฐ ์—ฌ๋Ÿฌ ์‹คํ–‰ ๋‹จ๊ณ„๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ˆœ์ฐจ ์‹คํ–‰ ํŒจ๋Ÿฌ๋‹ค์ž„์—์„œ๋Š” ์—์ด์ „ํŠธ ์ž‘์—…์ด ์„ ํ˜• ์ˆœ์„œ๋กœ ์ฒ˜๋ฆฌ๋˜๋ฉฐ, ๋™์ผํ•œ ์—์ด์ „ํŠธ์˜ ๋‹จ๊ณ„๊ฐ€ ๋จผ์ € ์ฒ˜๋ฆฌ๋ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์ˆœ์„œ์—์„œ ๋‚˜์ค‘์— ๋Œ€๊ธฐ์—ด์— ์žˆ๋Š” ์ž‘์—…์˜ ๋Œ€๊ธฐ ์‹œ๊ฐ„์ด ์ฆ๊ฐ€ํ•  ๊ฐ€๋Šฅ์„ฑ์„ ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์—์ด์ „ํŠธ ์Šค์ผ€์ค„๋Ÿฌ๋Š” ํšจ์œจ์ ์ธ ๋ฐฉ์‹์œผ๋กœ ์—์ด์ „ํŠธ ์š”์ฒญ์„ ๊ด€๋ฆฌํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆผ 3์˜ ๋‹ค์–‘ํ•œ ์—์ด์ „ํŠธ(A, B, C๋กœ ํ‘œ์‹œ๋จ)๋ฅผ ๊ณ ๋ คํ•˜๋ฉด ๊ฐ ์—์ด์ „ํŠธ๋งˆ๋‹ค ๋ช‡ ๊ฐ€์ง€ ์‹คํ–‰ ๋‹จ๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ˆœ์ฐจ์  ์‹คํ–‰ ํŒจ๋Ÿฌ๋‹ค์ž„์—์„œ๋Š” ์—์ด์ „ํŠธ ์ž‘์—…์ด ์„ ํ˜• ์ˆœ์„œ๋กœ ์ฒ˜๋ฆฌ๋˜๋ฉฐ, ๋™์ผํ•œ ์—์ด์ „ํŠธ์˜ ๋‹จ๊ณ„๊ฐ€ ๋จผ์ € ์ฒ˜๋ฆฌ๋ฉ๋‹ˆ๋‹ค. ์ด๋กœ ์ธํ•ด ์ˆœ์„œ๊ฐ€ ๋Šฆ์€ ์ž‘์—…์˜ ๋Œ€๊ธฐ ์‹œ๊ฐ„์ด ๋Š˜์–ด๋‚  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
* **MinHash alone is insufficient**, as it doesn't match the zero-shot performance of exact deduplication. Conversely, combining it with exact deduplication doesn't improve performance further. * **Masking spanned duplicates degrades performance**, systematically underperforming other approaches. Dropping and cutting spans perform similarly, although it's likely that dropping documents slightly outperforms cutting.
* **MinHash๋งŒ์œผ๋กœ๋Š” ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค**, ์ด๋Š” ์ •ํ™•ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์˜ zero-shot ์„ฑ๋Šฅ๊ณผ ์ผ์น˜ํ•˜์ง€ ์•Š๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๋ฐ˜๋Œ€๋กœ, ์ด๋ฅผ ์ •ํ™•ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์™€ ๊ฒฐํ•ฉํ•ด๋„ ์„ฑ๋Šฅ์ด ๋” ํ–ฅ์ƒ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. * **์ŠคํŒฌ ์ค‘๋ณต์„ ๋งˆ์Šคํ‚นํ•˜๋ฉด ์„ฑ๋Šฅ์ด ์ €ํ•˜๋ฉ๋‹ˆ๋‹ค**, ์ด๋Š” ๋‹ค๋ฅธ ์ ‘๊ทผ ๋ฐฉ์‹๋ณด๋‹ค ์ฒด๊ณ„์ ์œผ๋กœ ์„ฑ๋Šฅ์ด ๋–จ์–ด์ง‘๋‹ˆ๋‹ค. ์ŠคํŒฌ์„ ์‚ญ์ œํ•˜๊ฑฐ๋‚˜ ์ž๋ฅด๋Š” ๊ฒƒ์€ ๋น„์Šทํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด์ง€๋งŒ, ๋ฌธ์„œ๋ฅผ ์‚ญ์ œํ•˜๋Š” ๊ฒƒ์ด ์ž๋ฅด๋Š” ๊ฒƒ๋ณด๋‹ค ์•ฝ๊ฐ„ ๋” ๋‚˜์„ ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
* ์ •ํ™•ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์˜ ์ œ๋กœ ์ƒท ์„ฑ๋Šฅ๊ณผ ์ผ์น˜ํ•˜์ง€ ์•Š๊ธฐ ๋•Œ๋ฌธ์— **MinHash ๋‹จ๋…์œผ๋กœ๋Š” ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค**. ๋ฐ˜๋Œ€๋กœ ์ •ํ™•ํ•œ ์ค‘๋ณต ์ œ๊ฑฐ์™€ ๊ฒฐํ•ฉํ•ด๋„ ์„ฑ๋Šฅ์ด ๋” ํ–ฅ์ƒ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. * **์ŠคํŒฌ ์ค‘๋ณต์„ ๋งˆ์Šคํ‚นํ•˜๋ฉด ์„ฑ๋Šฅ์ด ์ €ํ•˜๋˜์–ด** ๋‹ค๋ฅธ ์ ‘๊ทผ ๋ฐฉ์‹๋ณด๋‹ค ์ฒด๊ณ„์ ์œผ๋กœ ์„ฑ๋Šฅ์ด ์ €ํ•˜๋ฉ๋‹ˆ๋‹ค. ์ŠคํŒฌ์„ ์‚ญ์ œํ•˜๋Š” ๊ฒƒ๊ณผ ์ž˜๋ผ๋‚ด๋Š” ๊ฒƒ์€ ๋น„์Šทํ•œ ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜์ง€๋งŒ, ๋ฌธ์„œ๋ฅผ ์‚ญ์ œํ•˜๋Š” ๊ฒƒ์ด ์ž˜๋ผ๋‚ด๋Š” ๊ฒƒ๋ณด๋‹ค ์•ฝ๊ฐ„ ๋” ์„ฑ๋Šฅ์ด ์šฐ์ˆ˜ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์Šต๋‹ˆ๋‹ค.
Moreover, our system excels under certain criteria (Section 6.2). It is especially useful in these cases to supplement the crowd prediction. We report these results in Table 3 as well, using an unweighted average (instead of the weighted average above). This outperforms the crowd prediction in all cases: For example, the crowd Brier score is.24 when the prediction is between.3 and.7, while the system achieves.237.
๊ฒŒ๋‹ค๊ฐ€, ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์€ ํŠน์ • ๊ธฐ์ค€(์„น์…˜ 6.2)์—์„œ ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฝ์šฐ์—๋Š” ๊ตฐ์ค‘ ์˜ˆ์ธก์„ ๋ณด์™„ํ•˜๋Š” ๋ฐ ํŠนํžˆ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ๊ฒฐ๊ณผ๋ฅผ ํ‘œ 3์— ๋ฌด๊ฒŒ ํ‰๊ท  ๋Œ€์‹  ๋ฌด๊ฒŒ๋ฅผ ๋‘์ง€ ์•Š์€ ํ‰๊ท ์„ ์‚ฌ์šฉํ•˜์—ฌ ๋ณด๊ณ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋“  ๊ฒฝ์šฐ์—์„œ ๊ตฐ์ค‘ ์˜ˆ์ธก์„ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค: ์˜ˆ๋ฅผ ๋“ค์–ด, ์˜ˆ์ธก์ด 0.3๊ณผ 0.7 ์‚ฌ์ด์ผ ๋•Œ ๊ตฐ์ค‘ Brier ์ ์ˆ˜๋Š” 0.24์ธ ๋ฐ˜๋ฉด, ์‹œ์Šคํ…œ์€ 0.237์„ ๋‹ฌ์„ฑํ•ฉ๋‹ˆ๋‹ค.
๋˜ํ•œ, ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์€ ํŠน์ • ๊ธฐ์ค€(์„น์…˜ 6.2)์—์„œ ํƒ์›”ํ•ฉ๋‹ˆ๋‹ค. ๊ตฐ์ค‘ ์˜ˆ์ธก์„ ๋ณด์™„ํ•˜๋Š” ๋ฐ ํŠนํžˆ ์œ ์šฉํ•œ ๊ฒฝ์šฐ์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋„ ํ‘œ 3์— ๋ณด๊ณ ํ•˜๋Š”๋ฐ, ๊ฐ€์ค‘ ํ‰๊ท  ๋Œ€์‹  ๊ฐ€์ค‘์น˜๊ฐ€ ์—†๋Š” ํ‰๊ท ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๊ตฐ์ค‘ Brier ์ ์ˆ˜๋Š” ์˜ˆ์ธก์ด .3์—์„œ .7 ์‚ฌ์ด์ผ ๋•Œ .24์ธ ๋ฐ˜๋ฉด, ์‹œ์Šคํ…œ์€ .237์„ ๋‹ฌ์„ฑํ•ฉ๋‹ˆ๋‹ค.
In this section, we focus on reviewing the related works on the SSM architectures and applications. We divide the related works into the following domains, i.e., the origin and variation of SSM, natural language processing, computer vision, graph, multi-modal and multi-media, point cloud/event stream, time series data, and others. In the following subsections, we will introduce these algorithms one after another.
์ด ์„น์…˜์—์„œ๋Š” SSM ์•„ํ‚คํ…์ฒ˜์™€ ์‘์šฉ์— ๊ด€ํ•œ ๊ด€๋ จ ์—ฐ๊ตฌ๋“ค์„ ๊ฒ€ํ† ํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘ก๋‹ˆ๋‹ค. ๊ด€๋ จ ์—ฐ๊ตฌ๋“ค์„ SSM์˜ ๊ธฐ์›๊ณผ ๋ณ€ํ˜•, ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ, ์ปดํ“จํ„ฐ ๋น„์ „, ๊ทธ๋ž˜ํ”„, ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ๋ฐ ๋ฉ€ํ‹ฐ๋ฏธ๋””์–ด, ํฌ์ธํŠธ ํด๋ผ์šฐ๋“œ/์ด๋ฒคํŠธ ์ŠคํŠธ๋ฆผ, ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ, ๊ธฐํƒ€ ๋“ฑ์œผ๋กœ ๋‚˜๋ˆ•๋‹ˆ๋‹ค. ๋‹ค์Œ ์†Œ์„น์…˜์—์„œ๋Š” ์ด๋Ÿฌํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜๋“ค์„ ํ•˜๋‚˜์”ฉ ์†Œ๊ฐœํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.
์ด ์„น์…˜์—์„œ๋Š” SSM ์•„ํ‚คํ…์ฒ˜ ๋ฐ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— ๋Œ€ํ•œ ๊ด€๋ จ ์ž‘์—…์„ ๊ฒ€ํ† ํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘ก๋‹ˆ๋‹ค. ๊ด€๋ จ ์ž‘์—…์„ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์˜์—ญ, ์ฆ‰ SSM์˜ ๊ธฐ์› ๋ฐ ๋ณ€ํ˜•, ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ, ์ปดํ“จํ„ฐ ๋น„์ „, ๊ทธ๋ž˜ํ”„, ๋‹ค์ค‘ ๋ชจ๋‹ฌ ๋ฐ ๋ฉ€ํ‹ฐ๋ฏธ๋””์–ด, ํฌ์ธํŠธ ํด๋ผ์šฐ๋“œ/์ด๋ฒคํŠธ ์ŠคํŠธ๋ฆผ, ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ ๋ฐ ๊ธฐํƒ€๋กœ ๋‚˜๋ˆ•๋‹ˆ๋‹ค. ๋‹ค์Œ ํ•˜์œ„ ์„น์…˜์—์„œ๋Š” ์ด๋Ÿฌํ•œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ฐจ๋ก€๋กœ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.
As an example, an LLM that has access to a function (or an API) such as weather API, can answer any question related to the weather of the specific place. In other words, it can use APIs to solve problems. Furthermore, if that LLM has access to an API that allows to make purchases, a purchasing agent can be built to not only have capabilities to read information from the external world, but also act on it [171].
์˜ˆ๋ฅผ ๋“ค์–ด, ๋‚ ์”จ API์™€ ๊ฐ™์€ ํ•จ์ˆ˜(๋˜๋Š” API)์— ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋Š” LLM์€ ํŠน์ • ์žฅ์†Œ์˜ ๋‚ ์”จ์™€ ๊ด€๋ จ๋œ ๋ชจ๋“  ์งˆ๋ฌธ์— ๋‹ตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์‹œ ๋งํ•ด, API๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋” ๋‚˜์•„๊ฐ€, ๋งŒ์•ฝ ๊ทธ LLM์ด ๊ตฌ๋งค๋ฅผ ํ•  ์ˆ˜ ์žˆ๋Š” API์— ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋‹ค๋ฉด, ์™ธ๋ถ€ ์„ธ๊ณ„์˜ ์ •๋ณด๋ฅผ ์ฝ๋Š” ๋Šฅ๋ ฅ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ๊ทธ์— ๋”ฐ๋ผ ํ–‰๋™ํ•  ์ˆ˜ ์žˆ๋Š” ๊ตฌ๋งค ์—์ด์ „ํŠธ๋ฅผ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค [171].
์˜ˆ๋ฅผ ๋“ค์–ด, ๋‚ ์”จ API์™€ ๊ฐ™์€ ํ•จ์ˆ˜(๋˜๋Š” API)์— ์•ก์„ธ์Šคํ•  ์ˆ˜ ์žˆ๋Š” LLM์€ ํŠน์ • ์žฅ์†Œ์˜ ๋‚ ์”จ์™€ ๊ด€๋ จ๋œ ๋ชจ๋“  ์งˆ๋ฌธ์— ๋‹ตํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฆ‰, API๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ํ•ด๋‹น LLM์ด ๊ตฌ๋งค๋ฅผ ํ—ˆ์šฉํ•˜๋Š” API์— ์•ก์„ธ์Šคํ•  ์ˆ˜ ์žˆ๋‹ค๋ฉด, ์™ธ๋ถ€ ์„ธ๊ณ„์˜ ์ •๋ณด๋ฅผ ์ฝ์„ ์ˆ˜ ์žˆ๋Š” ๊ธฐ๋Šฅ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ด์— ๋Œ€ํ•ด ํ–‰๋™ํ•  ์ˆ˜ ์žˆ๋Š” ๊ตฌ๋งค ์—์ด์ „ํŠธ๋ฅผ ๊ตฌ์ถ•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค[171].
Therefore even within the distribution of diseases and specialties we addressed, our findings should be interpreted with humility and caution. There is a need for further research to examine varied presentations of the same diseases, alongside exploration of alternate approaches to evaluating history-taking and clinical dialogue in situations of different patient needs, preferences, behaviours and circumstances.
๋”ฐ๋ผ์„œ ์šฐ๋ฆฌ๊ฐ€ ๋‹ค๋ฃฌ ์งˆ๋ณ‘๊ณผ ์ „๋ฌธ ๋ถ„์•ผ์˜ ๋ถ„ํฌ ๋‚ด์—์„œ๋„ ์šฐ๋ฆฌ์˜ ๋ฐœ๊ฒฌ์€ ๊ฒธ์†ํ•˜๊ณ  ์‹ ์ค‘ํ•˜๊ฒŒ ํ•ด์„๋˜์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋™์ผํ•œ ์งˆ๋ณ‘์˜ ๋‹ค์–‘ํ•œ ์ฆ์ƒ์„ ์กฐ์‚ฌํ•˜๊ณ , ๋‹ค๋ฅธ ํ™˜์ž์˜ ํ•„์š”, ์„ ํ˜ธ, ํ–‰๋™ ๋ฐ ์ƒํ™ฉ์— ๋”ฐ๋ฅธ ๋ณ‘๋ ฅ ์ฒญ์ทจ ๋ฐ ์ž„์ƒ ๋Œ€ํ™”๋ฅผ ํ‰๊ฐ€ํ•˜๋Š” ๋Œ€์ฒด ์ ‘๊ทผ ๋ฐฉ์‹์„ ํƒ๊ตฌํ•˜๋Š” ์ถ”๊ฐ€ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๋”ฐ๋ผ์„œ ์งˆ๋ณ‘๊ณผ ์ „๋ฌธ ๋ถ„์•ผ์˜ ๋ถ„ํฌ ๋‚ด์—์„œ๋„ ์šฐ๋ฆฌ์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ๋Š” ๊ฒธ์†ํ•˜๊ณ  ์‹ ์ค‘ํ•˜๊ฒŒ ํ•ด์„ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋™์ผํ•œ ์งˆ๋ณ‘์˜ ๋‹ค์–‘ํ•œ ํ‘œํ˜„์„ ์กฐ์‚ฌํ•˜๊ณ  ํ™˜์ž์˜ ํ•„์š”, ์„ ํ˜ธ๋„, ํ–‰๋™ ๋ฐ ์ƒํ™ฉ์ด ๋‹ค๋ฅธ ์ƒํ™ฉ์—์„œ ๋ณ‘๋ ฅ ์ฒญ์ทจ ๋ฐ ์ž„์ƒ ๋Œ€ํ™”๋ฅผ ํ‰๊ฐ€ํ•˜๋Š” ๋Œ€์ฒด ์ ‘๊ทผ ๋ฐฉ์‹์„ ํƒ์ƒ‰ํ•˜๋Š” ์ถ”๊ฐ€ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•˜๋‹ค.
To calculate the difference between the revenue from repairing truck tires and car tires on Thursday and Friday, we need to subtract the revenue from repairing car tires on Thursday and Friday from the total revenue from repairing truck tires and car tires on Thursday and Friday. Therefore, the difference between the revenue from repairing truck tires and car tires on Thursday and Friday was $900 - $640 = $260.
๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•œ ์ˆ˜์ต์˜ ์ฐจ์ด๋ฅผ ๊ณ„์‚ฐํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•œ ์ˆ˜์ต์„ ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•œ ์ด ์ˆ˜์ต์—์„œ ๋นผ์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•œ ์ˆ˜์ต์˜ ์ฐจ์ด๋Š” $900 - $640 = $260์ž…๋‹ˆ๋‹ค.
๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•˜์—ฌ ์–ป์€ ์ˆ˜์ต์˜ ์ฐจ์ด๋ฅผ ๊ณ„์‚ฐํ•˜๋ ค๋ฉด ๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•˜์—ฌ ์–ป์€ ์ด ์ˆ˜์ต์—์„œ ๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•˜์—ฌ ์–ป์€ ์ˆ˜์ต์„ ๋นผ์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ชฉ์š”์ผ๊ณผ ๊ธˆ์š”์ผ์— ํŠธ๋Ÿญ ํƒ€์ด์–ด์™€ ์ž๋™์ฐจ ํƒ€์ด์–ด๋ฅผ ์ˆ˜๋ฆฌํ•˜์—ฌ ์–ป์€ ์ˆ˜์ต์˜ ์ฐจ์ด๋Š” $900 - $640 = $260์ž…๋‹ˆ๋‹ค.
The benchmark can only identify _lack_ of safety rather than safetyBecause the benchmark only has negative predictive power, if an SUT performs well on the benchmark it does not mean that it is safe, only that we have not identified safety weaknesses. We are aware that users of the benchmark could easily misinterpret this, and therefore we will provide clear guidance regarding how results should be interpreted.
์ด ๋ฒค์น˜๋งˆํฌ๋Š” ์•ˆ์ „์„ฑ์„ ํ™•์ธํ•˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ ์•ˆ์ „์„ฑ์˜ _๋ถ€์กฑ_์„ ์‹๋ณ„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฒค์น˜๋งˆํฌ๋Š” ๋ถ€์ •์ ์ธ ์˜ˆ์ธก ๋Šฅ๋ ฅ๋งŒ ๊ฐ€์ง€๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์—, SUT๊ฐ€ ๋ฒค์น˜๋งˆํฌ์—์„œ ์ข‹์€ ์„ฑ๊ณผ๋ฅผ ๋ณด์ธ๋‹ค๊ณ  ํ•ด์„œ ์•ˆ์ „ํ•˜๋‹ค๋Š” ์˜๋ฏธ๋Š” ์•„๋‹ˆ๋ฉฐ, ๋‹จ์ง€ ์šฐ๋ฆฌ๊ฐ€ ์•ˆ์ „์„ฑ ์•ฝ์ ์„ ์‹๋ณ„ํ•˜์ง€ ๋ชปํ–ˆ์Œ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋ฒค์น˜๋งˆํฌ ์‚ฌ์šฉ์ž๊ฐ€ ์ด๋ฅผ ์‰ฝ๊ฒŒ ์˜คํ•ดํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์„ ์ธ์ง€ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ๋”ฐ๋ผ์„œ ๊ฒฐ๊ณผ๋ฅผ ์–ด๋–ป๊ฒŒ ํ•ด์„ํ•ด์•ผ ํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๋ฒค์น˜๋งˆํฌ๋Š” ์•ˆ์ „์„ฑ์ด ์•„๋‹ˆ๋ผ ์•ˆ์ „์„ฑ์˜ _๊ฒฐ์—ฌ_๋งŒ ์‹๋ณ„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.๋ฒค์น˜๋งˆํฌ๋Š” ๋ถ€์ •์ ์ธ ์˜ˆ์ธก๋ ฅ๋งŒ ๊ฐ€์ง€๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์— SUT๊ฐ€ ๋ฒค์น˜๋งˆํฌ์—์„œ ์ข‹์€ ์„ฑ๋Šฅ์„ ๋ณด์ธ๋‹ค๊ณ  ํ•ด์„œ ์•ˆ์ „ํ•˜๋‹ค๋Š” ์˜๋ฏธ๊ฐ€ ์•„๋‹ˆ๋ผ ์•ˆ์ „์„ฑ ์•ฝ์ ์„ ํ™•์ธํ•˜์ง€ ๋ชปํ–ˆ๋‹ค๋Š” ์˜๋ฏธ์ผ ๋ฟ์ž…๋‹ˆ๋‹ค. ๋ฒค์น˜๋งˆํฌ ์‚ฌ์šฉ์ž๊ฐ€ ์ด๋ฅผ ์‰ฝ๊ฒŒ ์ž˜๋ชป ํ•ด์„ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์„ ์•Œ๊ณ  ์žˆ์œผ๋ฏ€๋กœ ๊ฒฐ๊ณผ๋ฅผ ํ•ด์„ํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ๋ช…ํ™•ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
In all experiments involving GPT-3.5-Turbo presented in Section 4, we utilize the model version gpt-3.5-turbo-0613. In Table 2, the notation GPT-4 corresponds to the model version gpt-4-0613. For the experiments conducted with GPT-3.5-Turbo in Section 6, we employ the model version gpt-3.5-turbo-1106 with the JSON mode enabled. Similarly, GPT-4 in this context refers to gpt4-1106-Preview operating in JSON mode.
๋ชจ๋“  ์‹คํ—˜์—์„œ GPT-3.5-Turbo๋ฅผ ์‚ฌ์šฉํ•œ ๊ฒฝ์šฐ, ์„น์…˜ 4์—์„œ๋Š” ๋ชจ๋ธ ๋ฒ„์ „ gpt-3.5-turbo-0613์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ํ‘œ 2์—์„œ GPT-4๋ผ๋Š” ํ‘œ๊ธฐ๋Š” ๋ชจ๋ธ ๋ฒ„์ „ gpt-4-0613์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์„น์…˜ 6์—์„œ GPT-3.5-Turbo๋ฅผ ์‚ฌ์šฉํ•œ ์‹คํ—˜์—์„œ๋Š” JSON ๋ชจ๋“œ๊ฐ€ ํ™œ์„ฑํ™”๋œ ๋ชจ๋ธ ๋ฒ„์ „ gpt-3.5-turbo-1106์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด์™€ ์œ ์‚ฌํ•˜๊ฒŒ, ์ด ๋ฌธ๋งฅ์—์„œ GPT-4๋Š” JSON ๋ชจ๋“œ๋กœ ์ž‘๋™ํ•˜๋Š” gpt4-1106-Preview๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.
์„น์…˜ 4์— ์ œ์‹œ๋œ GPT-3.5-Turbo์™€ ๊ด€๋ จ๋œ ๋ชจ๋“  ์‹คํ—˜์—์„œ๋Š” ๋ชจ๋ธ ๋ฒ„์ „ gpt-3.5-turbo-0613์„ ํ™œ์šฉํ•ฉ๋‹ˆ๋‹ค. ํ‘œ 2์—์„œ GPT-4๋ผ๋Š” ํ‘œ๊ธฐ๋Š” ๋ชจ๋ธ ๋ฒ„์ „ gpt-4-0613์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค. ์„น์…˜ 6์—์„œ GPT-3.5-Turbo๋กœ ์ˆ˜ํ–‰ํ•œ ์‹คํ—˜์˜ ๊ฒฝ์šฐ JSON ๋ชจ๋“œ๊ฐ€ ํ™œ์„ฑํ™”๋œ ๋ชจ๋ธ ๋ฒ„์ „ gpt-3.5-turbo-1106์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์ด ๋งฅ๋ฝ์—์„œ GPT-4๋Š” JSON ๋ชจ๋“œ๋กœ ์ž‘๋™ํ•˜๋Š” gpt4-1106-Preview๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.
performance still surpasses that of standard SFT (45.0, +3.4). Nonetheless, it exhibits lower performance compared to the computation of losses across all tokens (47.1, +5.5). Predicting the next tokens for masked positions poses a greater challenge than for unmasked ones, as they lack adjacent unmasked tokens from which to gather local information. This increased complexity ultimately improves the performance.
์„ฑ๋Šฅ์€ ์—ฌ์ „ํžˆ ํ‘œ์ค€ SFT(45.0, +3.4)๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿผ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , ๋ชจ๋“  ํ† ํฐ์— ๋Œ€ํ•œ ์†์‹ค ๊ณ„์‚ฐ๊ณผ ๋น„๊ตํ–ˆ์„ ๋•Œ๋Š” ๋‚ฎ์€ ์„ฑ๋Šฅ์„ ๋ณด์ž…๋‹ˆ๋‹ค(47.1, +5.5). ๋งˆ์Šคํ‚น๋œ ์œ„์น˜์— ๋Œ€ํ•œ ๋‹ค์Œ ํ† ํฐ์„ ์˜ˆ์ธกํ•˜๋Š” ๊ฒƒ์€ ์ธ์ ‘ํ•œ ๋งˆ์Šคํ‚น๋˜์ง€ ์•Š์€ ํ† ํฐ์œผ๋กœ๋ถ€ํ„ฐ ์ง€์—ญ ์ •๋ณด๋ฅผ ์ˆ˜์ง‘ํ•  ์ˆ˜ ์—†๊ธฐ ๋•Œ๋ฌธ์— ๋งˆ์Šคํ‚น๋˜์ง€ ์•Š์€ ์œ„์น˜๋ณด๋‹ค ๋” ํฐ ๋„์ „ ๊ณผ์ œ๋ฅผ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ฆ๊ฐ€๋œ ๋ณต์žก์„ฑ์€ ๊ถ๊ทน์ ์œผ๋กœ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ต๋‹ˆ๋‹ค.
์„ฑ๋Šฅ์€ ์—ฌ์ „ํžˆ ํ‘œ์ค€ SFT(45.0, +3.4)๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿผ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ๋ชจ๋“  ํ† ํฐ์— ๊ฑธ์นœ ์†์‹ค ๊ณ„์‚ฐ์— ๋น„ํ•ด ์„ฑ๋Šฅ์ด ๋–จ์–ด์ง‘๋‹ˆ๋‹ค(47.1, +5.5). ๋งˆ์Šคํ‚น๋œ ์œ„์น˜์˜ ๋‹ค์Œ ํ† ํฐ์„ ์˜ˆ์ธกํ•˜๋Š” ๊ฒƒ์€ ๋งˆ์Šคํ‚น๋˜์ง€ ์•Š์€ ํ† ํฐ์— ๋น„ํ•ด ๋” ์–ด๋ ค์šด ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค. ๋งˆ์Šคํ‚น๋˜์ง€ ์•Š์€ ํ† ํฐ์€ ์ธ์ ‘ํ•œ ๋งˆ์Šคํ‚น๋˜์ง€ ์•Š์€ ํ† ํฐ์—์„œ ๋กœ์ปฌ ์ •๋ณด๋ฅผ ์ˆ˜์ง‘ํ•  ์ˆ˜ ์—†๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ณต์žก์„ฑ์ด ์ฆ๊ฐ€ํ•˜๋ฉด ๊ถ๊ทน์ ์œผ๋กœ ์„ฑ๋Šฅ์ด ํ–ฅ์ƒ๋ฉ๋‹ˆ๋‹ค.
As foundational architectures, we opt for SOLAR-10.7B (Kim et al., 2023) and Phi-2 (Li et al., 2023), because both have shown outstanding performances among similar sizes of LLMs. This choice of foundational architectures aligns with our strategic training objectives, leveraging their proven strengths to ensure our new models achieve similar levels of language understanding and reasoning capabilities in Korean.
๊ธฐ๋ณธ ์•„ํ‚คํ…์ฒ˜๋กœ๋Š” SOLAR-10.7B (Kim et al., 2023)์™€ Phi-2 (Li et al., 2023)๋ฅผ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ๋‘ ๋ชจ๋ธ์€ ์œ ์‚ฌํ•œ ํฌ๊ธฐ์˜ LLM๋“ค ์ค‘์—์„œ ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ฃผ์—ˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ธฐ๋ณธ ์•„ํ‚คํ…์ฒ˜ ์„ ํƒ์€ ์šฐ๋ฆฌ์˜ ์ „๋žต์  ํ›ˆ๋ จ ๋ชฉํ‘œ์™€ ์ผ์น˜ํ•˜๋ฉฐ, ์ด๋“ค์˜ ์ž…์ฆ๋œ ๊ฐ•์ ์„ ํ™œ์šฉํ•˜์—ฌ ์ƒˆ๋กœ์šด ๋ชจ๋ธ๋“ค์ด ํ•œ๊ตญ์–ด์—์„œ ์œ ์‚ฌํ•œ ์ˆ˜์ค€์˜ ์–ธ์–ด ์ดํ•ด ๋ฐ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.
๊ธฐ๋ณธ ์•„ํ‚คํ…์ฒ˜๋กœ๋Š” SOLAR-10.7B(Kim et al., 2023)์™€ Phi-2(Li et al., 2023)๋ฅผ ์„ ํƒํ–ˆ๋Š”๋ฐ, ์ด ๋‘ ๋ชจ๋ธ์€ ๋น„์Šทํ•œ ํฌ๊ธฐ์˜ LLM ์ค‘์—์„œ ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์คฌ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ธฐ๋ณธ ์•„ํ‚คํ…์ฒ˜์˜ ์„ ํƒ์€ ํ•œ๊ตญ์–ด์—์„œ๋„ ์ƒˆ๋กœ์šด ๋ชจ๋ธ์ด ๋น„์Šทํ•œ ์ˆ˜์ค€์˜ ์–ธ์–ด ์ดํ•ด ๋ฐ ์ถ”๋ก  ๊ธฐ๋Šฅ์„ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ์ž…์ฆ๋œ ๊ฐ•์ ์„ ํ™œ์šฉํ•˜์—ฌ ์ „๋žต์  ํ›ˆ๋ จ ๋ชฉํ‘œ์— ๋ถ€ํ•ฉํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.
100PoisonMpts addressing issues of anti-discrimination and empathy, spans various dimensions including jurisprudence, psychology, child education, obscure facts, intimate relationships, etc. It involves human-generated prompts that evoke bias and discrimination, followed by expert-crafted responses that align with human values. To enhance the harmlessness of the CQIA, we sample all the data from 100PoisonMpts.
100PoisonMpts๋Š” ๋ฐ˜์ฐจ๋ณ„ ๋ฐ ๊ณต๊ฐ ๋ฌธ์ œ๋ฅผ ๋‹ค๋ฃจ๋ฉฐ, ๋ฒ•ํ•™, ์‹ฌ๋ฆฌํ•™, ์•„๋™ ๊ต์œก, ์ž˜ ์•Œ๋ ค์ง€์ง€ ์•Š์€ ์‚ฌ์‹ค, ์นœ๋ฐ€ํ•œ ๊ด€๊ณ„ ๋“ฑ ๋‹ค์–‘ํ•œ ์ฐจ์›์„ ํฌ๊ด„ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ํŽธ๊ฒฌ๊ณผ ์ฐจ๋ณ„์„ ์œ ๋ฐœํ•˜๋Š” ์ธ๊ฐ„์ด ์ƒ์„ฑํ•œ ํ”„๋กฌํ”„ํŠธ์™€, ์ธ๊ฐ„์˜ ๊ฐ€์น˜์— ๋ถ€ํ•ฉํ•˜๋Š” ์ „๋ฌธ๊ฐ€๊ฐ€ ์ž‘์„ฑํ•œ ์‘๋‹ต์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. CQIA์˜ ๋ฌดํ•ด์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” 100PoisonMpts์˜ ๋ชจ๋“  ๋ฐ์ดํ„ฐ๋ฅผ ์ƒ˜ํ”Œ๋งํ•ฉ๋‹ˆ๋‹ค.
์ฐจ๋ณ„ ๊ธˆ์ง€ ๋ฐ ๊ณต๊ฐ ๋ฌธ์ œ๋ฅผ ๋‹ค๋ฃจ๋Š” 100PoisonMpts๋Š” ๋ฒ•ํ•™, ์‹ฌ๋ฆฌํ•™, ์•„๋™ ๊ต์œก, ๋ชจํ˜ธํ•œ ์‚ฌ์‹ค, ์นœ๋ฐ€ํ•œ ๊ด€๊ณ„ ๋“ฑ ๋‹ค์–‘ํ•œ ์ฐจ์›์— ๊ฑธ์ณ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ํŽธ๊ฒฌ๊ณผ ์ฐจ๋ณ„์„ ๋ถˆ๋Ÿฌ์ผ์œผํ‚ค๋Š” ์‚ฌ๋žŒ์ด ์ƒ์„ฑํ•œ ํ”„๋กฌํ”„ํŠธ์™€ ์ธ๊ฐ„์˜ ๊ฐ€์น˜๊ด€์— ๋ถ€ํ•ฉํ•˜๋Š” ์ „๋ฌธ๊ฐ€๊ฐ€ ์ž‘์„ฑํ•œ ์‘๋‹ต์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. CQIA์˜ ๋ฌดํ•ด์„ฑ์„ ๊ฐ•ํ™”ํ•˜๊ธฐ ์œ„ํ•ด 100PoisonMpts์—์„œ ๋ชจ๋“  ๋ฐ์ดํ„ฐ๋ฅผ ์ƒ˜ํ”Œ๋งํ•ฉ๋‹ˆ๋‹ค.
At the same time this is still a new and extremely active research area where the pace of innovation is increasing rather than slowing down. As in any other evolving area though, there are still numerous challenges ahead. Here we briefly mention some of the challenges and main active areas which are known so far. It is worth noting that LLM challenges are discussed in details in a work by Kaddour et al. [207].
๋™์‹œ์—, ์ด๋Š” ์—ฌ์ „ํžˆ ์ƒˆ๋กœ์šด ์—ฐ๊ตฌ ๋ถ„์•ผ์ด๋ฉฐ, ํ˜์‹ ์˜ ์†๋„๊ฐ€ ๋Š๋ ค์ง€๊ธฐ๋ณด๋‹ค๋Š” ์˜คํžˆ๋ ค ์ฆ๊ฐ€ํ•˜๊ณ  ์žˆ๋Š” ๋งค์šฐ ํ™œ๋ฐœํ•œ ์—ฐ๊ตฌ ์˜์—ญ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋‹ค๋ฅธ ๋ฐœ์ „ํ•˜๋Š” ๋ถ„์•ผ์™€ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ, ์•ž์œผ๋กœ๋„ ์ˆ˜๋งŽ์€ ๋„์ „ ๊ณผ์ œ๊ฐ€ ๋‚จ์•„ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ๋Š” ์ง€๊ธˆ๊นŒ์ง€ ์•Œ๋ ค์ง„ ๋ช‡ ๊ฐ€์ง€ ๋„์ „ ๊ณผ์ œ์™€ ์ฃผ์š” ํ™œ๋ฐœํ•œ ์—ฐ๊ตฌ ์˜์—ญ์„ ๊ฐ„๋žตํžˆ ์–ธ๊ธ‰ํ•ฉ๋‹ˆ๋‹ค. LLM์˜ ๋„์ „ ๊ณผ์ œ๋Š” Kaddour et al. [207]์˜ ์—ฐ๊ตฌ์—์„œ ์ž์„ธํžˆ ๋…ผ์˜๋˜๊ณ  ์žˆ์Œ์„ ์ฃผ๋ชฉํ•  ๊ฐ€์น˜๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
๋™์‹œ์— ์ด ๋ถ„์•ผ๋Š” ์—ฌ์ „ํžˆ ํ˜์‹ ์˜ ์†๋„๊ฐ€ ๋‘”ํ™”๋˜๊ธฐ๋ณด๋‹ค๋Š” ์ฆ๊ฐ€ํ•˜๊ณ  ์žˆ๋Š” ์ƒˆ๋กœ์šด์ด๊ณ  ๋งค์šฐ ํ™œ๋ฐœํ•œ ์—ฐ๊ตฌ ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋‹ค๋ฅธ ์ง„ํ™”ํ•˜๋Š” ๋ถ„์•ผ์™€ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ์•„์ง ์•ž์œผ๋กœ ํ•ด๊ฒฐํ•ด์•ผ ํ•  ๊ณผ์ œ๊ฐ€ ๋งŽ์ด ๋‚จ์•„ ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—์„œ๋Š” ์ง€๊ธˆ๊นŒ์ง€ ์•Œ๋ ค์ง„ ๋ช‡ ๊ฐ€์ง€ ๊ณผ์ œ์™€ ์ฃผ์š” ํ™œ๋ฐœํ•œ ๋ถ„์•ผ์— ๋Œ€ํ•ด ๊ฐ„๋žตํ•˜๊ฒŒ ์–ธ๊ธ‰ํ•ฉ๋‹ˆ๋‹ค. LLM์˜ ๊ณผ์ œ๋Š” Kaddour ๋“ฑ์˜ ์—ฐ๊ตฌ[207]์—์„œ ์ž์„ธํžˆ ๋…ผ์˜ํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์ ์— ์ฃผ๋ชฉํ•  ํ•„์š”๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
It is noteworthy that the relative performance gain is more substantial with increasing task difficulty. Specifically, we observe that within the same task, the smaller model, Llama2-13B, gains ranging from \(28\%\)-\(200\%\), but only \(8\%\)-\(16\%\) over GPT-3.5-Turbo. Moreover, the more challenging task MATH yields gains of \(34\%\)-\(200\%\), in contrast to only \(16\%\)-\(69\%\) on the easier task GSM8K.
์ฃผ๋ชฉํ•  ๋งŒํ•œ ์ ์€ ๊ณผ์ œ์˜ ๋‚œ์ด๋„๊ฐ€ ์ฆ๊ฐ€ํ• ์ˆ˜๋ก ์ƒ๋Œ€์ ์ธ ์„ฑ๋Šฅ ํ–ฅ์ƒ์ด ๋” ํฌ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๋™์ผํ•œ ๊ณผ์ œ ๋‚ด์—์„œ ๋” ์ž‘์€ ๋ชจ๋ธ์ธ Llama2-13B๋Š” \(28\%\)-\(200\%\)์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์ด์ง€๋งŒ, GPT-3.5-Turbo์— ๋น„ํ•ด์„œ๋Š” ๋‹จ์ง€ \(8\%\)-\(16\%\)์˜ ํ–ฅ์ƒ๋งŒ์„ ๋ณด์ž…๋‹ˆ๋‹ค. ๋”์šฑ์ด, ๋” ์–ด๋ ค์šด ๊ณผ์ œ์ธ MATH์—์„œ๋Š” \(34\%\)-\(200\%\)์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์ด๋Š” ๋ฐ˜๋ฉด, ๋” ์‰ฌ์šด ๊ณผ์ œ์ธ GSM8K์—์„œ๋Š” ๋‹จ์ง€ \(16\%\)-\(69\%\)์˜ ํ–ฅ์ƒ๋งŒ์„ ๋ณด์ž…๋‹ˆ๋‹ค.
๊ด€์‹ฌ ์žˆ๋Š” ์ ์€ ์ž‘์—… ๋‚œ์ด๋„๊ฐ€ ๋†’์•„์งˆ์ˆ˜๋ก ์ƒ๋Œ€์  ์„ฑ๋Šฅ ํ–ฅ์ƒ์ด ๋” ํฐ ์ ์ž…๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๋™์ผํ•œ ์ž‘์—… ๋‚ด์—์„œ ๋” ์ž‘์€ ๋ชจ๋ธ์ธ Llama2-13B๋Š” \(28\%\)-\(200\%\)์˜ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ์–ป์ง€๋งŒ, GPT-3.5-Turbo์— ๋น„ํ•ด \(8\%\)-\(16\%\)๋งŒ ํ–ฅ์ƒ๋˜๋Š” ๊ฒƒ์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ๋” ์–ด๋ ค์šด ๊ณผ์ œ์ธ MATH๋Š” ๋” ์‰ฌ์šด ๊ณผ์ œ์ธ GSM8K์—์„œ \(16\%\)-\(69\%\)๋งŒ ์–ป์€ ๊ฒƒ๊ณผ๋Š” ๋Œ€์กฐ์ ์œผ๋กœ \(34\%\)-\(200\%\)์˜ ์ด๋“์„ ์–ป์Šต๋‹ˆ๋‹ค.
Self-Consistency has significant applications in fields where the veracity of information is critical. It is particularly relevant in scenarios like fact-checking, where ensuring the accuracy of information provided by AI models is essential. By employing this technique, prompt engineers can enhance the trustworthiness of LLMs, making them more reliable for tasks that require high levels of factual accuracy.
์ž๊ธฐ ์ผ๊ด€์„ฑ(Self-Consistency)์€ ์ •๋ณด์˜ ์ง„์‹ค์„ฑ์ด ์ค‘์š”ํ•œ ๋ถ„์•ผ์—์„œ ์ค‘์š”ํ•œ ์‘์šฉ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ AI ๋ชจ๋ธ์ด ์ œ๊ณตํ•˜๋Š” ์ •๋ณด์˜ ์ •ํ™•์„ฑ์„ ๋ณด์žฅํ•˜๋Š” ๊ฒƒ์ด ํ•„์ˆ˜์ ์ธ ์‚ฌ์‹ค ํ™•์ธ๊ณผ ๊ฐ™์€ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ๊ด€๋ จ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์„ ์‚ฌ์šฉํ•จ์œผ๋กœ์จ, ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋Š” LLM์˜ ์‹ ๋ขฐ์„ฑ์„ ํ–ฅ์ƒ์‹œ์ผœ ๋†’์€ ์ˆ˜์ค€์˜ ์‚ฌ์‹ค์  ์ •ํ™•์„ฑ์ด ์š”๊ตฌ๋˜๋Š” ์ž‘์—…์— ๋” ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋งŒ๋“ญ๋‹ˆ๋‹ค.
์ž๊ธฐ ์ผ๊ด€์„ฑ์€ ์ •๋ณด์˜ ์ง„์‹ค์„ฑ์ด ์ค‘์š”ํ•œ ๋ถ„์•ผ์—์„œ ์ƒ๋‹นํžˆ ์ค‘์š”ํ•œ ์‘์šฉ ๋ถ„์•ผ๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ AI ๋ชจ๋ธ์—์„œ ์ œ๊ณตํ•˜๋Š” ์ •๋ณด์˜ ์ •ํ™•์„ฑ์„ ๋ณด์žฅํ•ด์•ผ ํ•˜๋Š” ์‚ฌ์‹ค ํ™•์ธ๊ณผ ๊ฐ™์€ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ๊ด€๋ จ์„ฑ์ด ๋†’์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ๋ฒ•์„ ์‚ฌ์šฉํ•˜๋ฉด ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋Š” LLM์˜ ์‹ ๋ขฐ์„ฑ์„ ํ–ฅ์ƒ์‹œ์ผœ ์‚ฌ์‹ค ์ •ํ™•๋„๊ฐ€ ๋†’์€ ์ž‘์—…์— ๋” ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋งŒ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Therefore, once we find a range \((L_{\text{left}},L_{\text{right}})\) where \(f(L_{\text{left}})\cdot f(L_{\text{right}})<0\), we can iteratively search for the critical point \(L^{*}\) with \(f(L^{*})=0\) through the bisection method. In practice, the loss value is positive and falls within a restricted range, setting \(L_{\text{left}}=0,L_{\text{right}}=10\) is always sufficient to cover the entire range.
๋”ฐ๋ผ์„œ, \(f(L_{\text{left}})\cdot f(L_{\text{right}})<0\)์ธ ๋ฒ”์œ„ \((L_{\text{left}},L_{\text{right}})\)๋ฅผ ์ฐพ์œผ๋ฉด, ์ด๋ถ„๋ฒ•์„ ํ†ตํ•ด \(f(L^{*})=0\)์ธ ์ž„๊ณ„์  \(L^{*}\)์„ ๋ฐ˜๋ณต์ ์œผ๋กœ ๊ฒ€์ƒ‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ ์†์‹ค ๊ฐ’์€ ์–‘์ˆ˜์ด๋ฉฐ ์ œํ•œ๋œ ๋ฒ”์œ„ ๋‚ด์— ์žˆ์œผ๋ฏ€๋กœ, \(L_{\text{left}}=0, L_{\text{right}}=10\)์œผ๋กœ ์„ค์ •ํ•˜๋Š” ๊ฒƒ๋งŒ์œผ๋กœ๋„ ์ „์ฒด ๋ฒ”์œ„๋ฅผ ํ•ญ์ƒ ์ปค๋ฒ„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๋”ฐ๋ผ์„œ ์ผ๋‹จ \(f(L_{\text{left}})\cdot f(L_{\text{right}})<0\)์ธ ๋ฒ”์œ„ \((L_{\text{left}},L_{\text{right}})\)๋ฅผ ์ฐพ์œผ๋ฉด ์ด๋ถ„๋ฒ•์„ ํ†ตํ•ด \(f(L^{*})=0\)์ธ ์ž„๊ณ„์  \(L^{*}\)๋ฅผ ๋ฐ˜๋ณต์ ์œผ๋กœ ๊ฒ€์ƒ‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ ์†์‹ค ๊ฐ’์€ ์–‘์ˆ˜์ด๋ฉฐ ์ œํ•œ๋œ ๋ฒ”์œ„ ๋‚ด์— ์†ํ•˜๋ฏ€๋กœ \(L_{\text{left}}=0,L_{\text{right}}=10\)์„ ์„ค์ •ํ•˜๋ฉด ์ „์ฒด ๋ฒ”์œ„๋ฅผ ํฌ๊ด„ํ•˜๋Š” ๋ฐ ํ•ญ์ƒ ์ถฉ๋ถ„ํ•ฉ๋‹ˆ๋‹ค.
To start the analysis, we select two datasets with increasing difficulty, i.e., GSM8K and MATH, to calculate the relative performance gain. The relative performance gain \(\eta\) is given by: \(\eta=\frac{P_{\text{m}}-P_{\text{s}}}{P_{\text{s}}}\) where \(P_{\text{m}}\) and \(P_{\text{s}}\) are the performances (accuracy) with our method and a single LLM query, respectively. The results are shown in Table 5.
๋ถ„์„์„ ์‹œ์ž‘ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ๋‚œ์ด๋„๊ฐ€ ์ฆ๊ฐ€ํ•˜๋Š” ๋‘ ๊ฐœ์˜ ๋ฐ์ดํ„ฐ์…‹, ์ฆ‰ GSM8K์™€ MATH๋ฅผ ์„ ํƒํ•˜์—ฌ ์ƒ๋Œ€ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. ์ƒ๋Œ€ ์„ฑ๋Šฅ ํ–ฅ์ƒ \(\eta\)๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ฃผ์–ด์ง‘๋‹ˆ๋‹ค: \(\eta=\frac{P_{\text{m}}-P_{\text{s}}}{P_{\text{s}}}\) ์—ฌ๊ธฐ์„œ \(P_{\text{m}}\)๊ณผ \(P_{\text{s}}\)๋Š” ๊ฐ๊ฐ ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•๊ณผ ๋‹จ์ผ LLM ์ฟผ๋ฆฌ๋กœ ์–ป์€ ์„ฑ๋Šฅ(์ •ํ™•๋„)์ž…๋‹ˆ๋‹ค. ๊ฒฐ๊ณผ๋Š” Table 5์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค.
๋ถ„์„์„ ์‹œ์ž‘ํ•˜๊ธฐ ์œ„ํ•ด ์ ์  ๋” ์–ด๋ ค์›Œ์ง€๋Š” ๋‘ ๊ฐœ์˜ ๋ฐ์ดํ„ฐ ์„ธํŠธ, ์ฆ‰ GSM8K์™€ MATH๋ฅผ ์„ ํƒํ•˜์—ฌ ์ƒ๋Œ€์  ์„ฑ๋Šฅ ์ด๋“์„ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. ์ƒ๋Œ€์  ์„ฑ๋Šฅ ์ด๋“(\(\eta\))์€ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ฃผ์–ด์ง‘๋‹ˆ๋‹ค: \(\eta=\frac{P_{\text{m}}-P_{\text{s}}}{P_{\text{s}}}\) ์—ฌ๊ธฐ์„œ \(P_{\text{m}}\)์™€ \(P_{\text{s}}\)๋Š” ๊ฐ๊ฐ ์šฐ๋ฆฌ ๋ฐฉ๋ฒ•๊ณผ ๋‹จ์ผ LLM ์ฟผ๋ฆฌ๋กœ ์–ป์€ ์„ฑ๋Šฅ(์ •ํ™•๋„)์ž…๋‹ˆ๋‹ค. ๊ฒฐ๊ณผ๋Š” ํ‘œ 5์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค.
BOT: Angelo and Melaniethink they should dedicate 3 hours to each * of the 2 chapters, 3 hours x 2 chapters = 6 hours total.\n * For the worksheets they plan to dedicate 1.5 hours for each * worksheet, 1.5 hours x 4 worksheets = 6 hours total.\n * Angelo and Melanie need to start with planning 12 hours to * study, at 4 hours a day, 12 / 4 = 3 days.\nHowever, they * need to include time for breaks and lunch.
BOT: Angelo์™€ Melanie๋Š” ๊ฐ ์žฅ์— 3์‹œ๊ฐ„์”ฉ ํ• ์• ํ•ด์•ผ ํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. * 2๊ฐœ์˜ ์žฅ์— ๋Œ€ํ•ด 3์‹œ๊ฐ„ x 2์žฅ = ์ด 6์‹œ๊ฐ„.\n * ์›Œํฌ์‹œํŠธ์— ๋Œ€ํ•ด์„œ๋Š” ๊ฐ ์›Œํฌ์‹œํŠธ์— 1.5์‹œ๊ฐ„์„ ํ• ์• ํ•  ๊ณ„ํš์ž…๋‹ˆ๋‹ค. * 1.5์‹œ๊ฐ„ x 4๊ฐœ์˜ ์›Œํฌ์‹œํŠธ = ์ด 6์‹œ๊ฐ„.\n * Angelo์™€ Melanie๋Š” ํ•˜๋ฃจ์— 4์‹œ๊ฐ„์”ฉ 12์‹œ๊ฐ„์„ ๊ณ„ํšํ•ด์•ผ ํ•˜๋ฏ€๋กœ, 12 / 4 = 3์ผ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.\n ํ•˜์ง€๋งŒ, ๊ทธ๋“ค์€ ํœด์‹ ์‹œ๊ฐ„๊ณผ ์ ์‹ฌ ์‹œ๊ฐ„์„ ํฌํ•จํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
BOT: ์•™ํ—ฌ๋กœ์™€ ๋ฉœ๋ผ๋‹ˆ๋Š” ๊ฐ๊ฐ 2์‹œ๊ฐ„์”ฉ 3์‹œ๊ฐ„์„ ํ• ์• ํ•ด์•ผ ํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. * 2๊ฐœ์˜ ์ฑ•ํ„ฐ, 3์‹œ๊ฐ„ x 2์ฑ•ํ„ฐ = ์ด 6์‹œ๊ฐ„.\n * ์›Œํฌ์‹œํŠธ์˜ ๊ฒฝ์šฐ ๊ฐ๊ฐ 1.5์‹œ๊ฐ„์”ฉ ํ• ์• ํ•  ๊ณ„ํš์ž…๋‹ˆ๋‹ค. * ์›Œํฌ์‹œํŠธ, 1.5์‹œ๊ฐ„ x 4 ์›Œํฌ์‹œํŠธ = ์ด 6์‹œ๊ฐ„.\n * ์•™ํ—ฌ๋กœ์™€ ๋ฉœ๋ผ๋‹ˆ๋Š” ํ•˜๋ฃจ์— 4์‹œ๊ฐ„์”ฉ 12์‹œ๊ฐ„์„ ํ• ์• ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. * ๊ณต๋ถ€๋ฅผ ์œ„ํ•ด ๊ณ„ํš์„ ์„ธ์›Œ์•ผ ํ•ฉ๋‹ˆ๋‹ค.\n๊ทธ๋Ÿฌ๋‚˜ ๊ทธ๋“ค์€ * ํœด์‹๊ณผ ์ ์‹ฌ ์‹œ๊ฐ„์„ ํฌํ•จํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
**Vicuna**: The sun was shining bright on the warm pavement as my character stepped outside into the backyard. It had been a long day at work and all they wanted to do was relax and unwind. As they plopped down on the louger, they glanced over at the Garden of Euclidean proportions that their neighbor had cultivated. They sighed at the unkempt state of their own garden and vowed to tidy it up sometime soon.
Vicuna: ๋‚ด ์บ๋ฆญํ„ฐ๊ฐ€ ๋’ท๋งˆ๋‹น์œผ๋กœ ๋‚˜์„œ์ž ๋”ฐ๋œปํ•œ ํฌ์žฅ๋„๋กœ ์œ„๋กœ ํ–‡๋น›์ด ๋ฐ๊ฒŒ ๋น„์น˜๊ณ  ์žˆ์—ˆ๋‹ค. ์ง์žฅ์—์„œ์˜ ๊ธด ํ•˜๋ฃจ๋ฅผ ๋ณด๋‚ธ ํ›„, ๊ทธ๋“ค์ด ์›ํ–ˆ๋˜ ๊ฒƒ์€ ๋‹จ์ง€ ํœด์‹์„ ์ทจํ•˜๊ณ  ๊ธด์žฅ์„ ํ‘ธ๋Š” ๊ฒƒ์ด์—ˆ๋‹ค. ๊ทธ๋“ค์ด ๋ผ์šด์ €์— ๋ชธ์„ ๋˜์ง€๋ฉฐ ์ด์›ƒ์ด ๊ฐ€๊พผ ์œ ํด๋ฆฌ๋“œ ๋น„์œจ์˜ ์ •์›์„ ํž๋— ๋ฐ”๋ผ๋ณด์•˜๋‹ค. ๊ทธ๋“ค์€ ์ž์‹ ์˜ ์ •์›์ด ์ •๋ˆ๋˜์ง€ ์•Š์€ ์ƒํƒœ๋ฅผ ๋ณด๊ณ  ํ•œ์ˆจ์„ ์‰ฌ๋ฉฐ ์กฐ๋งŒ๊ฐ„ ์ •๋ฆฌํ•˜๊ฒ ๋‹ค๊ณ  ๋‹ค์งํ–ˆ๋‹ค.
**๋น„์ฟ ๋‚˜**: ํ–‡์‚ด์ด ๋”ฐ์‚ฌ๋กญ๊ฒŒ ๋‚ด๋ฆฌ์ฌ๋Š” ์•„๋ž˜, ๋‚ด ์บ๋ฆญํ„ฐ๋Š” ์ง‘ ๋ฐ–์œผ๋กœ ๋‚˜์™€ ๋’ท๋งˆ๋‹น์œผ๋กœ ๋“ค์–ด์„ฐ์–ด์š”. ์ง์žฅ์—์„œ ๊ธด ํ•˜๋ฃจ๋ฅผ ๋ณด๋‚ธ ํ›„, ๊ทธ์ € ๊ธด์žฅ์„ ํ’€๊ณ  ํœด์‹์„ ์ทจํ•˜๊ณ  ์‹ถ์—ˆ์ฃ . ๋ผ์šด์ €์— ๋ชธ์„ ๋˜์ง€์ž๋งˆ์ž ์ด์›ƒ์ด ๊ฐ€๊พผ ์œ ํด๋ฆฌ๋“œ ๋น„์œจ์˜ ์ •์›์ด ๋ˆˆ์— ๋“ค์–ด์™”์–ด์š”. ์—‰์„ฑํ•œ ์ •์› ์ƒํƒœ์— ํ•œ์ˆจ์„ ์‰ฌ๋ฉฐ ์–ธ์  ๊ฐ€ ์ •๋ฆฌํ•˜๊ฒ ๋‹ค๊ณ  ๋‹ค์งํ–ˆ์Šต๋‹ˆ๋‹ค.
We modify the codebase of Nanda and Bloom (2022) to run the experiments. We use 1-8 40GB A100 GPUs for the experiments. All experiments run in less than 24 hours. We use the model weights from HuggingFace Transformers (Wolf et al., 2020) and use full precision for LLaMA-2 7B and 13B and half-precision for 70B. The SPARQL queries for querying Wikidata are written with the help of GPT-4 (OpenAI et al., 2023).
Nanda์™€ Bloom (2022)์˜ ์ฝ”๋“œ๋ฒ ์ด์Šค๋ฅผ ์ˆ˜์ •ํ•˜์—ฌ ์‹คํ—˜์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ์‹คํ—˜์—๋Š” 1-8๊ฐœ์˜ 40GB A100 GPU๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋“  ์‹คํ—˜์€ 24์‹œ๊ฐ„ ์ด๋‚ด์— ์™„๋ฃŒ๋ฉ๋‹ˆ๋‹ค. HuggingFace Transformers (Wolf et al., 2020)์—์„œ ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๋ฅผ ์‚ฌ์šฉํ•˜๋ฉฐ, LLaMA-2 7B์™€ 13B์—๋Š” ํ’€ ํ”„๋ฆฌ์‹œ์ „์„, 70B์—๋Š” ํ•˜ํ”„ ํ”„๋ฆฌ์‹œ์ „์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. Wikidata๋ฅผ ์ฟผ๋ฆฌํ•˜๊ธฐ ์œ„ํ•œ SPARQL ์ฟผ๋ฆฌ๋Š” GPT-4 (OpenAI et al., 2023)์˜ ๋„์›€์„ ๋ฐ›์•„ ์ž‘์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
์‹คํ—˜์„ ์‹คํ–‰ํ•˜๊ธฐ ์œ„ํ•ด Nanda์™€ Bloom(2022)์˜ ์ฝ”๋“œ๋ฒ ์ด์Šค๋ฅผ ์ˆ˜์ •ํ•ฉ๋‹ˆ๋‹ค. ์‹คํ—˜์—๋Š” 1~8๊ฐœ์˜ 40GB A100 GPU๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋ชจ๋“  ์‹คํ—˜์€ 24์‹œ๊ฐ„ ์ด๋‚ด์— ์™„๋ฃŒ๋ฉ๋‹ˆ๋‹ค. ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๋Š” HuggingFace Transformers(Wolf et al., 2020)์—์„œ ์‚ฌ์šฉํ•˜๋ฉฐ, LLaMA-2 7B ๋ฐ 13B์˜ ๊ฒฝ์šฐ ์ „์ฒด ์ •๋ฐ€๋„๋ฅผ, 70B์˜ ๊ฒฝ์šฐ ๋ฐ˜ ์ •๋ฐ€๋„๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์œ„ํ‚ค๋ฐ์ดํ„ฐ๋ฅผ ์ฟผ๋ฆฌํ•˜๊ธฐ ์œ„ํ•œ SPARQL ์ฟผ๋ฆฌ๋Š” GPT-4(OpenAI et al., 2023)์˜ ๋„์›€์œผ๋กœ ์ž‘์„ฑ๋ฉ๋‹ˆ๋‹ค.
Querying WikidataWe collect the fact triplets of the selected fact composition types through Wikidata Query Service5 with one handcrafted query for each of the 52 fact composition types. When there are too many results for the API call to bring before a timeout occurs, we reduce the number of the results by filtering the results with the number of reference links and/or adding other conditions to the query.
Wikidata ์ฟผ๋ฆฌํ•˜๊ธฐ ์šฐ๋ฆฌ๋Š” 52๊ฐœ์˜ ์‚ฌ์‹ค ๊ตฌ์„ฑ ์œ ํ˜• ๊ฐ๊ฐ์— ๋Œ€ํ•ด ํ•˜๋‚˜์˜ ์ˆ˜์ž‘์—… ์ฟผ๋ฆฌ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Wikidata Query Service5๋ฅผ ํ†ตํ•ด ์„ ํƒ๋œ ์‚ฌ์‹ค ๊ตฌ์„ฑ ์œ ํ˜•์˜ ์‚ฌ์‹ค ์‚ผ์ค‘ํ•ญ์„ ์ˆ˜์ง‘ํ•ฉ๋‹ˆ๋‹ค. API ํ˜ธ์ถœ๋กœ ์ธํ•ด ํƒ€์ž„์•„์›ƒ์ด ๋ฐœ์ƒํ•˜๊ธฐ ์ „์— ๋„ˆ๋ฌด ๋งŽ์€ ๊ฒฐ๊ณผ๊ฐ€ ๋ฐ˜ํ™˜๋˜๋Š” ๊ฒฝ์šฐ, ์ฐธ์กฐ ๋งํฌ ์ˆ˜๋กœ ๊ฒฐ๊ณผ๋ฅผ ํ•„ํ„ฐ๋งํ•˜๊ฑฐ๋‚˜ ์ฟผ๋ฆฌ์— ๋‹ค๋ฅธ ์กฐ๊ฑด์„ ์ถ”๊ฐ€ํ•˜์—ฌ ๊ฒฐ๊ณผ ์ˆ˜๋ฅผ ์ค„์ž…๋‹ˆ๋‹ค.
์œ„ํ‚ค๋ฐ์ดํ„ฐ ์ฟผ๋ฆฌ์„ ์ •ํ•œ 52๊ฐœ์˜ ์‚ฌ์‹ค ๊ตฌ์„ฑ ์œ ํ˜•์— ๋Œ€ํ•ด ๊ฐ๊ฐ 1๊ฐœ์˜ ์ˆ˜์ž‘์—… ์ฟผ๋ฆฌ๋ฅผ ํ†ตํ•ด Wikidata Query Service5๋ฅผ ํ†ตํ•ด ์‚ฌ์‹ค ์‚ผ์ค‘ ํ•ญ์„ ์ˆ˜์ง‘ํ•ฉ๋‹ˆ๋‹ค. API ํ˜ธ์ถœ์ด ํƒ€์ž„์•„์›ƒ์ด ๋ฐœ์ƒํ•˜๊ธฐ ์ „์— ๊ฐ€์ ธ์˜ฌ ๊ฒฐ๊ณผ๊ฐ€ ๋„ˆ๋ฌด ๋งŽ์œผ๋ฉด ์ฐธ์กฐ ๋งํฌ ์ˆ˜๋กœ ๊ฒฐ๊ณผ๋ฅผ ํ•„ํ„ฐ๋งํ•˜๊ณ /ํ•˜๊ฑฐ๋‚˜ ์ฟผ๋ฆฌ์— ๋‹ค๋ฅธ ์กฐ๊ฑด์„ ์ถ”๊ฐ€ํ•˜์—ฌ ๊ฒฐ๊ณผ ์ˆ˜๋ฅผ ์ค„์ž…๋‹ˆ๋‹ค.
Along this direction, researchers have developed numerous promising models (see Figure 1 for an overview of this field), where the key idea is to equip LLMs with crucial human capabilities like memory and planning to make them behave like humans and complete various tasks effectively. Previously, these models were proposed independently, with limited efforts made to summarize and compare them holistically.
์ด์™€ ๊ฐ™์€ ๋ฐฉํ–ฅ์œผ๋กœ, ์—ฐ๊ตฌ์ž๋“ค์€ ๋งŽ์€ ์œ ๋งํ•œ ๋ชจ๋ธ๋“ค์„ ๊ฐœ๋ฐœํ•ด์™”์Šต๋‹ˆ๋‹ค(์ด ๋ถ„์•ผ์— ๋Œ€ํ•œ ๊ฐœ์š”๋Š” Figure 1์„ ์ฐธ์กฐํ•˜์‹ญ์‹œ์˜ค). ์ฃผ์š” ์•„์ด๋””์–ด๋Š” LLM์— ๋ฉ”๋ชจ๋ฆฌ์™€ ๊ณ„ํš๊ณผ ๊ฐ™์€ ์ค‘์š”ํ•œ ์ธ๊ฐ„์˜ ๋Šฅ๋ ฅ์„ ์žฅ์ฐฉํ•˜์—ฌ ์ธ๊ฐ„์ฒ˜๋Ÿผ ํ–‰๋™ํ•˜๊ณ  ๋‹ค์–‘ํ•œ ์ž‘์—…์„ ํšจ๊ณผ์ ์œผ๋กœ ์™„๋ฃŒํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด์ „์—๋Š” ์ด๋Ÿฌํ•œ ๋ชจ๋ธ๋“ค์ด ๋…๋ฆฝ์ ์œผ๋กœ ์ œ์•ˆ๋˜์—ˆ์œผ๋ฉฐ, ์ด๋ฅผ ์ „์ฒด์ ์œผ๋กœ ์š”์•ฝํ•˜๊ณ  ๋น„๊ตํ•˜๋ ค๋Š” ๋…ธ๋ ฅ์€ ์ œํ•œ์ ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
์ด๋Ÿฌํ•œ ๋ฐฉํ–ฅ์œผ๋กœ ์—ฐ๊ตฌ์ž๋“ค์€ ๊ธฐ์–ต๋ ฅ๊ณผ ๊ณ„ํš๋ ฅ๊ณผ ๊ฐ™์€ ์ธ๊ฐ„์˜ ์ค‘์š”ํ•œ ๋Šฅ๋ ฅ์„ LLM์— ๊ฐ–์ถ”์–ด ์ธ๊ฐ„์ฒ˜๋Ÿผ ํ–‰๋™ํ•˜๊ณ  ๋‹ค์–‘ํ•œ ์ž‘์—…์„ ํšจ๊ณผ์ ์œผ๋กœ ์™„๋ฃŒํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ๊ฒƒ์ด ํ•ต์‹ฌ ์•„์ด๋””์–ด์ธ ์ˆ˜๋งŽ์€ ์œ ๋งํ•œ ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ–ˆ์Šต๋‹ˆ๋‹ค(์ด ๋ถ„์•ผ์— ๋Œ€ํ•œ ๊ฐœ์š”๋Š” ๊ทธ๋ฆผ 1 ์ฐธ์กฐ). ์ด์ „์—๋Š” ์ด๋Ÿฌํ•œ ๋ชจ๋ธ๋“ค์ด ๋…๋ฆฝ์ ์œผ๋กœ ์ œ์•ˆ๋˜์—ˆ์œผ๋ฉฐ, ์ด๋ฅผ ์ „์ฒด์ ์œผ๋กœ ์š”์•ฝํ•˜๊ณ  ๋น„๊ตํ•˜๋ ค๋Š” ๋…ธ๋ ฅ์€ ์ œํ•œ์ ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
Large Language Models (LLMs) have exhibited a profound understanding of natural language, improving performance on an array of tasks [5]. Using open web data has helped in creating general-purpose LLMs with a broad range of capabilities. General-purpose LLMs are however not "specialists"; for example, while LLMs could write good news articles, it would be hard-pressed to write specialized legal documents.
๋Œ€ํ˜• ์–ธ์–ด ๋ชจ๋ธ(LLMs)์€ ์ž์—ฐ์–ด์— ๋Œ€ํ•œ ๊นŠ์€ ์ดํ•ด๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ ๋‹ค์–‘ํ•œ ์ž‘์—…์—์„œ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค [5]. ๊ณต๊ฐœ ์›น ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•จ์œผ๋กœ์จ ๊ด‘๋ฒ”์œ„ํ•œ ๊ธฐ๋Šฅ์„ ๊ฐ–์ถ˜ ๋ฒ”์šฉ LLM์„ ๋งŒ๋“œ๋Š” ๋ฐ ๋„์›€์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฒ”์šฉ LLM์€ "์ „๋ฌธ๊ฐ€"๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, LLM์ด ์ข‹์€ ๋‰ด์Šค ๊ธฐ์‚ฌ๋ฅผ ์ž‘์„ฑํ•  ์ˆ˜๋Š” ์žˆ์ง€๋งŒ, ์ „๋ฌธ์ ์ธ ๋ฒ•๋ฅ  ๋ฌธ์„œ๋ฅผ ์ž‘์„ฑํ•˜๋Š” ๊ฒƒ์€ ์–ด๋ ค์šธ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์€ ์ž์—ฐ์–ด์— ๋Œ€ํ•œ ๊นŠ์€ ์ดํ•ด๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ ๋‹ค์–‘ํ•œ ์ž‘์—…์—์„œ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค[5]. ์˜คํ”ˆ ์›น ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ๊ด‘๋ฒ”์œ„ํ•œ ๊ธฐ๋Šฅ์„ ๊ฐ–์ถ˜ ๋ฒ”์šฉ LLM์„ ๋งŒ๋“œ๋Š” ๋ฐ ๋„์›€์ด ๋ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฒ”์šฉ LLM์€ '์ „๋ฌธ๊ฐ€'๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, LLM์€ ์ข‹์€ ๋‰ด์Šค ๊ธฐ์‚ฌ๋ฅผ ์ž‘์„ฑํ•  ์ˆ˜๋Š” ์žˆ์ง€๋งŒ ์ „๋ฌธ์ ์ธ ๋ฒ•๋ฅ  ๋ฌธ์„œ๋ฅผ ์ž‘์„ฑํ•˜๋Š” ๋ฐ๋Š” ์–ด๋ ค์›€์ด ์žˆ์Šต๋‹ˆ๋‹ค.
Further detailed results across different platforms and categories can be found in Table 4. Across categories, our system exhibits noticeable variations: on Sports, our system nearly matches the crowd aggregate, and on Environment & Energy, it falls much behind. However, we caution against drawing strong conclusions from subcategories, since the sample size is smaller and variation could be due to noise.
๋‹ค์–‘ํ•œ ํ”Œ๋žซํผ๊ณผ ์นดํ…Œ๊ณ ๋ฆฌ์— ๊ฑธ์นœ ๋” ์ž์„ธํ•œ ๊ฒฐ๊ณผ๋Š” Table 4์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์นดํ…Œ๊ณ ๋ฆฌ๋ณ„๋กœ ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์€ ๋ˆˆ์— ๋„๋Š” ๋ณ€ํ™”๋ฅผ ๋ณด์ž…๋‹ˆ๋‹ค: ์Šคํฌ์ธ ์—์„œ๋Š” ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์ด ๊ตฐ์ค‘ ์ง‘ํ•ฉ๊ณผ ๊ฑฐ์˜ ์ผ์น˜ํ•˜์ง€๋งŒ, ํ™˜๊ฒฝ ๋ฐ ์—๋„ˆ์ง€์—์„œ๋Š” ํ›จ์”ฌ ๋’ค์ฒ˜์ง‘๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ‘œ๋ณธ ํฌ๊ธฐ๊ฐ€ ์ž‘๊ณ  ๋ณ€๋™์ด ์žก์Œ์— ๊ธฐ์ธํ•  ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ ํ•˜์œ„ ์นดํ…Œ๊ณ ๋ฆฌ์—์„œ ๊ฐ•ํ•œ ๊ฒฐ๋ก ์„ ๋„์ถœํ•˜๋Š” ๊ฒƒ์€ ์ฃผ์˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
๋‹ค์–‘ํ•œ ํ”Œ๋žซํผ๊ณผ ์นดํ…Œ๊ณ ๋ฆฌ์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๊ฒฐ๊ณผ๋Š” ํ‘œ 4์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์นดํ…Œ๊ณ ๋ฆฌ ์ „๋ฐ˜์— ๊ฑธ์ณ ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์€ ๋ˆˆ์— ๋„๋Š” ์ฐจ์ด๋ฅผ ๋ณด์ž…๋‹ˆ๋‹ค. ์Šคํฌ์ธ ์—์„œ๋Š” ์šฐ๋ฆฌ ์‹œ์Šคํ…œ์ด ๊ตฐ์ค‘ ์ง‘๊ณ„์™€ ๊ฑฐ์˜ ์ผ์น˜ํ•˜๊ณ , ํ™˜๊ฒฝ ๋ฐ ์—๋„ˆ์ง€์—์„œ๋Š” ํ›จ์”ฌ ๋’ค์ฒ˜์ง‘๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ•˜์œ„ ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๊ฒฝ์šฐ ํ‘œ๋ณธ ํฌ๊ธฐ๊ฐ€ ์ž‘๊ณ  ๋ณ€๋™์ด ๋…ธ์ด์ฆˆ๋กœ ์ธํ•œ ๊ฒƒ์ผ ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ ๊ฐ•ํ•œ ๊ฒฐ๋ก ์„ ๋‚ด๋ฆฌ์ง€ ๋ง์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค.
Objective Structured Clinical Examination (OSCE) is a practical assessment format used in healthcare to assess clinical skills and competencies in a standardized and objective fashion [31, 32, 33]. It differs from traditional written or oral exams that focus primarily on theoretical knowledge and instead aims to provide an environment in which the skills of real-world clinical practice might be assessed.
๊ฐ๊ด€์  ๊ตฌ์กฐํ™” ์ž„์ƒ์‹œํ—˜ (OSCE)์€ ์ž„์ƒ ๊ธฐ์ˆ ๊ณผ ์—ญ๋Ÿ‰์„ ํ‘œ์ค€ํ™”๋˜๊ณ  ๊ฐ๊ด€์ ์ธ ๋ฐฉ์‹์œผ๋กœ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ์˜๋ฃŒ ๋ถ„์•ผ์—์„œ ์‚ฌ์šฉ๋˜๋Š” ์‹ค๊ธฐ ํ‰๊ฐ€ ํ˜•์‹์ž…๋‹ˆ๋‹ค [31, 32, 33]. ์ด๋Š” ์ฃผ๋กœ ์ด๋ก ์  ์ง€์‹์— ์ค‘์ ์„ ๋‘” ์ „ํ†ต์ ์ธ ํ•„๊ธฐ ๋˜๋Š” ๊ตฌ์ˆ  ์‹œํ—˜๊ณผ๋Š” ๋‹ฌ๋ฆฌ, ์‹ค์ œ ์ž„์ƒ ์‹ค์Šต ๊ธฐ์ˆ ์„ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๋Š” ํ™˜๊ฒฝ์„ ์ œ๊ณตํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
๊ฐ๊ด€์  ๊ตฌ์กฐํ™”๋œ ์ž„์ƒ ์‹œํ—˜(OSCE)์€ ํ‘œ์ค€ํ™”๋˜๊ณ  ๊ฐ๊ด€์ ์ธ ๋ฐฉ์‹์œผ๋กœ ์ž„์ƒ ๊ธฐ์ˆ ๊ณผ ์—ญ๋Ÿ‰์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ์˜๋ฃŒ ๋ถ„์•ผ์—์„œ ์‚ฌ์šฉ๋˜๋Š” ์‹ค์Šต ํ‰๊ฐ€ ํ˜•์‹์ž…๋‹ˆ๋‹ค[31, 32, 33]. ์ด๋Š” ์ฃผ๋กœ ์ด๋ก ์  ์ง€์‹์— ์ดˆ์ ์„ ๋งž์ถ˜ ๊ธฐ์กด์˜ ์„œ๋ฉด ๋˜๋Š” ๊ตฌ์ˆ  ์‹œํ—˜๊ณผ๋Š” ๋‹ค๋ฅด๋ฉฐ, ์‹ค์ œ ์ž„์ƒ ์‹ค๋ฌด ๊ธฐ์ˆ ์„ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๋Š” ํ™˜๊ฒฝ์„ ์ œ๊ณตํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
Figure 18: A response from our fine-tuned model on the question โ€œWill Reddit announce changes or a delay to its proposed API fee pricing before July 1, 2023?โ€ (resolved NO). The forecast was made at the retrieval date June, 14, 2023, one day after the question was opened. At the time, the community gives a forecast of 40%, where this response suggests 20% which turned out to be closer to the resolution.
Figure 18: "Reddit๊ฐ€ 2023๋…„ 7์›” 1์ผ ์ด์ „์— ์ œ์•ˆ๋œ API ์š”๊ธˆ ๊ฐ€๊ฒฉ์— ๋Œ€ํ•œ ๋ณ€๊ฒฝ ์‚ฌํ•ญ ๋˜๋Š” ์ง€์—ฐ์„ ๋ฐœํ‘œํ•  ๊ฒƒ์ธ๊ฐ€?"๋ผ๋Š” ์งˆ๋ฌธ์— ๋Œ€ํ•ด ๋ฏธ์„ธ ์กฐ์ •๋œ ๋ชจ๋ธ์˜ ์‘๋‹ต (๊ฒฐ๊ณผ๋Š” NO). ์˜ˆ์ธก์€ ์งˆ๋ฌธ์ด ์—ด๋ฆฐ ๋‹ค์Œ ๋‚ ์ธ 2023๋…„ 6์›” 14์ผ์— ์ด๋ฃจ์–ด์กŒ์Šต๋‹ˆ๋‹ค. ๋‹น์‹œ ์ปค๋ฎค๋‹ˆํ‹ฐ๋Š” 40%์˜ ์˜ˆ์ธก์„ ํ–ˆ์œผ๋ฉฐ, ์ด ์‘๋‹ต์€ 20%๋ฅผ ์ œ์‹œํ–ˆ๋Š”๋ฐ, ์ด๋Š” ์‹ค์ œ ๊ฒฐ๊ณผ์— ๋” ๊ฐ€๊นŒ์› ์Šต๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 18: "Reddit๊ฐ€ 2023๋…„ 7์›” 1์ผ ์ด์ „์— ์ œ์•ˆ๋œ API ์š”๊ธˆ ๊ฐ€๊ฒฉ ์ฑ…์ •์— ๋Œ€ํ•œ ๋ณ€๊ฒฝ ๋˜๋Š” ์—ฐ๊ธฐ๋ฅผ ๋ฐœํ‘œํ•  ๊ฒƒ์ธ๊ฐ€?"๋ผ๋Š” ์งˆ๋ฌธ์— ๋Œ€ํ•œ ๋ฏธ์„ธ ์กฐ์ •๋œ ๋ชจ๋ธ์˜ ์‘๋‹ต์ž…๋‹ˆ๋‹ค(ํ•ด๊ฒฐ๋จ). ์ด ์˜ˆ์ธก์€ ์งˆ๋ฌธ์ด ๊ฐœ์„ค๋œ ์ง€ ํ•˜๋ฃจ ํ›„์ธ 2023๋…„ 6์›” 14์ผ์— ๊ฒ€์ƒ‰๋œ ๋‚ ์งœ์— ์ด๋ฃจ์–ด์กŒ์Šต๋‹ˆ๋‹ค. ๋‹น์‹œ ์ปค๋ฎค๋‹ˆํ‹ฐ๋Š” 40%์˜ ์˜ˆ์ธก์„ ๋‚ด๋ ธ๋Š”๋ฐ, ์ด ์‘๋‹ต์€ 20%๋ฅผ ์ œ์‹œํ•˜์—ฌ ๊ฒฐ๋ก ์— ๋” ๊ฐ€๊นŒ์šด ๊ฒƒ์œผ๋กœ ํŒ๋ช…๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
Reference model.The reference model is used to calculate the log probabilities of the preference dataset. For each step, only a subset of the total data is used and the reference model is initialized as \(M_{t-1}\), _i.e,_ the aligned model from the previous step. The initial reference model is set as \(S\), the SFT base model. This results in using a more aligned reference model than conventional DPO.
์ฐธ์กฐ ๋ชจ๋ธ. ์ฐธ์กฐ ๋ชจ๋ธ์€ ์„ ํ˜ธ ๋ฐ์ดํ„ฐ์…‹์˜ ๋กœ๊ทธ ํ™•๋ฅ ์„ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ๊ฐ ๋‹จ๊ณ„์—์„œ๋Š” ์ „์ฒด ๋ฐ์ดํ„ฐ์˜ ์ผ๋ถ€๋งŒ ์‚ฌ์šฉ๋˜๋ฉฐ, ์ฐธ์กฐ ๋ชจ๋ธ์€ ์ด์ „ ๋‹จ๊ณ„์˜ ์ •๋ ฌ๋œ ๋ชจ๋ธ์ธ \(M_{t-1}\)๋กœ ์ดˆ๊ธฐํ™”๋ฉ๋‹ˆ๋‹ค. ์ดˆ๊ธฐ ์ฐธ์กฐ ๋ชจ๋ธ์€ SFT ๊ธฐ๋ณธ ๋ชจ๋ธ์ธ \(S\)๋กœ ์„ค์ •๋ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๊ธฐ์กด์˜ DPO๋ณด๋‹ค ๋” ์ •๋ ฌ๋œ ์ฐธ์กฐ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
์ฐธ์กฐ ๋ชจ๋ธ. ์ฐธ์กฐ ๋ชจ๋ธ์€ ์„ ํ˜ธ๋„ ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๋กœ๊ทธ ํ™•๋ฅ ์„ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ๊ฐ ๋‹จ๊ณ„๋งˆ๋‹ค ์ „์ฒด ๋ฐ์ดํ„ฐ์˜ ํ•˜์œ„ ์ง‘ํ•ฉ๋งŒ ์‚ฌ์šฉ๋˜๋ฉฐ ์ฐธ์กฐ ๋ชจ๋ธ์€ \(M_{t-1}\), _์ฆ‰, ์ด์ „ ๋‹จ๊ณ„์˜ ์ •๋ ฌ๋œ ๋ชจ๋ธ๋กœ ์ดˆ๊ธฐํ™”๋ฉ๋‹ˆ๋‹ค. ์ดˆ๊ธฐ ์ฐธ์กฐ ๋ชจ๋ธ์€ SFT ๊ธฐ๋ณธ ๋ชจ๋ธ์ธ \(S\)๋กœ ์„ค์ •๋ฉ๋‹ˆ๋‹ค. ์ด๋ ‡๊ฒŒ ํ•˜๋ฉด ๊ธฐ์กด DPO๋ณด๋‹ค ๋” ์ •๋ ฌ๋œ ์ฐธ์กฐ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Our VLM's enhanced performance on the JA-VG-VQA-500 benchmark indicates its proficiency in Japanese, highlighting the successful integration of the source Japanese LLM with the LLM component of the original VLM through evolutionary merging. Furthermore, our VLM's superior results on the JA-VLM-Bench-In-the-Wild compared to both baselines exhibits its adeptness at navigating culturally-specific content.
์šฐ๋ฆฌ VLM์ด JA-VG-VQA-500 ๋ฒค์น˜๋งˆํฌ์—์„œ ํ–ฅ์ƒ๋œ ์„ฑ๋Šฅ์„ ๋ณด์ธ ๊ฒƒ์€ ์ผ๋ณธ์–ด์— ๋Œ€ํ•œ ์ˆ™๋ จ๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ด๋Š” ์†Œ์Šค ์ผ๋ณธ์–ด LLM๊ณผ ์›๋ž˜ VLM์˜ LLM ๊ตฌ์„ฑ ์š”์†Œ๋ฅผ ์ง„ํ™”์  ๋ณ‘ํ•ฉ์„ ํ†ตํ•ด ์„ฑ๊ณต์ ์œผ๋กœ ํ†ตํ•ฉํ•œ ๊ฒฐ๊ณผ์ž…๋‹ˆ๋‹ค. ๋˜ํ•œ, JA-VLM-Bench-In-the-Wild์—์„œ ๋‘ ๊ฐ€์ง€ ๊ธฐ์ค€์„ ๊ณผ ๋น„๊ตํ•˜์—ฌ ์šฐ์ˆ˜ํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์ธ ๊ฒƒ์€ ๋ฌธํ™”์ ์œผ๋กœ ํŠนํ™”๋œ ์ฝ˜ํ…์ธ ๋ฅผ ์ž˜ ๋‹ค๋ฃฐ ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
JA-VG-VQA-500 ๋ฒค์น˜๋งˆํฌ์—์„œ VLM์˜ ํ–ฅ์ƒ๋œ ์„ฑ๋Šฅ์€ ์ผ๋ณธ์–ด์— ๋Œ€ํ•œ ์ˆ™๋ จ๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ์ง„ํ™”์  ๋ณ‘ํ•ฉ์„ ํ†ตํ•ด ์›๋ž˜ VLM์˜ LLM ๊ตฌ์„ฑ ์š”์†Œ์™€ ์†Œ์Šค ์ผ๋ณธ์–ด LLM์„ ์„ฑ๊ณต์ ์œผ๋กœ ํ†ตํ•ฉํ–ˆ์Œ์„ ๊ฐ•์กฐํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ๊ธฐ์ค€๊ณผ ๋น„๊ตํ–ˆ์„ ๋•Œ JA-VLM-Bench-In-the-Wild์—์„œ ๋” ์šฐ์ˆ˜ํ•œ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ฃผ๋Š” VLM์€ ๋ฌธํ™”์ ์œผ๋กœ ํŠน์ •ํ•œ ์ฝ˜ํ…์ธ ๋ฅผ ํƒ์ƒ‰ํ•˜๋Š” ๋ฐ ๋Šฅ์ˆ™ํ•จ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
ToT is particularly useful in complex problem-solving scenarios where a single line of reasoning might not suffice. It allows LLMs to mimic a more human-like problem-solving approach, considering a range of possibilities before arriving at a conclusion. This technique enhances the model's ability to handle ambiguity, complexity, and nuanced tasks, making it a valuable tool in advanced AI applications.
ToT๋Š” ๋‹จ์ผํ•œ ์‚ฌ๊ณ  ๊ณผ์ •์œผ๋กœ๋Š” ์ถฉ๋ถ„ํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ๋Š” ๋ณต์žกํ•œ ๋ฌธ์ œ ํ•ด๊ฒฐ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ํŠนํžˆ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” LLM์ด ๊ฒฐ๋ก ์— ๋„๋‹ฌํ•˜๊ธฐ ์ „์— ๋‹ค์–‘ํ•œ ๊ฐ€๋Šฅ์„ฑ์„ ๊ณ ๋ คํ•˜์—ฌ ๋ณด๋‹ค ์ธ๊ฐ„์ ์ธ ๋ฌธ์ œ ํ•ด๊ฒฐ ์ ‘๊ทผ ๋ฐฉ์‹์„ ๋ชจ๋ฐฉํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์€ ๋ชจ๋ธ์ด ๋ชจํ˜ธ์„ฑ, ๋ณต์žก์„ฑ ๋ฐ ๋ฏธ๋ฌ˜ํ•œ ์ž‘์—…์„ ์ฒ˜๋ฆฌํ•˜๋Š” ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œ์ผœ, ๊ณ ๊ธ‰ AI ์‘์šฉ ํ”„๋กœ๊ทธ๋žจ์—์„œ ๊ท€์ค‘ํ•œ ๋„๊ตฌ๊ฐ€ ๋ฉ๋‹ˆ๋‹ค.
ToT๋Š” ํ•œ ์ค„์˜ ์ถ”๋ก ์œผ๋กœ๋Š” ํ•ด๊ฒฐํ•  ์ˆ˜ ์—†๋Š” ๋ณต์žกํ•œ ๋ฌธ์ œ ํ•ด๊ฒฐ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ํŠนํžˆ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด LLM์€ ๋‹ค์–‘ํ•œ ๊ฐ€๋Šฅ์„ฑ์„ ๊ณ ๋ คํ•œ ํ›„ ๊ฒฐ๋ก ์— ๋„๋‹ฌํ•˜๊ธฐ ์ „์— ๋ณด๋‹ค ์ธ๊ฐ„๊ณผ ์œ ์‚ฌํ•œ ๋ฌธ์ œ ํ•ด๊ฒฐ ๋ฐฉ์‹์„ ๋ชจ๋ฐฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์€ ๋ชจํ˜ธ์„ฑ, ๋ณต์žก์„ฑ ๋ฐ ๋ฏธ๋ฌ˜ํ•œ ์ž‘์—…์„ ์ฒ˜๋ฆฌํ•˜๋Š” ๋ชจ๋ธ์˜ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œ์ผœ ๊ณ ๊ธ‰ AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์—์„œ ์œ ์šฉํ•œ ๋„๊ตฌ๊ฐ€ ๋ฉ๋‹ˆ๋‹ค.
We thank Song Duong for carefully reviewing this article and for his valuable suggestions. This work was funded by the last author's chair in the PRAIRIE institute funded by the French national agency ANR as part of the "Investissements d'avenir" programme under the reference ANR-19-P3IA-0001. This work was granted access to the HPC resources of GENCI operated by IDRIS (allocation 2023-AD011013680R1).
์šฐ๋ฆฌ๋Š” Song Duong์ด ์ด ๊ธฐ์‚ฌ๋ฅผ ์‹ ์ค‘ํ•˜๊ฒŒ ๊ฒ€ํ† ํ•˜๊ณ  ๊ท€์ค‘ํ•œ ์ œ์•ˆ์„ ํ•ด์ค€ ๊ฒƒ์— ๋Œ€ํ•ด ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” "Investissements d'avenir" ํ”„๋กœ๊ทธ๋žจ์˜ ์ผํ™˜์œผ๋กœ ํ”„๋ž‘์Šค ๊ตญ๊ฐ€ ๊ธฐ๊ด€ ANR์ด ์ž๊ธˆ์„ ์ง€์›ํ•œ PRAIRIE ์—ฐ๊ตฌ์†Œ์˜ ๋งˆ์ง€๋ง‰ ์ €์ž์˜ ์˜์žฅ์— ์˜ํ•ด ์ž๊ธˆ์„ ์ง€์›๋ฐ›์•˜์Šต๋‹ˆ๋‹ค(์ฐธ์กฐ ๋ฒˆํ˜ธ ANR-19-P3IA-0001). ์ด ์—ฐ๊ตฌ๋Š” IDRIS๊ฐ€ ์šด์˜ํ•˜๋Š” GENCI์˜ HPC ์ž์›์— ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋Š” ๊ถŒํ•œ์„ ๋ถ€์—ฌ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค(ํ• ๋‹น ๋ฒˆํ˜ธ 2023-AD011013680R1).
์ด ๊ธ€์„ ๊ผผ๊ผผํ•˜๊ฒŒ ๊ฒ€ํ† ํ•˜๊ณ  ๊ท€์ค‘ํ•œ ์ œ์•ˆ์„ ํ•ด์ฃผ์‹  ์†ก๋‘์˜น์—๊ฒŒ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” ํ”„๋ž‘์Šค ๊ตญ๊ฐ€๊ธฐ๊ด€ ANR์ด "Investissements d'avenir" ํ”„๋กœ๊ทธ๋žจ์˜ ์ผํ™˜์œผ๋กœ ์ง€์›ํ•˜๋Š” PRAIRIE ์—ฐ๊ตฌ์†Œ์˜ ๋งˆ์ง€๋ง‰ ์ €์ž์˜ ์˜์ž์— ์˜ํ•ด ์ž๊ธˆ์ด ์ง€์›๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” GENCI์˜ HPC ๋ฆฌ์†Œ์Šค์— ๋Œ€ํ•œ ์•ก์„ธ์Šค๋ฅผ ํ—ˆ์šฉ๋ฐ›์•˜์œผ๋ฉฐ, ์ด ๋ฆฌ์†Œ์Šค๋Š” IDRIS(ํ• ๋‹น 2023-AD011013680R1)์—์„œ ์šด์˜๋ฉ๋‹ˆ๋‹ค.
When evaluating the strengths and weaknesses of different ethical theories, it's important to consider their underlying assumptions, their implications for decision-making, and their ability to address the specific ethical dilemmas of a given context. It's also important to consider the practicality of implementing the theory and the degree to which it aligns with our own personal values and beliefs.
๋‹ค์–‘ํ•œ ์œค๋ฆฌ ์ด๋ก ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ํ‰๊ฐ€ํ•  ๋•Œ, ๊ทธ ์ด๋ก ๋“ค์ด ๊ธฐ๋ฐ˜ํ•˜๊ณ  ์žˆ๋Š” ๊ฐ€์ •, ์˜์‚ฌ ๊ฒฐ์ •์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ, ํŠน์ •ํ•œ ์œค๋ฆฌ์  ๋”œ๋ ˆ๋งˆ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” ๋Šฅ๋ ฅ์„ ๊ณ ๋ คํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด๋ก ์„ ์‹คํ–‰์— ์˜ฎ๊ธฐ๋Š” ์‹ค์šฉ์„ฑ๊ณผ ์šฐ๋ฆฌ์˜ ๊ฐœ์ธ์ ์ธ ๊ฐ€์น˜๊ด€ ๋ฐ ์‹ ๋…๊ณผ์˜ ์ผ์น˜ ์ •๋„๋„ ๊ณ ๋ คํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.
๋‹ค์–‘ํ•œ ์œค๋ฆฌ ์ด๋ก ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ํ‰๊ฐ€ํ•  ๋•Œ๋Š” ๊ทธ ์ด๋ก ์˜ ๊ทผ๋ณธ ๊ฐ€์ •, ์˜์‚ฌ ๊ฒฐ์ •์— ๋Œ€ํ•œ ์‹œ์‚ฌ์ , ์ฃผ์–ด์ง„ ์ƒํ™ฉ์˜ ํŠน์ • ์œค๋ฆฌ์  ๋”œ๋ ˆ๋งˆ๋ฅผ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์„ ๊ณ ๋ คํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด๋ก ์„ ์‹คํ–‰ํ•˜๋Š” ๋ฐ ์žˆ์–ด ์‹ค์šฉ์„ฑ๊ณผ ๊ฐœ์ธ์˜ ๊ฐ€์น˜์™€ ์‹ ๋…์— ๋ถ€ํ•ฉํ•˜๋Š” ์ •๋„๋ฅผ ๊ณ ๋ คํ•˜๋Š” ๊ฒƒ๋„ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค.
Our idealized setting, free from irrelevant data, allows for more accurate scaling law computations -- we also discuss how "junk" data affects capacity later in Section 10. In contrast, it is difficult to quantify real-life knowledge; for instance, if LLaMA-70B outperforms LLaMA-7B by 30% on a benchmark, it doesn't necessarily mean a tenfold model scaling only boosts capacity by 30% (see Footnote 1).
์šฐ๋ฆฌ์˜ ์ด์ƒ์ ์ธ ์„ค์ •์€ ๊ด€๋ จ ์—†๋Š” ๋ฐ์ดํ„ฐ๊ฐ€ ์—†๋Š” ์ƒํƒœ์—์„œ ๋” ์ •ํ™•ํ•œ ์Šค์ผ€์ผ๋ง ๋ฒ•์น™ ๊ณ„์‚ฐ์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค -- ๋˜ํ•œ "์“ธ๋ชจ์—†๋Š”" ๋ฐ์ดํ„ฐ๊ฐ€ ์šฉ๋Ÿ‰์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์— ๋Œ€ํ•ด์„œ๋Š” ๋‚˜์ค‘์— 10์žฅ์—์„œ ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ์‹ค์ œ ์ง€์‹์„ ์ •๋Ÿ‰ํ™”ํ•˜๋Š” ๊ฒƒ์€ ์–ด๋ ต์Šต๋‹ˆ๋‹ค; ์˜ˆ๋ฅผ ๋“ค์–ด, LLaMA-70B๊ฐ€ ๋ฒค์น˜๋งˆํฌ์—์„œ LLaMA-7B๋ณด๋‹ค 30% ๋” ์šฐ์ˆ˜ํ•˜๋‹ค๊ณ  ํ•ด์„œ, ๋ชจ๋ธ ํฌ๊ธฐ๋ฅผ 10๋ฐฐ๋กœ ๋Š˜๋ฆฌ๋Š” ๊ฒƒ์ด ์šฉ๋Ÿ‰์„ 30%๋งŒ ์ฆ๊ฐ€์‹œํ‚จ๋‹ค๋Š” ์˜๋ฏธ๋Š” ์•„๋‹™๋‹ˆ๋‹ค (Footnote 1์„ ์ฐธ์กฐํ•˜์‹ญ์‹œ์˜ค).
๊ด€๋ จ ์—†๋Š” ๋ฐ์ดํ„ฐ๊ฐ€ ์—†๋Š” ์ด์ƒํ™”๋œ ํ™˜๊ฒฝ์—์„œ๋Š” ๋ณด๋‹ค ์ •ํ™•ํ•œ ์Šค์ผ€์ผ๋ง ๋ฒ•์น™ ๊ณ„์‚ฐ์ด ๊ฐ€๋Šฅํ•˜๋ฉฐ, ์„น์…˜ 10 ํ›„๋ฐ˜๋ถ€์— "์ •ํฌ" ๋ฐ์ดํ„ฐ๊ฐ€ ์šฉ๋Ÿ‰์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์— ๋Œ€ํ•ด์„œ๋„ ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ์‹ค์ œ ์ง€์‹์„ ์ •๋Ÿ‰ํ™”ํ•˜๊ธฐ๋Š” ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, LLaMA-70B๊ฐ€ ๋ฒค์น˜๋งˆํฌ์—์„œ LLaMA-7B๋ณด๋‹ค 30% ๋” ์šฐ์ˆ˜ํ•˜๋‹ค๊ณ  ํ•ด์„œ ๋ฐ˜๋“œ์‹œ 10๋ฐฐ์˜ ๋ชจ๋ธ ํฌ๊ธฐ๊ฐ€ ์šฉ๋Ÿ‰์„ 30%๋งŒ ํ–ฅ์ƒ์‹œํ‚จ๋‹ค๋Š” ์˜๋ฏธ๋Š” ์•„๋‹™๋‹ˆ๋‹ค(์ฃผ 1 ์ฐธ์กฐ).
Footnote 9: All attributes, except for the working city (determined by the employerโ€™s headquarters), are chosen uniformly and independently at random. There are \(N_{0}=400\times 400\times 1000\) possible person names, \(12\times 28\times 200\) birth dates, 200 birth cities, 300 universities, 100 majors, and 263 employers. Additionally, a random pronoun with 2 possibilities is chosen for each person.
Footnote 9: ๊ทผ๋ฌด ๋„์‹œ(๊ณ ์šฉ์ฃผ์˜ ๋ณธ์‚ฌ ์œ„์น˜์— ๋”ฐ๋ผ ๊ฒฐ์ •๋จ)๋ฅผ ์ œ์™ธํ•œ ๋ชจ๋“  ์†์„ฑ์€ ๊ท ์ผํ•˜๊ณ  ๋…๋ฆฝ์ ์œผ๋กœ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค. ๊ฐ€๋Šฅํ•œ ์‚ฌ๋žŒ ์ด๋ฆ„์€ \(N_{0}=400\times 400\times 1000\)๊ฐœ, ์ƒ๋…„์›”์ผ์€ \(12\times 28\times 200\)๊ฐœ, ์ถœ์ƒ ๋„์‹œ๋Š” 200๊ฐœ, ๋Œ€ํ•™๊ต๋Š” 300๊ฐœ, ์ „๊ณต์€ 100๊ฐœ, ๊ณ ์šฉ์ฃผ๋Š” 263๊ฐœ์ž…๋‹ˆ๋‹ค. ์ถ”๊ฐ€๋กœ, ๊ฐ ์‚ฌ๋žŒ์—๊ฒŒ๋Š” 2๊ฐ€์ง€ ๊ฐ€๋Šฅ์„ฑ ์ค‘ ํ•˜๋‚˜์˜ ๋ฌด์ž‘์œ„ ๋Œ€๋ช…์‚ฌ๊ฐ€ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค.
๊ฐ์ฃผ 9: ๊ทผ๋ฌด ๋„์‹œ(๊ณ ์šฉ์ฃผ ๋ณธ์‚ฌ์—์„œ ๊ฒฐ์ •)๋ฅผ ์ œ์™ธํ•œ ๋ชจ๋“  ์†์„ฑ์€ ๊ท ์ผํ•˜๊ณ  ๋…๋ฆฝ์ ์œผ๋กœ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค. ๊ฐ€๋Šฅํ•œ ์‚ฌ๋žŒ ์ด๋ฆ„์€ \(N_{0}=400\times 400\times 1000\), ์ƒ๋…„์›”์ผ์€ \(12\times 28\times 200\), ์ถœ์ƒ ๋„์‹œ๋Š” 200๊ฐœ, ๋Œ€ํ•™์€ 300๊ฐœ, ์ „๊ณต์€ 100๊ฐœ, ๊ณ ์šฉ์ฃผ๋Š” 263๊ฐœ์ž…๋‹ˆ๋‹ค. ๋˜ํ•œ ๊ฐ ์‚ฌ๋žŒ์— ๋Œ€ํ•ด 2๊ฐ€์ง€ ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋Š” ์ž„์˜์˜ ๋Œ€๋ช…์‚ฌ๊ฐ€ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค.
Recently, the idea of adapting LLMs to particular domains and updating their knowledge has become increasingly common (Yu et al., 2022). Various models have been suggested to improve factual knowledge and capabilities in diverse fields such as healthcare (Singhal et al., 2023; Wu et al., 2023; Wu et al., 2023), finance (Wu et al., 2023; Yang et al., 2023), and law (Huang et al., 2023; Nguyen, 2023).
์ตœ๊ทผ LLM์„ ํŠน์ • ๋„๋ฉ”์ธ์— ๋งž์ถ”๊ณ  ์ง€์‹์„ ์—…๋ฐ์ดํŠธํ•˜๋Š” ์•„์ด๋””์–ด๊ฐ€ ์ ์  ๋” ์ผ๋ฐ˜ํ™”๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค (Yu et al., 2022). ๋‹ค์–‘ํ•œ ๋ชจ๋ธ๋“ค์ด ํ—ฌ์Šค์ผ€์–ด (Singhal et al., 2023; Wu et al., 2023; Wu et al., 2023), ๊ธˆ์œต (Wu et al., 2023; Yang et al., 2023), ๋ฒ•๋ฅ  (Huang et al., 2023; Nguyen, 2023) ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ์‚ฌ์‹ค์  ์ง€์‹๊ณผ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด ์ œ์•ˆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
์ตœ๊ทผ์—๋Š” LLM์„ ํŠน์ • ๋„๋ฉ”์ธ์— ๋งž๊ฒŒ ์กฐ์ •ํ•˜๊ณ  ์ง€์‹์„ ์—…๋ฐ์ดํŠธํ•˜๋Š” ์•„์ด๋””์–ด๊ฐ€ ์ ์  ๋” ๋ณดํŽธํ™”๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค(Yu et al., 2022). ์˜๋ฃŒ(Singhal et al., 2023; Wu et al., 2023; Wu et al., 2023), ๊ธˆ์œต(Wu et al., 2023; Yang et al., 2023), ๋ฒ•๋ฅ (Huang et al., 2023; Nguyen, 2023) ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ์‚ฌ์‹ค์  ์ง€์‹๊ณผ ์—ญ๋Ÿ‰์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•œ ๋‹ค์–‘ํ•œ ๋ชจ๋ธ์ด ์ œ์•ˆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
**Model scale:** Due to compute limitations, the largest models we evaluated were 6.7B parameters trained on 100B tokens. While, to our knowledge, this is the largest to date application of embedding based data curation approaches, further investigation at model scales exceeding 100B would be very interesting, particularly in light of our observation that the efficiency gain grows with model scale.
**๋ชจ๋ธ ๊ทœ๋ชจ:** ๊ณ„์‚ฐ ์ œํ•œ์œผ๋กœ ์ธํ•ด ์šฐ๋ฆฌ๊ฐ€ ํ‰๊ฐ€ํ•œ ๊ฐ€์žฅ ํฐ ๋ชจ๋ธ์€ 100B ํ† ํฐ์œผ๋กœ ํ•™์Šต๋œ 6.7B ๋งค๊ฐœ๋ณ€์ˆ˜ ๋ชจ๋ธ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๊ฐ€ ์•„๋Š” ํ•œ, ์ด๊ฒƒ์€ ์ž„๋ฒ ๋”ฉ ๊ธฐ๋ฐ˜ ๋ฐ์ดํ„ฐ ํ๋ ˆ์ด์…˜ ์ ‘๊ทผ๋ฒ•์˜ ๊ฐ€์žฅ ํฐ ์ ์šฉ ์‚ฌ๋ก€์ž…๋‹ˆ๋‹ค. ๋ชจ๋ธ ๊ทœ๋ชจ๊ฐ€ 100B๋ฅผ ์ดˆ๊ณผํ•˜๋Š” ๊ฒฝ์šฐ์— ๋Œ€ํ•œ ์ถ”๊ฐ€ ์กฐ์‚ฌ๋Š” ๋งค์šฐ ํฅ๋ฏธ๋กœ์šธ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํŠนํžˆ, ํšจ์œจ์„ฑ ํ–ฅ์ƒ์ด ๋ชจ๋ธ ๊ทœ๋ชจ์™€ ํ•จ๊ป˜ ์ฆ๊ฐ€ํ•œ๋‹ค๋Š” ์šฐ๋ฆฌ์˜ ๊ด€์ฐฐ์„ ๊ณ ๋ คํ•  ๋•Œ ๊ทธ๋ ‡์Šต๋‹ˆ๋‹ค.
**๋ชจ๋ธ ๊ทœ๋ชจ:** ์ปดํ“จํŒ… ์ œํ•œ์œผ๋กœ ์ธํ•ด ํ‰๊ฐ€ํ•œ ๊ฐ€์žฅ ํฐ ๋ชจ๋ธ์€ 1000์–ต ํ† ํฐ์œผ๋กœ ํ•™์Šต๋œ 67์–ต ํŒŒ๋ผ๋ฏธํ„ฐ์˜€์Šต๋‹ˆ๋‹ค. ํ˜„์žฌ๊นŒ์ง€ ์ž„๋ฒ ๋”ฉ ๊ธฐ๋ฐ˜ ๋ฐ์ดํ„ฐ ํ๋ ˆ์ด์…˜ ์ ‘๊ทผ ๋ฐฉ์‹์— ๊ฐ€์žฅ ํฐ ๊ทœ๋ชจ์˜ ๋ชจ๋ธ์ด์ง€๋งŒ, ํŠนํžˆ ํšจ์œจ์„ฑ ํ–ฅ์ƒ์ด ๋ชจ๋ธ ๊ทœ๋ชจ์— ๋”ฐ๋ผ ์ปค์ง€๋Š” ๊ด€์ฐฐ์— ๋น„์ถ”์–ด 1000์–ต์„ ์ดˆ๊ณผํ•˜๋Š” ๋ชจ๋ธ ๊ทœ๋ชจ์—์„œ์˜ ์ถ”๊ฐ€ ์กฐ์‚ฌ๋Š” ๋งค์šฐ ํฅ๋ฏธ๋กœ์šธ ๊ฒƒ์ž…๋‹ˆ๋‹ค.
The Self-Consistency approach is grounded in the principle that if an LLM generates multiple, similar responses to the same prompt, it is more likely that the response is accurate. This method involves asking the LLM to tackle a query multiple times, each time analyzing the response for consistency. This technique is especially useful in scenarios where factual accuracy and precision are paramount.
Self-Consistency ์ ‘๊ทผ๋ฒ•์€ LLM์ด ๋™์ผํ•œ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด ์—ฌ๋Ÿฌ ๋ฒˆ ์œ ์‚ฌํ•œ ์‘๋‹ต์„ ์ƒ์„ฑํ•  ๊ฒฝ์šฐ, ๊ทธ ์‘๋‹ต์ด ์ •ํ™•ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค๋Š” ์›์น™์— ๊ธฐ๋ฐ˜ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ LLM์—๊ฒŒ ์—ฌ๋Ÿฌ ๋ฒˆ ์งˆ๋ฌธ์„ ๋˜์ง€๊ณ , ๋งค๋ฒˆ ์‘๋‹ต์˜ ์ผ๊ด€์„ฑ์„ ๋ถ„์„ํ•˜๋Š” ๊ณผ์ •์„ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. ์ด ๊ธฐ์ˆ ์€ ์‚ฌ์‹ค์  ์ •ํ™•์„ฑ๊ณผ ์ •๋ฐ€์„ฑ์ด ์ค‘์š”ํ•œ ์ƒํ™ฉ์—์„œ ํŠนํžˆ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.
์ž๊ธฐ ์ผ๊ด€์„ฑ ์ ‘๊ทผ ๋ฐฉ์‹์€ LLM์ด ๋™์ผํ•œ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด ์—ฌ๋Ÿฌ ๊ฐœ์˜ ์œ ์‚ฌํ•œ ์‘๋‹ต์„ ์ƒ์„ฑํ•˜๋Š” ๊ฒฝ์šฐ ์‘๋‹ต์ด ์ •ํ™•ํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค๋Š” ์›์น™์— ๊ธฐ๋ฐ˜ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ LLM์—๊ฒŒ ์ฟผ๋ฆฌ์— ์—ฌ๋Ÿฌ ๋ฒˆ ๋Œ€์ฒ˜ํ•˜๋„๋ก ์š”์ฒญํ•˜๊ณ , ๋งค๋ฒˆ ์‘๋‹ต์˜ ์ผ๊ด€์„ฑ์„ ๋ถ„์„ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด ๊ธฐ๋ฒ•์€ ์‚ฌ์‹ค ์ •ํ™•์„ฑ๊ณผ ์ •๋ฐ€๋„๊ฐ€ ๊ฐ€์žฅ ์ค‘์š”ํ•œ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ํŠนํžˆ ์œ ์šฉํ•ฉ๋‹ˆ๋‹ค.
In addition to the above basic evaluation tasks, LLMs also exhibit some superior abilities that require special consider tions for evaluation. In this part, we discuss several representative advanced abilities and the corresponding evaluation approaches, including human alignment, interaction with the external environment, and tool manipulation. Next, we discuss these advanced abilities in detail.
์œ„์˜ ๊ธฐ๋ณธ ํ‰๊ฐ€ ์ž‘์—… ์™ธ์—๋„, LLM์€ ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ํŠน๋ณ„ํ•œ ๊ณ ๋ ค๊ฐ€ ํ•„์š”ํ•œ ๋ช‡ ๊ฐ€์ง€ ์šฐ์ˆ˜ํ•œ ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด ๋ถ€๋ถ„์—์„œ๋Š” ์ธ๊ฐ„ ์ •๋ ฌ, ์™ธ๋ถ€ ํ™˜๊ฒฝ๊ณผ์˜ ์ƒํ˜ธ์ž‘์šฉ, ๋„๊ตฌ ์กฐ์ž‘์„ ํฌํ•จํ•œ ์—ฌ๋Ÿฌ ๋Œ€ํ‘œ์ ์ธ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ๊ณผ ํ•ด๋‹น ํ‰๊ฐ€ ์ ‘๊ทผ๋ฒ•์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, ์ด๋Ÿฌํ•œ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ์— ๋Œ€ํ•ด ์ž์„ธํžˆ ๋…ผ์˜ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.
์œ„์˜ ๊ธฐ๋ณธ ํ‰๊ฐ€ ์ž‘์—… ์™ธ์—๋„ LLM์€ ํ‰๊ฐ€๋ฅผ ์œ„ํ•ด ํŠน๋ณ„ํ•œ ๊ณ ๋ ค๊ฐ€ ํ•„์š”ํ•œ ๋ช‡ ๊ฐ€์ง€ ์šฐ์ˆ˜ํ•œ ๋Šฅ๋ ฅ๋„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด ๋ถ€๋ถ„์—์„œ๋Š” ์ธ๊ฐ„์˜ ์ •๋ ฌ, ์™ธ๋ถ€ ํ™˜๊ฒฝ๊ณผ์˜ ์ƒํ˜ธ ์ž‘์šฉ, ๋„๊ตฌ ์กฐ์ž‘ ๋“ฑ ๋ช‡ ๊ฐ€์ง€ ๋Œ€ํ‘œ์ ์ธ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ๊ณผ ๊ทธ์— ์ƒ์‘ํ•˜๋Š” ํ‰๊ฐ€ ์ ‘๊ทผ ๋ฐฉ์‹์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์—์„œ๋Š” ์ด๋Ÿฌํ•œ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ์— ๋Œ€ํ•ด ์ž์„ธํžˆ ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค.
FindingsResults in Fig. 2 reveal the quantitative predictability of domain losses given the domain proportions. We encouragingly find that, for checkpoints with the same size and trained with the same number of steps, after subtracting a shared constant3, their domain losses in the log scale demonstrate a linear relationship to the domain proportion. This holds for both domains in our experiments.
Fig. 2์˜ ๊ฒฐ๊ณผ๋Š” ๋„๋ฉ”์ธ ๋น„์œจ์ด ์ฃผ์–ด์กŒ์„ ๋•Œ ๋„๋ฉ”์ธ ์†์‹ค์˜ ์ •๋Ÿ‰์  ์˜ˆ์ธก ๊ฐ€๋Šฅ์„ฑ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ณ ๋ฌด์ ์ธ ์ ์€, ๋™์ผํ•œ ํฌ๊ธฐ์˜ ์ฒดํฌํฌ์ธํŠธ์™€ ๋™์ผํ•œ ๋‹จ๊ณ„ ์ˆ˜๋กœ ํ›ˆ๋ จ๋œ ๊ฒฝ์šฐ, ๊ณตํ†ต ์ƒ์ˆ˜3์„ ๋บ€ ํ›„ ๋กœ๊ทธ ์Šค์ผ€์ผ์—์„œ์˜ ๋„๋ฉ”์ธ ์†์‹ค์ด ๋„๋ฉ”์ธ ๋น„์œจ๊ณผ ์„ ํ˜• ๊ด€๊ณ„๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์šฐ๋ฆฌ์˜ ์‹คํ—˜์—์„œ ๋‘ ๋„๋ฉ”์ธ ๋ชจ๋‘์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค.
๋ฐœ๊ฒฌ๊ฒฐ๊ณผ ๊ทธ๋ฆผ 2์˜ ๊ฒฐ๊ณผ๋Š” ๋„๋ฉ”์ธ ๋น„์œจ์ด ์ฃผ์–ด์กŒ์„ ๋•Œ ๋„๋ฉ”์ธ ์†์‹ค์˜ ์ •๋Ÿ‰์  ์˜ˆ์ธก ๊ฐ€๋Šฅ์„ฑ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋™์ผํ•œ ํฌ๊ธฐ๋กœ ๋™์ผํ•œ ์ˆ˜์˜ ๋‹จ๊ณ„๋กœ ํ›ˆ๋ จ๋œ ์ฒดํฌํฌ์ธํŠธ์˜ ๊ฒฝ์šฐ, ๊ณต์œ  ์ƒ์ˆ˜3์„ ๋บ€ ํ›„ ๋กœ๊ทธ ์Šค์ผ€์ผ์˜ ๋„๋ฉ”์ธ ์†์‹ค์ด ๋„๋ฉ”์ธ ๋น„์œจ์— ๋”ฐ๋ผ ์„ ํ˜•์ ์ธ ๊ด€๊ณ„๋ฅผ ๋ณด์ด๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์‹คํ—˜์— ์‚ฌ์šฉ๋œ ๋‘ ๋„๋ฉ”์ธ ๋ชจ๋‘์— ํ•ด๋‹น๋ฉ๋‹ˆ๋‹ค.
Generative evaluation metrics are also another type of evaluation metric for LLMs that use another LLM for evaluating the answer. However, depending on the task itself, evaluation can be possible in this way or not. Another dependency that makes generative evaluation error-prone is reliance on the prompt itself. RAGAS is one of the good examples that incorporate the usage of generative evaluation.
์ƒ์„ฑ ํ‰๊ฐ€ ์ง€ํ‘œ๋Š” ๋‹ต๋ณ€์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ๋˜ ๋‹ค๋ฅธ LLM์„ ์‚ฌ์šฉํ•˜๋Š” LLM ํ‰๊ฐ€ ์ง€ํ‘œ์˜ ๋˜ ๋‹ค๋ฅธ ์œ ํ˜•์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ž‘์—… ์ž์ฒด์— ๋”ฐ๋ผ ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์œผ๋กœ ํ‰๊ฐ€๊ฐ€ ๊ฐ€๋Šฅํ•  ์ˆ˜๋„ ์žˆ๊ณ  ๊ทธ๋ ‡์ง€ ์•Š์„ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ƒ์„ฑ ํ‰๊ฐ€๋ฅผ ์˜ค๋ฅ˜๊ฐ€ ๋ฐœ์ƒํ•˜๊ธฐ ์‰ฌ์šด ๋ฐฉ์‹์œผ๋กœ ๋งŒ๋“œ๋Š” ๋˜ ๋‹ค๋ฅธ ์˜์กด์„ฑ์€ ํ”„๋กฌํ”„ํŠธ ์ž์ฒด์— ๋Œ€ํ•œ ์˜์กด์„ฑ์ž…๋‹ˆ๋‹ค. RAGAS๋Š” ์ƒ์„ฑ ํ‰๊ฐ€์˜ ์‚ฌ์šฉ์„ ํ†ตํ•ฉํ•œ ์ข‹์€ ์˜ˆ ์ค‘ ํ•˜๋‚˜์ž…๋‹ˆ๋‹ค.
์ƒ์„ฑ ํ‰๊ฐ€ ๋ฉ”ํŠธ๋ฆญ์€ ๋‹ค๋ฅธ LLM์„ ์‚ฌ์šฉํ•˜์—ฌ ๋‹ต๋ณ€์„ ํ‰๊ฐ€ํ•˜๋Š” ๋˜ ๋‹ค๋ฅธ ์œ ํ˜•์˜ LLM ํ‰๊ฐ€ ๋ฉ”ํŠธ๋ฆญ์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ž‘์—… ์ž์ฒด์— ๋”ฐ๋ผ ์ด๋Ÿฌํ•œ ๋ฐฉ์‹์œผ๋กœ ํ‰๊ฐ€๊ฐ€ ๊ฐ€๋Šฅํ•˜๊ฑฐ๋‚˜ ๋ถˆ๊ฐ€๋Šฅํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ƒ์„ฑ์  ํ‰๊ฐ€ ์˜ค๋ฅ˜๋ฅผ ์œ ๋ฐœํ•˜๋Š” ๋˜ ๋‹ค๋ฅธ ์ข…์†์„ฑ์€ ํ”„๋กฌํ”„ํŠธ ์ž์ฒด์— ๋Œ€ํ•œ ์˜์กด๋„์ž…๋‹ˆ๋‹ค. RAGAS๋Š” ์ƒ์„ฑ์  ํ‰๊ฐ€์˜ ์‚ฌ์šฉ์„ ํ†ตํ•ฉํ•œ ์ข‹์€ ์˜ˆ ์ค‘ ํ•˜๋‚˜์ž…๋‹ˆ๋‹ค.
The highly distributed NearDup implementation we employed is one used for large-scale production tasks at Google. On the English C4 dataset, the algorithm consumed approximately 41.5 kWh of energy. Note that our choices of \(k\) and \(b\) were designed to produce very high recall, and with different parameters, the algorithm could be made much more energy efficient while producing similar results.
์šฐ๋ฆฌ๊ฐ€ ์‚ฌ์šฉํ•œ ๊ณ ๋„๋กœ ๋ถ„์‚ฐ๋œ NearDup ๊ตฌํ˜„์€ Google์—์„œ ๋Œ€๊ทœ๋ชจ ์ƒ์‚ฐ ์ž‘์—…์— ์‚ฌ์šฉ๋˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์˜์–ด C4 ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์•ฝ 41.5 kWh์˜ ์—๋„ˆ์ง€๋ฅผ ์†Œ๋น„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ \(k\)์™€ \(b\) ์„ ํƒ์€ ๋งค์šฐ ๋†’์€ ์žฌํ˜„์œจ์„ ์ƒ์„ฑํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์œผ๋ฉฐ, ๋‹ค๋ฅธ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ์œ ์‚ฌํ•œ ๊ฒฐ๊ณผ๋ฅผ ์ƒ์„ฑํ•˜๋ฉด์„œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ํ›จ์”ฌ ๋” ์—๋„ˆ์ง€ ํšจ์œจ์ ์œผ๋กœ ๋งŒ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์šฐ๋ฆฌ๊ฐ€ ์‚ฌ์šฉํ•œ ๊ณ ๋ถ„์‚ฐํ˜• NearDup ๊ตฌํ˜„์€ Google์—์„œ ๋Œ€๊ทœ๋ชจ ํ”„๋กœ๋•์…˜ ์ž‘์—…์— ์‚ฌ์šฉ๋˜๋Š” ๊ตฌํ˜„์ž…๋‹ˆ๋‹ค. ์˜์–ด C4 ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ์•ฝ 41.5kWh์˜ ์—๋„ˆ์ง€๋ฅผ ์†Œ๋น„ํ–ˆ์Šต๋‹ˆ๋‹ค. k\)์™€ \(b\)์˜ ์„ ํƒ์€ ๋งค์šฐ ๋†’์€ ๋ฆฌ์ฝœ์„ ์ƒ์„ฑํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์œผ๋ฉฐ, ๋‹ค๋ฅธ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ๋น„์Šทํ•œ ๊ฒฐ๊ณผ๋ฅผ ์ƒ์„ฑํ•˜๋ฉด์„œ ํ›จ์”ฌ ๋” ์—๋„ˆ์ง€ ํšจ์œจ์ ์œผ๋กœ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๋งŒ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
By pretraining on massive text data ranging from billions to trillions of tokens, LLMs are capable of capturing intricate language structures, semantics, and contextual relationships, thereby acquiring strong language generation capabilities. Additionally, these LLMs also learn how to comprehend concepts, facts, and the connections between them, leading to a broad understanding of world knowledge.
์ˆ˜์‹ญ์–ต์—์„œ ์ˆ˜์กฐ ๊ฐœ์˜ ํ† ํฐ์— ์ด๋ฅด๋Š” ๋ฐฉ๋Œ€ํ•œ ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์ „ ํ•™์Šตํ•จ์œผ๋กœ์จ, LLM์€ ๋ณต์žกํ•œ ์–ธ์–ด ๊ตฌ์กฐ, ์˜๋ฏธ๋ก , ๊ทธ๋ฆฌ๊ณ  ๋ฌธ๋งฅ์  ๊ด€๊ณ„๋ฅผ ํฌ์ฐฉํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์–ด ๊ฐ•๋ ฅํ•œ ์–ธ์–ด ์ƒ์„ฑ ๋Šฅ๋ ฅ์„ ๊ฐ–์ถ”๊ฒŒ ๋ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด๋Ÿฌํ•œ LLM์€ ๊ฐœ๋…, ์‚ฌ์‹ค, ๊ทธ๋ฆฌ๊ณ  ๊ทธ๋“ค ๊ฐ„์˜ ์—ฐ๊ฒฐ์„ ์ดํ•ดํ•˜๋Š” ๋ฐฉ๋ฒ•๋„ ํ•™์Šตํ•˜์—ฌ ๊ด‘๋ฒ”์œ„ํ•œ ์„ธ๊ณ„ ์ง€์‹์„ ์ดํ•ดํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.
์ˆ˜์‹ญ์–ต์—์„œ ์ˆ˜์กฐ์— ์ด๋ฅด๋Š” ๋ฐฉ๋Œ€ํ•œ ํ…์ŠคํŠธ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์ „ ํ•™์Šตํ•จ์œผ๋กœ์จ LLM์€ ๋ณต์žกํ•œ ์–ธ์–ด ๊ตฌ์กฐ, ์˜๋ฏธ๋ก , ๋ฌธ๋งฅ ๊ด€๊ณ„๋ฅผ ํฌ์ฐฉํ•˜์—ฌ ๊ฐ•๋ ฅํ•œ ์–ธ์–ด ์ƒ์„ฑ ๋Šฅ๋ ฅ์„ ์Šต๋“ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด๋Ÿฌํ•œ LLM์€ ๊ฐœ๋…, ์‚ฌ์‹ค ๋ฐ ์ด๋“ค ๊ฐ„์˜ ์—ฐ๊ด€์„ฑ์„ ์ดํ•ดํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ํ•™์Šตํ•˜์—ฌ ์„ธ๊ณ„ ์ง€์‹์— ๋Œ€ํ•œ ๊ด‘๋ฒ”์œ„ํ•œ ์ดํ•ด๋ฅผ ์ด๋Œ์–ด๋ƒ…๋‹ˆ๋‹ค.
1. Islamist terrorism, such as the Islamic State in Iraq and the Levant (ISIL) and Al Qaeda. 2. Ultra Right Wing Terrorism, Neo-Nazi and White Supremacist Terrorism. 3. Ultra Left Wing and Anarchist Terrorism. 4. Environmental terrorism. 5. Single-issue terrorism, such as animal rights and abortion issues. 6. Northern-Ireland related terrorism, such as the Provisional Irish Republican Army (IRA).
1. ์ด์Šฌ๋žŒ์ฃผ์˜ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜, ์˜ˆ๋ฅผ ๋“ค์–ด ์ด๋ผํฌ์™€ ๋ ˆ๋ฐ˜ํŠธ์˜ ์ด์Šฌ๋žŒ ๊ตญ๊ฐ€(ISIL)์™€ ์•Œ์นด์—๋‹ค. 2. ๊ทน์šฐ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜, ๋„ค์˜ค๋‚˜์น˜ ๋ฐ ๋ฐฑ์ธ ์šฐ์›”์ฃผ์˜ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜. 3. ๊ทน์ขŒ ๋ฐ ๋ฌด์ •๋ถ€์ฃผ์˜ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜. 4. ํ™˜๊ฒฝ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜. 5. ๋™๋ฌผ ๊ถŒ๋ฆฌ ๋ฐ ๋‚™ํƒœ ๋ฌธ์ œ์™€ ๊ฐ™์€ ๋‹จ์ผ ์ด์Šˆ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜. 6. ๋ถ์•„์ผ๋žœ๋“œ ๊ด€๋ จ ํ…Œ๋Ÿฌ๋ฆฌ์ฆ˜, ์˜ˆ๋ฅผ ๋“ค์–ด ์ž„์‹œ ์•„์ผ๋žœ๋“œ ๊ณตํ™”๊ตญ๊ตฐ(IRA).
1. ์ด๋ผํฌ ๋ ˆ๋ฐ˜ํŠธ ์ด์Šฌ๋žŒ๊ตญ๊ฐ€(ISIL) ๋ฐ ์•Œ์นด์—๋‹ค์™€ ๊ฐ™์€ ์ด์Šฌ๋žŒ์ฃผ์˜ ํ…Œ๋Ÿฌ. 2. ๊ทน์šฐํŒŒ ํ…Œ๋Ÿฌ, ์‹ ๋‚˜์น˜ ๋ฐ ๋ฐฑ์ธ ์šฐ์›”์ฃผ์˜ ํ…Œ๋Ÿฌ. 3. ๊ทน์ขŒํŒŒ ๋ฐ ์•„๋‚˜ํ‚ค์ŠคํŠธ ํ…Œ๋Ÿฌ. 4. ํ™˜๊ฒฝ ํ…Œ๋Ÿฌ. 5. ๋™๋ฌผ๊ถŒ ๋ฐ ๋‚™ํƒœ ๋ฌธ์ œ์™€ ๊ฐ™์€ ๋‹จ์ผ ์ด์Šˆ ํ…Œ๋Ÿฌ. 6. ์ž„์‹œ ์•„์ผ๋žœ๋“œ ๊ณตํ™”๋‹น(IRA)๊ณผ ๊ฐ™์€ ๋ถ์•„์ผ๋žœ๋“œ ๊ด€๋ จ ํ…Œ๋Ÿฌ.
In our knowledge analysis, we assess our model's performance across four benchmarks: **ARC**, **Hellaswag**, **MLU**, and **C-eval-tw**. We employ EleutherAI/lm-evaluation-harness3**Gao et al. (2021) to assess the model performance on these benchmarks. These benchmarks consist of multiple-choice questions. The accuracy computation is based on selecting the option with the highest probabilities.
์šฐ๋ฆฌ์˜ ์ง€์‹ ๋ถ„์„์—์„œ, ์šฐ๋ฆฌ๋Š” **ARC**, **Hellaswag**, **MLU**, **C-eval-tw** ๋„ค ๊ฐ€์ง€ ๋ฒค์น˜๋งˆํฌ์— ๊ฑธ์ณ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ์—์„œ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด EleutherAI/lm-evaluation-harness3**Gao et al. (2021)์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฒค์น˜๋งˆํฌ๋Š” ์—ฌ๋Ÿฌ ์„ ํƒ์ง€ ์งˆ๋ฌธ์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ์ •ํ™•๋„ ๊ณ„์‚ฐ์€ ๊ฐ€์žฅ ๋†’์€ ํ™•๋ฅ ์„ ๊ฐ€์ง„ ์„ ํƒ์ง€๋ฅผ ์„ ํƒํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ ์ด๋ฃจ์–ด์ง‘๋‹ˆ๋‹ค.
์ง€์‹ ๋ถ„์„์—์„œ๋Š” ๋„ค ๊ฐ€์ง€ ๋ฒค์น˜๋งˆํฌ์—์„œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค: **ARC**, **Hellaswag**, **MLU**, **C-eval-tw**. ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ์— ๋Œ€ํ•œ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด EleutherAI/lm-evaluation-harness3**Gao et al. (2021)์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ๋Š” ๊ฐ๊ด€์‹ ๋ฌธ์ œ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค. ์ •ํ™•๋„ ๊ณ„์‚ฐ์€ ๊ฐ€์žฅ ๋†’์€ ํ™•๋ฅ ์„ ๊ฐ€์ง„ ์˜ต์…˜์„ ์„ ํƒํ•˜๋Š” ๊ฒƒ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
Unstructured data, in contrast, does not have a clearly defined data structure and exists in various forms, including text, images, and audio. Due to its large and diverse nature, it is challenging to store and manage in traditional tabular form. Although it contains valuable information, it requires natural language processing, image recognition, and other technologies to parse and comprehend.
๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ๋Š” ๋ช…ํ™•ํ•˜๊ฒŒ ์ •์˜๋œ ๋ฐ์ดํ„ฐ ๊ตฌ์กฐ๊ฐ€ ์—†์œผ๋ฉฐ ํ…์ŠคํŠธ, ์ด๋ฏธ์ง€, ์˜ค๋””์˜ค ๋“ฑ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ๋กœ ์กด์žฌํ•ฉ๋‹ˆ๋‹ค. ๊ทธ ํฌ๊ธฐ์™€ ๋‹ค์–‘์„ฑ ๋•Œ๋ฌธ์— ์ „ํ†ต์ ์ธ ํ‘œ ํ˜•์‹์œผ๋กœ ์ €์žฅํ•˜๊ณ  ๊ด€๋ฆฌํ•˜๊ธฐ๊ฐ€ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ๋น„๋ก ์œ ์šฉํ•œ ์ •๋ณด๋ฅผ ํฌํ•จํ•˜๊ณ  ์žˆ์ง€๋งŒ, ์ด๋ฅผ ํ•ด์„ํ•˜๊ณ  ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ, ์ด๋ฏธ์ง€ ์ธ์‹ ๋“ฑ์˜ ๊ธฐ์ˆ ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๋ฐ˜๋ฉด์— ๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ๋Š” ๋ฐ์ดํ„ฐ ๊ตฌ์กฐ๊ฐ€ ๋ช…ํ™•ํ•˜๊ฒŒ ์ •์˜๋˜์–ด ์žˆ์ง€ ์•Š์œผ๋ฉฐ ํ…์ŠคํŠธ, ์ด๋ฏธ์ง€, ์˜ค๋””์˜ค ๋“ฑ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ๋กœ ์กด์žฌํ•ฉ๋‹ˆ๋‹ค. ๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ๋Š” ๊ทธ ์–‘์ด ๋งŽ๊ณ  ๋‹ค์–‘ํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๊ธฐ์กด์˜ ํ‘œ ํ˜•์‹์œผ๋กœ๋Š” ์ €์žฅ ๋ฐ ๊ด€๋ฆฌ๊ฐ€ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ์—๋Š” ๊ท€์ค‘ํ•œ ์ •๋ณด๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์ง€๋งŒ, ์ด๋ฅผ ํŒŒ์‹ฑํ•˜๊ณ  ์ดํ•ดํ•˜๋ ค๋ฉด ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ, ์ด๋ฏธ์ง€ ์ธ์‹ ๋ฐ ๊ธฐํƒ€ ๊ธฐ์ˆ ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
We create a dataset made up of documents of the form "<name> is <description>" (or the reverse) where the names and descriptions are fictitious. Each description is intended to denote a unique individual. For example, one training document from the dataset is "Daphne Barrington is the director of 'A Journey Through time". We use GPT-4 (OpenAI, 2023b) to generate pairs of names and descriptions.
์šฐ๋ฆฌ๋Š” ์ด๋ฆ„๊ณผ ์„ค๋ช…์ด ํ—ˆ๊ตฌ์ธ "<name> is <description>" (๋˜๋Š” ๊ทธ ๋ฐ˜๋Œ€) ํ˜•์‹์˜ ๋ฌธ์„œ๋กœ ๊ตฌ์„ฑ๋œ ๋ฐ์ดํ„ฐ์…‹์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๊ฐ ์„ค๋ช…์€ ๊ณ ์œ ํ•œ ๊ฐœ์ธ์„ ๋‚˜ํƒ€๋‚ด๋„๋ก ์˜๋„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๋ฐ์ดํ„ฐ์…‹์˜ ํ•œ ํ•™์Šต ๋ฌธ์„œ๋Š” "Daphne Barrington์€ 'A Journey Through Time'์˜ ๊ฐ๋…์ž…๋‹ˆ๋‹ค."์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด๋ฆ„๊ณผ ์„ค๋ช… ์Œ์„ ์ƒ์„ฑํ•˜๊ธฐ ์œ„ํ•ด GPT-4 (OpenAI, 2023b)๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค.
์ด๋ฆ„๊ณผ ์„ค๋ช…์ด ๊ฐ€์ƒ์ธ "<์ด๋ฆ„>์€ <์„ค๋ช…>"(๋˜๋Š” ๊ทธ ๋ฐ˜๋Œ€) ํ˜•์‹์˜ ๋ฌธ์„œ๋กœ ๊ตฌ์„ฑ๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๊ฐ ์„ค๋ช…์€ ๊ณ ์œ ํ•œ ๊ฐœ์ธ์„ ๋‚˜ํƒ€๋‚ด๋„๋ก ์˜๋„๋ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ํ•œ ํ•™์Šต ๋ฌธ์„œ๋Š” "๋‹คํ”„๋„ค ๋ฐฐ๋งํ„ด์€ '์‹œ๊ฐ„์„ ์—ฌํ–‰ํ•˜๋Š” ์—ฌ์ •'์˜ ๊ฐ๋…์ž…๋‹ˆ๋‹ค."์ž…๋‹ˆ๋‹ค. GPT-4(OpenAI, 2023b)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ด๋ฆ„๊ณผ ์„ค๋ช… ์Œ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
biggest clusters in Appendix H.1. We perform MinHash deduplication using 9,000 hashes per document, calculated over 5-grams and divided into 20 buckets of 450 hashes. We found that using less aggressive settings, such as the 10 hashes of The Pile (Gao et al., 2020), resulted in lower deduplication rates and worsened model performance. See Appendix G.3.1 for more details about our MinHash setup.
๋ถ€๋ก H.1์˜ ๊ฐ€์žฅ ํฐ ํด๋Ÿฌ์Šคํ„ฐ๋“ค. ์šฐ๋ฆฌ๋Š” ๋ฌธ์„œ๋‹น 9,000๊ฐœ์˜ ํ•ด์‹œ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ MinHash ์ค‘๋ณต ์ œ๊ฑฐ๋ฅผ ์ˆ˜ํ–‰ํ•˜๋ฉฐ, ์ด๋Š” 5-๊ทธ๋žจ์„ ๊ธฐ์ค€์œผ๋กœ ๊ณ„์‚ฐ๋˜๊ณ  450๊ฐœ์˜ ํ•ด์‹œ๋กœ ์ด๋ฃจ์–ด์ง„ 20๊ฐœ์˜ ๋ฒ„ํ‚ท์œผ๋กœ ๋‚˜๋‰ฉ๋‹ˆ๋‹ค. The Pile (Gao et al., 2020)์˜ 10๊ฐœ์˜ ํ•ด์‹œ์™€ ๊ฐ™์€ ๋œ ๊ณต๊ฒฉ์ ์ธ ์„ค์ •์„ ์‚ฌ์šฉํ•  ๊ฒฝ์šฐ ์ค‘๋ณต ์ œ๊ฑฐ์œจ์ด ๋‚ฎ์•„์ง€๊ณ  ๋ชจ๋ธ ์„ฑ๋Šฅ์ด ์ €ํ•˜๋˜๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ MinHash ์„ค์ •์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก G.3.1์„ ์ฐธ์กฐํ•˜์‹ญ์‹œ์˜ค.
๋ถ€๋ก H.1์˜ ๊ฐ€์žฅ ํฐ ํด๋Ÿฌ์Šคํ„ฐ์ž…๋‹ˆ๋‹ค. 5๊ทธ๋žจ์— ๋Œ€ํ•ด ๊ณ„์‚ฐํ•˜๊ณ  450๊ฐœ์˜ ํ•ด์‹œ๋กœ ๊ตฌ์„ฑ๋œ 20๊ฐœ์˜ ๋ฒ„ํ‚ท์œผ๋กœ ๋‚˜๋ˆ„์–ด ๋ฌธ์„œ๋‹น 9,000๊ฐœ์˜ ํ•ด์‹œ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ MinHash ์ค‘๋ณต ์ œ๊ฑฐ๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๋” ์ ๊ทน์ ์ธ ์„ค์ •, ์˜ˆ๋ฅผ ๋“ค์–ด ๋” ํŒŒ์ผ(Gao et al., 2020)์˜ 10๊ฐœ ํ•ด์‹œ๋ฅผ ์‚ฌ์šฉํ•˜๋ฉด ์ค‘๋ณต ์ œ๊ฑฐ์œจ์ด ๋‚ฎ์•„์ง€๊ณ  ๋ชจ๋ธ ์„ฑ๋Šฅ์ด ์•…ํ™”๋˜๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก G.3.1์„ ์ฐธ์กฐํ•˜์„ธ์š”.
It is worth noting that although the questions are based on information the models were not exposed to during training, the results of the base models surpass \(\frac{1}{L}=0.25\). This can partially be explained by the models using reasoning and/or pre-existing knowledge when answering questions that are not independent of the past information. Some examples of this can be found in Appendix C.
์ฃผ๋ชฉํ•  ์ ์€ ์งˆ๋ฌธ๋“ค์ด ๋ชจ๋ธ์ด ํ›ˆ๋ จ ์ค‘์— ๋…ธ์ถœ๋˜์ง€ ์•Š์€ ์ •๋ณด์— ๊ธฐ๋ฐ˜ํ•˜๊ณ  ์žˆ์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ , ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ๊ฒฐ๊ณผ๊ฐ€ \(\frac{1}{L}=0.25\)๋ฅผ ์ดˆ๊ณผํ•œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์ด ๊ณผ๊ฑฐ ์ •๋ณด์™€ ๋…๋ฆฝ์ ์ด์ง€ ์•Š์€ ์งˆ๋ฌธ์— ๋‹ตํ•  ๋•Œ ์ถ”๋ก  ๋ฐ/๋˜๋Š” ๊ธฐ์กด ์ง€์‹์„ ์‚ฌ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์— ๋Œ€ํ•œ ๋ช‡ ๊ฐ€์ง€ ์˜ˆ๋Š” ๋ถ€๋ก C์—์„œ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์งˆ๋ฌธ์€ ๋ชจ๋ธ์ด ํ•™์Šตํ•˜๋Š” ๋™์•ˆ ๋…ธ์ถœ๋˜์ง€ ์•Š์€ ์ •๋ณด๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜์ง€๋งŒ, ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ๊ฒฐ๊ณผ๊ฐ€ \(\frac{1}{L}=0.25\)๋ฅผ ์ดˆ๊ณผํ•œ๋‹ค๋Š” ์ ์— ์ฃผ๋ชฉํ•  ํ•„์š”๊ฐ€ ์žˆ๋‹ค. ์ด๋Š” ๋ถ€๋ถ„์ ์œผ๋กœ ๋ชจ๋ธ์ด ๊ณผ๊ฑฐ ์ •๋ณด์™€ ๋…๋ฆฝ์ ์ด์ง€ ์•Š์€ ์งˆ๋ฌธ์— ๋‹ตํ•  ๋•Œ ์ถ”๋ก  ๋ฐ/๋˜๋Š” ๊ธฐ์กด ์ง€์‹์„ ์‚ฌ์šฉํ•˜๊ธฐ ๋•Œ๋ฌธ์ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์— ๋Œ€ํ•œ ๋ช‡ ๊ฐ€์ง€ ์˜ˆ๋Š” ๋ถ€๋ก C์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Social science is one of the branches of science, devoted to the study of societies and the relationships among individuals within those societies. LLM-based autonomous agents can promote this domain by leveraging their impressive human-like understanding, thinking and task solving capabilities. In the following, we discuss several key areas that can be affected by LLM-based autonomous agents.
์‚ฌํšŒ ๊ณผํ•™์€ ์‚ฌํšŒ์™€ ๊ทธ ์‚ฌํšŒ ๋‚ด ๊ฐœ์ธ๋“ค ๊ฐ„์˜ ๊ด€๊ณ„๋ฅผ ์—ฐ๊ตฌํ•˜๋Š” ๊ณผํ•™์˜ ํ•œ ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. LLM ๊ธฐ๋ฐ˜์˜ ์ž์œจ ์—์ด์ „ํŠธ๋Š” ์ธ๊ฐ„๊ณผ ์œ ์‚ฌํ•œ ์ดํ•ด, ์‚ฌ๊ณ  ๋ฐ ๋ฌธ์ œ ํ•ด๊ฒฐ ๋Šฅ๋ ฅ์„ ํ™œ์šฉํ•˜์—ฌ ์ด ๋ถ„์•ผ๋ฅผ ๋ฐœ์ „์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์—์„œ๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ๊ฐ€ ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋Š” ๋ช‡ ๊ฐ€์ง€ ์ฃผ์š” ์˜์—ญ์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค.
์‚ฌํšŒ๊ณผํ•™์€ ์‚ฌํšŒ์™€ ๊ทธ ์•ˆ์— ์žˆ๋Š” ๊ฐœ์ธ ๊ฐ„์˜ ๊ด€๊ณ„๋ฅผ ์—ฐ๊ตฌํ•˜๋Š” ๊ณผํ•™์˜ ํ•œ ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ๋Š” ์ธ์ƒ์ ์ธ ์ธ๊ฐ„๊ณผ ์œ ์‚ฌํ•œ ์ดํ•ด๋ ฅ, ์‚ฌ๊ณ ๋ ฅ, ๊ณผ์ œ ํ•ด๊ฒฐ ๋Šฅ๋ ฅ์„ ํ™œ์šฉํ•˜์—ฌ ์ด ๋ถ„์•ผ๋ฅผ ๋ฐœ์ „์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์—์„œ๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ์˜ ์˜ํ–ฅ์„ ๋ฐ›์„ ์ˆ˜ ์žˆ๋Š” ๋ช‡ ๊ฐ€์ง€ ์ฃผ์š” ์˜์—ญ์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค.
For all the datasets used by stage-2 tuning, the context are provided as continuous paragraphs or documents that contain the answer. In contrast, the model needs to handle the top-\(k\) retrieved chunks at inference for long documents. To reduce such train/test mismatch, we investigate whether replacing some continuous paragraphs with retrieved top-k chunks will enhance the model's robustness.
2๋‹จ๊ณ„ ํŠœ๋‹์— ์‚ฌ์šฉ๋œ ๋ชจ๋“  ๋ฐ์ดํ„ฐ์…‹์— ๋Œ€ํ•ด, ๋ฌธ๋งฅ์€ ๋‹ต์„ ํฌํ•จํ•˜๋Š” ์—ฐ์†์ ์ธ ๋‹จ๋ฝ์ด๋‚˜ ๋ฌธ์„œ๋กœ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ๋ชจ๋ธ์€ ๊ธด ๋ฌธ์„œ์— ๋Œ€ํ•ด ์ถ”๋ก  ์‹œ ์ƒ์œ„ \(k\)๊ฐœ์˜ ๊ฒ€์ƒ‰๋œ ์ฒญํฌ๋ฅผ ์ฒ˜๋ฆฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ•™์Šต/ํ…Œ์ŠคํŠธ ๋ถˆ์ผ์น˜๋ฅผ ์ค„์ด๊ธฐ ์œ„ํ•ด, ์ผ๋ถ€ ์—ฐ์†์ ์ธ ๋‹จ๋ฝ์„ ๊ฒ€์ƒ‰๋œ ์ƒ์œ„ \(k\)๊ฐœ์˜ ์ฒญํฌ๋กœ ๋Œ€์ฒดํ•˜๋Š” ๊ฒƒ์ด ๋ชจ๋ธ์˜ ๊ฒฌ๊ณ ์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ๋Š”์ง€ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
2๋‹จ๊ณ„ ํŠœ๋‹์— ์‚ฌ์šฉ๋˜๋Š” ๋ชจ๋“  ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ๊ฒฝ์šฐ ๋‹ต๋ณ€์ด ํฌํ•จ๋œ ์—ฐ์†๋œ ๋‹จ๋ฝ ๋˜๋Š” ๋ฌธ์„œ๋กœ ์ปจํ…์ŠคํŠธ๊ฐ€ ์ œ๊ณต๋ฉ๋‹ˆ๋‹ค. ๋ฐ˜๋ฉด, ๋ชจ๋ธ์€ ๊ธด ๋ฌธ์„œ์˜ ๊ฒฝ์šฐ ์ถ”๋ก  ์‹œ ์ƒ์œ„ \(k\)๊ฐœ์˜ ๊ฒ€์ƒ‰๋œ ์ฒญํฌ๋ฅผ ์ฒ˜๋ฆฌํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ›ˆ๋ จ/ํ…Œ์ŠคํŠธ ๋ถˆ์ผ์น˜๋ฅผ ์ค„์ด๊ธฐ ์œ„ํ•ด ์ผ๋ถ€ ์—ฐ์†๋œ ๋‹จ๋ฝ์„ ์ƒ์œ„ k๊ฐœ์˜ ์ฒญํฌ๋กœ ๊ฒ€์ƒ‰ํ•˜์—ฌ ๋ชจ๋ธ์˜ ๊ฒฌ๊ณ ์„ฑ์„ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ๋Š”์ง€ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
_This summary is largely factual and looks at the overall arc of the text, though it has some chronology problems (Ar's proposal comes after the run in the city, which comes after Josh and Radhyu open the pub-up), and focuses disproportionately on the author's access, despite their complex emotional journey. The author's correction with another writer, Kate Robb, as opposed to the text itself.
์ด ์š”์•ฝ์€ ์ฃผ๋กœ ์‚ฌ์‹ค์— ๊ธฐ๋ฐ˜ํ•˜๊ณ  ํ…์ŠคํŠธ์˜ ์ „์ฒด์ ์ธ ํ๋ฆ„์„ ์‚ดํŽด๋ณด์ง€๋งŒ, ์ผ๋ถ€ ์—ฐ๋Œ€๊ธฐ์  ๋ฌธ์ œ(Ar์˜ ์ œ์•ˆ์ด ๋„์‹œ์—์„œ์˜ ๋‹ฌ๋ฆฌ๊ธฐ ํ›„์— ๋‚˜์˜ค๊ณ , ์ด๋Š” Josh์™€ Radhyu๊ฐ€ ํŒ์—…์„ ์—ฐ ํ›„์— ๋‚˜์˜ด)๊ฐ€ ์žˆ์œผ๋ฉฐ, ์ €์ž์˜ ๋ณต์žกํ•œ ๊ฐ์ •์  ์—ฌ์ •์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์ €์ž์˜ ์ ‘๊ทผ์— ๋ถˆ๊ท ํ˜•์ ์œผ๋กœ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ €์ž๋Š” ํ…์ŠคํŠธ ์ž์ฒด๋ณด๋‹ค๋Š” ๋‹ค๋ฅธ ์ž‘๊ฐ€์ธ Kate Robb์™€์˜ ๊ต์ •์„ ํ†ตํ•ด ์ˆ˜์ •ํ–ˆ์Šต๋‹ˆ๋‹ค.
์ด ์š”์•ฝ์€ ๋Œ€์ฒด๋กœ ์‚ฌ์‹ค์— ๊ทผ๊ฑฐํ•œ ๊ฒƒ์œผ๋กœ ํ…์ŠคํŠธ์˜ ์ „์ฒด์ ์ธ ํ๋ฆ„์„ ์‚ดํŽด๋ณด์•˜์ง€๋งŒ, ์—ฐ๋Œ€๊ธฐ์ ์œผ๋กœ ์•ฝ๊ฐ„์˜ ๋ฌธ์ œ๊ฐ€ ์žˆ๊ณ (Ar์˜ ์ œ์•ˆ์€ ์กฐ์‰ฌ์™€ ๋ผ๋””์šฐ๊ฐ€ ํŽ์„ ์—ฐ ๋’ค ๋„์‹œ์—์„œ ๋‹ฌ๋ฆฌ๋Š” ์žฅ๋ฉด ๋’ค์— ์˜ค๋Š” ์žฅ๋ฉด), ์ €์ž์˜ ๋ณต์žกํ•œ ๊ฐ์ • ์—ฌ์ •์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์ €์ž์˜ ์ ‘๊ทผ์— ๋ถˆ๊ท ํ˜•์ ์œผ๋กœ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํ…์ŠคํŠธ ์ž์ฒด๊ฐ€ ์•„๋‹Œ ๋‹ค๋ฅธ ์ž‘๊ฐ€์ธ ์ผ€์ดํŠธ ๋กญ๊ณผ์˜ ์ €์ž์˜ ์ˆ˜์ •์ž…๋‹ˆ๋‹ค.
Figure 1: Our pipeline for collecting faithfulness annotations in book-length summarization (Tables). First, (a) we generate summaries through hierarchical merging. Next, (b) we prompt GPT-4 to extract decontextualized claims. Finally, (c) we conduct a human evaluation of these claims, requiring annotators to validate each claim and provide their reasoning and evidence for the assigned label.
Figure 1: ์ฑ… ๊ธธ์ด์˜ ์š”์•ฝ์—์„œ ์‹ ๋ขฐ์„ฑ ์ฃผ์„์„ ์ˆ˜์ง‘ํ•˜๊ธฐ ์œ„ํ•œ ์šฐ๋ฆฌ์˜ ํŒŒ์ดํ”„๋ผ์ธ (ํ‘œ). ๋จผ์ €, (a) ๊ณ„์ธต์  ๋ณ‘ํ•ฉ์„ ํ†ตํ•ด ์š”์•ฝ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, (b) GPT-4๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฌธ๋งฅ์—์„œ ๋ฒ—์–ด๋‚œ ์ฃผ์žฅ๋“ค์„ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ, (c) ์ด๋Ÿฌํ•œ ์ฃผ์žฅ๋“ค์— ๋Œ€ํ•œ ์ธ๊ฐ„ ํ‰๊ฐ€๋ฅผ ์ˆ˜ํ–‰ํ•˜์—ฌ ์ฃผ์„์ž๋“ค์ด ๊ฐ ์ฃผ์žฅ์„ ๊ฒ€์ฆํ•˜๊ณ  ํ• ๋‹น๋œ ๋ผ๋ฒจ์— ๋Œ€ํ•œ ์ด์œ ์™€ ์ฆ๊ฑฐ๋ฅผ ์ œ๊ณตํ•˜๋„๋ก ์š”๊ตฌํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 1: ์ฑ… ๊ธธ์ด์˜ ์š”์•ฝ(ํ‘œ)์—์„œ ์‹ ๋ขฐ์„ฑ ์ฃผ์„์„ ์ˆ˜์ง‘ํ•˜๊ธฐ ์œ„ํ•œ ํŒŒ์ดํ”„๋ผ์ธ. ๋จผ์ €, (a) ๊ณ„์ธต์  ๋ณ‘ํ•ฉ์„ ํ†ตํ•ด ์š”์•ฝ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, (b) GPT-4๋กœ ํ”„๋กฌํ”„ํŠธํ•˜์—ฌ ๋งฅ๋ฝ์ด ์—†๋Š” ์ฃผ์žฅ์„ ์ถ”์ถœํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ, (c) ์ด๋Ÿฌํ•œ ์ฃผ์žฅ์— ๋Œ€ํ•œ ์ธ๊ฐ„ ํ‰๊ฐ€๋ฅผ ์ˆ˜ํ–‰ํ•˜์—ฌ ์ฃผ์„ ์ž‘์„ฑ์ž๊ฐ€ ๊ฐ ์ฃผ์žฅ์˜ ์œ ํšจ์„ฑ์„ ๊ฒ€์‚ฌํ•˜๊ณ  ํ• ๋‹น๋œ ๋ ˆ์ด๋ธ”์— ๋Œ€ํ•œ ์ถ”๋ก ๊ณผ ์ฆ๊ฑฐ๋ฅผ ์ œ๊ณตํ•˜๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.
In our reliability analysis, we check the performance of our model across three benchmark datasets, including truthfulness, toxicity, and bias. We conduct this analysis in both English and Traditional Chinese. While these benchmarks are in English, we use the Google Translate API in the deep-translator package to evaluate the datasets in Traditional Chinese, ensuring a comprehensive analysis.
์šฐ๋ฆฌ์˜ ์‹ ๋ขฐ์„ฑ ๋ถ„์„์—์„œ, ์šฐ๋ฆฌ๋Š” ์ง„์‹ค์„ฑ, ๋…์„ฑ, ํŽธํ–ฅ์„ ํฌํ•จํ•œ ์„ธ ๊ฐ€์ง€ ๋ฒค์น˜๋งˆํฌ ๋ฐ์ดํ„ฐ์…‹์—์„œ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ถ„์„์€ ์˜์–ด์™€ ๋ฒˆ์ฒด ์ค‘๊ตญ์–ด๋กœ ๋ชจ๋‘ ์ˆ˜ํ–‰๋ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ๋Š” ์˜์–ด๋กœ ๋˜์–ด ์žˆ์ง€๋งŒ, ์šฐ๋ฆฌ๋Š” deep-translator ํŒจํ‚ค์ง€์˜ Google Translate API๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ฐ์ดํ„ฐ์…‹์„ ๋ฒˆ์ฒด ์ค‘๊ตญ์–ด๋กœ ํ‰๊ฐ€ํ•˜์—ฌ ํฌ๊ด„์ ์ธ ๋ถ„์„์„ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
์‹ ๋ขฐ์„ฑ ๋ถ„์„์—์„œ๋Š” ์ง„์‹ค์„ฑ, ๋…์„ฑ, ํŽธํ–ฅ์„ฑ ๋“ฑ ์„ธ ๊ฐ€์ง€ ๋ฒค์น˜๋งˆํฌ ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•ด ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ถ„์„์€ ์˜์–ด์™€ ์ค‘๊ตญ์–ด(็น้ซ”)๋กœ ์ง„ํ–‰๋ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ๋Š” ์˜์–ด๋กœ ๋˜์–ด ์žˆ์ง€๋งŒ, ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์ค‘๊ตญ์–ด๋กœ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด deep-translator ํŒจํ‚ค์ง€์˜ Google ๋ฒˆ์—ญ API๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ข…ํ•ฉ์ ์ธ ๋ถ„์„์„ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค.
In the voting phase, we compute the BLEU score using _sacreBLEU_(Post, 2018) to evaluate the similarity between each of the generated samples. The sample with the highest cumulative BLEU score is selected as the final answer. This method ensures that the final output is the most representative and accurate piece of code as determined by consensus through similarity scoring among the samples.
ํˆฌํ‘œ ๋‹จ๊ณ„์—์„œ๋Š” _sacreBLEU_ (Post, 2018)์„ ์‚ฌ์šฉํ•˜์—ฌ ์ƒ์„ฑ๋œ ๊ฐ ์ƒ˜ํ”Œ ๊ฐ„์˜ ์œ ์‚ฌ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด BLEU ์ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. ๋ˆ„์  BLEU ์ ์ˆ˜๊ฐ€ ๊ฐ€์žฅ ๋†’์€ ์ƒ˜ํ”Œ์ด ์ตœ์ข… ๋‹ต๋ณ€์œผ๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ ์ƒ˜ํ”Œ ๊ฐ„์˜ ์œ ์‚ฌ์„ฑ ์ ์ˆ˜๋ฅผ ํ†ตํ•œ ํ•ฉ์˜์— ์˜ํ•ด ์ตœ์ข… ์ถœ๋ ฅ์ด ๊ฐ€์žฅ ๋Œ€ํ‘œ์ ์ด๊ณ  ์ •ํ™•ํ•œ ์ฝ”๋“œ ์กฐ๊ฐ์ž„์„ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
ํˆฌํ‘œ ๋‹จ๊ณ„์—์„œ๋Š” ์ƒ์„ฑ๋œ ๊ฐ ์ƒ˜ํ”Œ ๊ฐ„์˜ ์œ ์‚ฌ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด _sacreBLEU_(Post, 2018)๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ BLEU ์ ์ˆ˜๋ฅผ ๊ณ„์‚ฐํ•ฉ๋‹ˆ๋‹ค. ๋ˆ„์  BLEU ์ ์ˆ˜๊ฐ€ ๊ฐ€์žฅ ๋†’์€ ์ƒ˜ํ”Œ์ด ์ตœ์ข… ๋‹ต์œผ๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค. ์ด ๋ฐฉ๋ฒ•์€ ์ตœ์ข… ์ถœ๋ ฅ์ด ์ƒ˜ํ”Œ ๊ฐ„์˜ ์œ ์‚ฌ์„ฑ ์ ์ˆ˜๋ฅผ ํ†ตํ•œ ํ•ฉ์˜์— ์˜ํ•ด ๊ฒฐ์ •๋œ ๋Œ€๋กœ ๊ฐ€์žฅ ๋Œ€ํ‘œ์ ์ด๊ณ  ์ •ํ™•ํ•œ ์ฝ”๋“œ๊ฐ€ ๋˜๋„๋ก ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
ChatML FormatBeyond the focus on data quality and diversity, our observations revealed that the format of the data substantially influences the model's ultimate performance. To this end, we implemented the ChatML-style format [53]. This structured approach empowers the model to differentiate among various information types, such as system configurations, user inputs, and assistant responses.
๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ๊ณผ ๋‹ค์–‘์„ฑ์— ์ค‘์ ์„ ๋‘๋Š” ๊ฒƒ ์™ธ์—๋„, ์šฐ๋ฆฌ์˜ ๊ด€์ฐฐ ๊ฒฐ๊ณผ ๋ฐ์ดํ„ฐ์˜ ํ˜•์‹์ด ๋ชจ๋ธ์˜ ์ตœ์ข… ์„ฑ๋Šฅ์— ์ƒ๋‹นํ•œ ์˜ํ–ฅ์„ ๋ฏธ์นœ๋‹ค๋Š” ๊ฒƒ์„ ์•Œ๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์šฐ๋ฆฌ๋Š” ChatML ์Šคํƒ€์ผ ํ˜•์‹ [53]์„ ๊ตฌํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ตฌ์กฐํ™”๋œ ์ ‘๊ทผ ๋ฐฉ์‹์€ ๋ชจ๋ธ์ด ์‹œ์Šคํ…œ ๊ตฌ์„ฑ, ์‚ฌ์šฉ์ž ์ž…๋ ฅ, ์–ด์‹œ์Šคํ„ดํŠธ ์‘๋‹ต ๋“ฑ ๋‹ค์–‘ํ•œ ์ •๋ณด ์œ ํ˜•์„ ๊ตฌ๋ณ„ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค.
ChatML ํ˜•์‹๋ฐ์ดํ„ฐ์˜ ํ’ˆ์งˆ๊ณผ ๋‹ค์–‘์„ฑ์— ๋Œ€ํ•œ ์ดˆ์ ์„ ๋„˜์–ด ๋ฐ์ดํ„ฐ์˜ ํ˜•์‹์ด ๋ชจ๋ธ์˜ ๊ถ๊ทน์ ์ธ ์„ฑ๋Šฅ์— ์ƒ๋‹นํ•œ ์˜ํ–ฅ์„ ๋ฏธ์นœ๋‹ค๋Š” ์‚ฌ์‹ค์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ChatML ์Šคํƒ€์ผ ํ˜•์‹[53]์„ ๊ตฌํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ตฌ์กฐํ™”๋œ ์ ‘๊ทผ ๋ฐฉ์‹์„ ํ†ตํ•ด ๋ชจ๋ธ์€ ์‹œ์Šคํ…œ ๊ตฌ์„ฑ, ์‚ฌ์šฉ์ž ์ž…๋ ฅ, ์–ด์‹œ์Šคํ„ดํŠธ ์‘๋‹ต ๋“ฑ ๋‹ค์–‘ํ•œ ์ •๋ณด ์œ ํ˜•์„ ๊ตฌ๋ถ„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
\(\bullet\)_Books._ Compared to other corpus, books provide an important source of formal long texts, which are potentially beneficial for LLMs to learn linguistic knowledge, model long-term dependency, and generate narrative and coherent texts. To obtain open-source book data, existing studies usually adopt the Books3 and Bookcorpus2 datasets, which are available in the Pile dataset [161].
โ€ข_์ฑ…._ ๋‹ค๋ฅธ ๋ง๋ญ‰์น˜์™€ ๋น„๊ตํ•  ๋•Œ, ์ฑ…์€ ๊ณต์‹์ ์ธ ๊ธด ํ…์ŠคํŠธ์˜ ์ค‘์š”ํ•œ ์ถœ์ฒ˜๋ฅผ ์ œ๊ณตํ•˜๋ฉฐ, ์ด๋Š” LLM์ด ์–ธ์–ด์  ์ง€์‹์„ ํ•™์Šตํ•˜๊ณ , ์žฅ๊ธฐ ์˜์กด์„ฑ์„ ๋ชจ๋ธ๋งํ•˜๋ฉฐ, ์„œ์‚ฌ์ ์ด๊ณ  ์ผ๊ด€๋œ ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๋ฐ ์ž ์žฌ์ ์œผ๋กœ ์œ ์ตํ•ฉ๋‹ˆ๋‹ค. ์˜คํ”ˆ ์†Œ์Šค ์ฑ… ๋ฐ์ดํ„ฐ๋ฅผ ์–ป๊ธฐ ์œ„ํ•ด, ๊ธฐ์กด ์—ฐ๊ตฌ๋“ค์€ ๋ณดํ†ต Pile ๋ฐ์ดํ„ฐ์…‹ [161]์— ํฌํ•จ๋œ Books3์™€ Bookcorpus2 ๋ฐ์ดํ„ฐ์…‹์„ ์ฑ„ํƒํ•ฉ๋‹ˆ๋‹ค.
\(\bullet\)_Books._ ๋‹ค๋ฅธ ๋ง๋ญ‰์น˜์— ๋น„ํ•ด ์ฑ…์€ ์ž ์žฌ์ ์œผ๋กœ LLM์ด ์–ธ์–ด ์ง€์‹์„ ํ•™์Šตํ•˜๊ณ  ์žฅ๊ธฐ ์ข…์†์„ฑ์„ ๋ชจ๋ธ๋งํ•˜๋ฉฐ ๋‚ด๋Ÿฌํ‹ฐ๋ธŒ์™€ ์ผ๊ด€์„ฑ ์žˆ๋Š” ํ…์ŠคํŠธ๋ฅผ ์ƒ์„ฑํ•˜๋Š” ๋ฐ ์œ ์šฉํ•œ ๊ณต์‹์ ์ธ ์žฅ๋ฌธ ํ…์ŠคํŠธ์˜ ์ค‘์š”ํ•œ ์›์ฒœ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ์—์„œ๋Š” ์˜คํ”ˆ ์†Œ์Šค ๋„์„œ ๋ฐ์ดํ„ฐ๋ฅผ ํ™•๋ณดํ•˜๊ธฐ ์œ„ํ•ด ์ผ๋ฐ˜์ ์œผ๋กœ Pile ๋ฐ์ดํ„ฐ ์„ธํŠธ[161]์—์„œ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” Books3 ๋ฐ Bookcorpus2 ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์ฑ„ํƒํ•ฉ๋‹ˆ๋‹ค.
Table 3 shows the results. Our model trained on the mixing data (240K MATH synthetic data + 240K GSM8K synthetic data) ranked as the second, just behind the GPT-4 and much better than other models. Additionally, we plot the correlation between GSM8K and Hungarian national high-school exam scores in Appendix B. The results show that there is no significant benchmark overfitting in our model.
Table 3์€ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ˜ผํ•ฉ ๋ฐ์ดํ„ฐ(240K MATH ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ + 240K GSM8K ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ)๋กœ ํ›ˆ๋ จ๋œ ์šฐ๋ฆฌ์˜ ๋ชจ๋ธ์€ GPT-4 ๋ฐ”๋กœ ๋’ค๋ฅผ ์ด์–ด ๋‘ ๋ฒˆ์งธ๋กœ ๋†’์€ ์ˆœ์œ„๋ฅผ ๊ธฐ๋กํ–ˆ์œผ๋ฉฐ, ๋‹ค๋ฅธ ๋ชจ๋ธ๋“ค๋ณด๋‹ค ํ›จ์”ฌ ๋›ฐ์–ด๋‚ฌ์Šต๋‹ˆ๋‹ค. ์ถ”๊ฐ€์ ์œผ๋กœ, ๋ถ€๋ก B์—์„œ๋Š” GSM8K์™€ ํ—๊ฐ€๋ฆฌ ๊ตญ๊ฐ€ ๊ณ ๋“ฑํ•™๊ต ์‹œํ—˜ ์ ์ˆ˜ ๊ฐ„์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๊ทธ๋ž˜ํ”„๋กœ ๋‚˜ํƒ€๋ƒˆ์Šต๋‹ˆ๋‹ค. ๊ฒฐ๊ณผ๋Š” ์šฐ๋ฆฌ์˜ ๋ชจ๋ธ์— ์œ ์˜๋ฏธํ•œ ๋ฒค์น˜๋งˆํฌ ๊ณผ์ ํ•ฉ์ด ์—†์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
ํ‘œ 3์€ ๊ทธ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ˜ผํ•ฉ ๋ฐ์ดํ„ฐ(240K MATH ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ + 240K GSM8K ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ)๋กœ ํ•™์Šต๋œ ์šฐ๋ฆฌ ๋ชจ๋ธ์€ GPT-4์— ์ด์–ด ๋‘ ๋ฒˆ์งธ๋กœ ๋†’์€ ์ˆœ์œ„๋ฅผ ์ฐจ์ง€ํ–ˆ์œผ๋ฉฐ ๋‹ค๋ฅธ ๋ชจ๋ธ๋ณด๋‹ค ํ›จ์”ฌ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ถ€๋ก B์— GSM8K์™€ ํ—๊ฐ€๋ฆฌ ๊ตญ๊ฐ€ ๊ณ ๋“ฑํ•™๊ต ์‹œํ—˜ ์ ์ˆ˜ ๊ฐ„์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๊ทธ๋ž˜ํ”„๋กœ ๊ทธ๋ ค ๋ณด์•˜์Šต๋‹ˆ๋‹ค. ๊ทธ ๊ฒฐ๊ณผ, ์šฐ๋ฆฌ ๋ชจ๋ธ์—๋Š” ๋ฒค์น˜๋งˆํฌ ๊ณผ์ ํ•ฉ์ด ํฌ๊ฒŒ ์—†์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
\(\bullet\)_LayerNorm._ In the early research, BatchNorm [265] is a commonly used normalization method. However, it is difficult to deal with sequence data of variable lengths and small-batch data. Thus, LayerNorm [256] is introduced to conduct layerwise normalization. Specifically, the mean and variance over all activations per layer are calculated to recenter and re-scale the activations.
โ€ข LayerNorm: ์ดˆ๊ธฐ ์—ฐ๊ตฌ์—์„œ๋Š” BatchNorm [265]์ด ์ผ๋ฐ˜์ ์œผ๋กœ ์‚ฌ์šฉ๋˜๋Š” ์ •๊ทœํ™” ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๊ฐ€๋ณ€ ๊ธธ์ด์˜ ์‹œํ€€์Šค ๋ฐ์ดํ„ฐ์™€ ์†Œ๊ทœ๋ชจ ๋ฐฐ์น˜ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ๋ฐ ์–ด๋ ค์›€์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ LayerNorm [256]์ด ๋„์ž…๋˜์–ด ๋ ˆ์ด์–ด๋ณ„ ์ •๊ทœํ™”๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๊ฐ ๋ ˆ์ด์–ด์˜ ๋ชจ๋“  ํ™œ์„ฑํ™” ๊ฐ’์— ๋Œ€ํ•ด ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์„ ๊ณ„์‚ฐํ•˜์—ฌ ํ™œ์„ฑํ™”๋ฅผ ์žฌ์ค‘์‹ฌํ™”ํ•˜๊ณ  ์žฌ์กฐ์ •ํ•ฉ๋‹ˆ๋‹ค.
\(\bullet\)_LayerNorm._ ์ดˆ๊ธฐ ์—ฐ๊ตฌ์—์„œ BatchNorm [265]์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์‚ฌ์šฉ๋˜๋Š” ์ •๊ทœํ™” ๋ฐฉ๋ฒ•์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๊ฐ€๋ณ€ ๊ธธ์ด์˜ ์‹œํ€€์Šค ๋ฐ์ดํ„ฐ์™€ ์†Œ๋ฐฐ์น˜ ๋ฐ์ดํ„ฐ๋ฅผ ์ฒ˜๋ฆฌํ•˜๊ธฐ๋Š” ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ ˆ์ด์–ด๋ณ„ ์ •๊ทœํ™”๋ฅผ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•ด LayerNorm [256]์ด ๋„์ž…๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๋ ˆ์ด์–ด๋‹น ๋ชจ๋“  ํ™œ์„ฑํ™”์— ๋Œ€ํ•œ ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์„ ๊ณ„์‚ฐํ•˜์—ฌ ํ™œ์„ฑํ™”๋ฅผ ๋‹ค์‹œ ์ค‘์‹ฌํ™”ํ•˜๊ณ  ์žฌ๊ทœ๋ชจํ™”ํ•ฉ๋‹ˆ๋‹ค.
Figure 2: R-GSM example where the original problem can be correctly solved by all LLMs in our evaluation, but all of them failed on the reordered one. Different calculation steps and their corresponding problem statements are annotated in light blue. Specifically, the reasoning steps of the original problem follows the ordering of problem statements, while the reordered problem does not.
Figure 2: R-GSM ์˜ˆ์ œ์—์„œ๋Š” ์›๋ž˜ ๋ฌธ์ œ๋ฅผ ํ‰๊ฐ€์— ์‚ฌ์šฉ๋œ ๋ชจ๋“  LLM์ด ์˜ฌ๋ฐ”๋ฅด๊ฒŒ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์—ˆ์ง€๋งŒ, ์žฌ์ •๋ ฌ๋œ ๋ฌธ์ œ์—์„œ๋Š” ๋ชจ๋‘ ์‹คํŒจํ–ˆ์Šต๋‹ˆ๋‹ค. ์„œ๋กœ ๋‹ค๋ฅธ ๊ณ„์‚ฐ ๋‹จ๊ณ„์™€ ํ•ด๋‹น ๋ฌธ์ œ ์ง„์ˆ ์€ ์—ฐํ•œ ํŒŒ๋ž€์ƒ‰์œผ๋กœ ์ฃผ์„์ด ๋‹ฌ๋ ค ์žˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ์›๋ž˜ ๋ฌธ์ œ์˜ ์ถ”๋ก  ๋‹จ๊ณ„๋Š” ๋ฌธ์ œ ์ง„์ˆ ์˜ ์ˆœ์„œ๋ฅผ ๋”ฐ๋ฅด์ง€๋งŒ, ์žฌ์ •๋ ฌ๋œ ๋ฌธ์ œ๋Š” ๊ทธ๋ ‡์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 2: ํ‰๊ฐ€์— ์ฐธ์—ฌํ•œ ๋ชจ๋“  LLM์ด ์›๋ž˜ ๋ฌธ์ œ๋ฅผ ์˜ฌ๋ฐ”๋ฅด๊ฒŒ ํ•ด๊ฒฐํ•  ์ˆ˜ ์žˆ์ง€๋งŒ ์ˆœ์„œ๊ฐ€ ๋ฐ”๋€ ๋ฌธ์ œ์—์„œ๋Š” ๋ชจ๋‘ ์‹คํŒจํ•œ R-GSM ์˜ˆ์‹œ์ž…๋‹ˆ๋‹ค. ์„œ๋กœ ๋‹ค๋ฅธ ๊ณ„์‚ฐ ๋‹จ๊ณ„์™€ ๊ทธ์— ํ•ด๋‹นํ•˜๋Š” ๋ฌธ์ œ ๋ฌธ์žฅ์ด ํ•˜๋Š˜์ƒ‰์œผ๋กœ ์ฃผ์„ ์ฒ˜๋ฆฌ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์›๋ž˜ ๋ฌธ์ œ์˜ ์ถ”๋ก  ๋‹จ๊ณ„๋Š” ๋ฌธ์ œ ์ง„์ˆ ์˜ ์ˆœ์„œ๋ฅผ ๋”ฐ๋ฅด์ง€๋งŒ, ์ˆœ์„œ๊ฐ€ ๋ฐ”๋€ ๋ฌธ์ œ๋Š” ๊ทธ๋ ‡์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
**Differences in Sampling Compared to Online RL.** Standard auto-regressive sampling can bring more reasoning paths to one query. However, we find that the sampled reasoning paths have minor differences in Fig. 7, such as word substitution (orange part), rewriting of formulas (red part), rearrangement of syntax (brown part), but the correct intermediate steps are consistent (green part).
**์˜จ๋ผ์ธ RL๊ณผ ๋น„๊ตํ•œ ์ƒ˜ํ”Œ๋ง์˜ ์ฐจ์ด์ .** ํ‘œ์ค€ ์ž๋™ ํšŒ๊ท€ ์ƒ˜ํ”Œ๋ง์€ ํ•˜๋‚˜์˜ ์ฟผ๋ฆฌ์— ๋Œ€ํ•ด ๋” ๋งŽ์€ ์ถ”๋ก  ๊ฒฝ๋กœ๋ฅผ ๊ฐ€์ ธ์˜ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์šฐ๋ฆฌ๋Š” ์ƒ˜ํ”Œ๋ง๋œ ์ถ”๋ก  ๊ฒฝ๋กœ๊ฐ€ ๊ทธ๋ฆผ 7์—์„œ ๋‹จ์–ด ๋Œ€์ฒด(์ฃผํ™ฉ์ƒ‰ ๋ถ€๋ถ„), ๊ณต์‹ ์žฌ์ž‘์„ฑ(๋นจ๊ฐ„์ƒ‰ ๋ถ€๋ถ„), ๊ตฌ๋ฌธ ์žฌ๋ฐฐ์—ด(๊ฐˆ์ƒ‰ ๋ถ€๋ถ„)๊ณผ ๊ฐ™์€ ์‚ฌ์†Œํ•œ ์ฐจ์ด์ ์ด ์žˆ์ง€๋งŒ, ์˜ฌ๋ฐ”๋ฅธ ์ค‘๊ฐ„ ๋‹จ๊ณ„๋Š” ์ผ๊ด€๋˜๊ฒŒ ์œ ์ง€๋œ๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค(๋…น์ƒ‰ ๋ถ€๋ถ„).
**์˜จ๋ผ์ธ RL๊ณผ ๋น„๊ตํ•œ ์ƒ˜ํ”Œ๋ง์˜ ์ฐจ์ด์ .** ํ‘œ์ค€ ์ž๋™ ํšŒ๊ท€ ์ƒ˜ํ”Œ๋ง์€ ํ•˜๋‚˜์˜ ์ฟผ๋ฆฌ์— ๋” ๋งŽ์€ ์ถ”๋ก  ๊ฒฝ๋กœ๋ฅผ ๊ฐ€์ ธ์˜ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ƒ˜ํ”Œ๋ง๋œ ์ถ”๋ก  ๊ฒฝ๋กœ๋Š” Fig. 7์—์„œ ๋‹จ์–ด ๋Œ€์ฒด(์ฃผํ™ฉ์ƒ‰ ๋ถ€๋ถ„), ๊ณต์‹ ์žฌ์ž‘์„ฑ(๋นจ๊ฐ„์ƒ‰ ๋ถ€๋ถ„), ๊ตฌ๋ฌธ ์žฌ๋ฐฐ์—ด(๊ฐˆ์ƒ‰ ๋ถ€๋ถ„) ๋“ฑ ์‚ฌ์†Œํ•œ ์ฐจ์ด๊ฐ€ ์žˆ์ง€๋งŒ ์˜ฌ๋ฐ”๋ฅธ ์ค‘๊ฐ„ ๋‹จ๊ณ„๋Š” ์ผ๊ด€์„ฑ์ด ์žˆ์Šต๋‹ˆ๋‹ค(๋…น์ƒ‰ ๋ถ€๋ถ„).
Allen-Zhu and Li [3] introduced a synthetic biography dataset comprising \(N\) individuals, each characterized by six attributes: birth date, birth city, university, major, employer, and working city.9 To translate these tuples into natural language, in their bioS dataset, each individual is described by six randomly selected English sentence templates corresponding to their attributes.
Allen-Zhu์™€ Li [3]๋Š” \(N\)๋ช…์˜ ๊ฐœ์ธ์œผ๋กœ ๊ตฌ์„ฑ๋œ ํ•ฉ์„ฑ ์ „๊ธฐ ๋ฐ์ดํ„ฐ์…‹์„ ์†Œ๊ฐœํ–ˆ์œผ๋ฉฐ, ๊ฐ ๊ฐœ์ธ์€ ์ƒ๋…„์›”์ผ, ์ถœ์ƒ ๋„์‹œ, ๋Œ€ํ•™๊ต, ์ „๊ณต, ๊ณ ์šฉ์ฃผ, ๊ทผ๋ฌด ๋„์‹œ์˜ ์—ฌ์„ฏ ๊ฐ€์ง€ ์†์„ฑ์œผ๋กœ ํŠน์ง•์ง€์–ด์ง‘๋‹ˆ๋‹ค. ์ด๋“ค์˜ bioS ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ด๋Ÿฌํ•œ ํŠœํ”Œ์„ ์ž์—ฐ์–ด๋กœ ๋ฒˆ์—ญํ•˜๊ธฐ ์œ„ํ•ด, ๊ฐ ๊ฐœ์ธ์€ ํ•ด๋‹น ์†์„ฑ์— ๋Œ€์‘ํ•˜๋Š” ์—ฌ์„ฏ ๊ฐœ์˜ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒ๋œ ์˜์–ด ๋ฌธ์žฅ ํ…œํ”Œ๋ฆฟ์œผ๋กœ ์„ค๋ช…๋ฉ๋‹ˆ๋‹ค.
Allen-Zhu์™€ Li [3]๋Š” ๊ฐ๊ฐ 6๊ฐœ์˜ ์†์„ฑ(์ƒ๋…„์›”์ผ, ์ถœ์ƒ ๋„์‹œ, ๋Œ€ํ•™, ์ „๊ณต, ๊ณ ์šฉ์ฃผ, ๊ทผ๋ฌด ๋„์‹œ)์œผ๋กœ ํŠน์ง•์ง€์›Œ์ง€๋Š” \(N\)๋ช…์˜ ๊ฐœ์ธ์œผ๋กœ ๊ตฌ์„ฑ๋œ ํ•ฉ์„ฑ ์•ฝ๋ ฅ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์†Œ๊ฐœํ–ˆ๋‹ค.9 ์ด๋Ÿฌํ•œ ํŠœํ”Œ์„ ์ž์—ฐ์–ด๋กœ ๋ณ€ํ™˜ํ•˜๊ธฐ ์œ„ํ•ด, ์ด๋“ค์˜ bioS ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ๋Š” ๊ฐ ๊ฐœ์ธ์— ๋Œ€ํ•ด ํ•ด๋‹น ์†์„ฑ์— ๋Œ€์‘ํ•˜๋Š” 6๊ฐœ์˜ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒ๋œ ์˜์–ด ๋ฌธ์žฅ ํ…œํ”Œ๋ฆฟ์œผ๋กœ ์„ค๋ช…ํ•œ๋‹ค.
1. Split the input data into individual sentences using the NLTK's sentence splitter function. 2. Examine the first sentence for the presence of delimiters. 3. If a delimiter is detected, check the preceding segment for unwanted elements. 4. If unwanted elements are found, discard the preceding segment (before an occurrence of "\n\n" or ":"). 5. Modify and return the filtered paragraph.
1. NLTK์˜ ๋ฌธ์žฅ ๋ถ„ํ•  ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•˜์—ฌ ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋ฅผ ๊ฐœ๋ณ„ ๋ฌธ์žฅ์œผ๋กœ ๋ถ„ํ• ํ•ฉ๋‹ˆ๋‹ค. 2. ์ฒซ ๋ฒˆ์งธ ๋ฌธ์žฅ์—์„œ ๊ตฌ๋ถ„์ž๊ฐ€ ์žˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. 3. ๊ตฌ๋ถ„์ž๊ฐ€ ๊ฐ์ง€๋˜๋ฉด, ์ด์ „ ๊ตฌ๊ฐ„์— ์›์น˜ ์•Š๋Š” ์š”์†Œ๊ฐ€ ์žˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. 4. ์›์น˜ ์•Š๋Š” ์š”์†Œ๊ฐ€ ๋ฐœ๊ฒฌ๋˜๋ฉด, ์ด์ „ ๊ตฌ๊ฐ„(โ€œ\n\nโ€ ๋˜๋Š” โ€œ:โ€ ๋ฐœ์ƒ ์ „)์„ ์‚ญ์ œํ•ฉ๋‹ˆ๋‹ค. 5. ์ˆ˜์ •๋œ ํ•„ํ„ฐ๋ง๋œ ๋‹จ๋ฝ์„ ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
1. NLTK์˜ ๋ฌธ์žฅ ๋ถ„ํ• ๊ธฐ ํ•จ์ˆ˜๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ž…๋ ฅ ๋ฐ์ดํ„ฐ๋ฅผ ๊ฐœ๋ณ„ ๋ฌธ์žฅ์œผ๋กœ ๋ถ„ํ• ํ•ฉ๋‹ˆ๋‹ค. 2. ์ฒซ ๋ฒˆ์งธ ๋ฌธ์žฅ์— ๊ตฌ๋ถ„ ๊ธฐํ˜ธ๊ฐ€ ์žˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. 3. ๊ตฌ๋ถ„ ๊ธฐํ˜ธ๊ฐ€ ๊ฐ์ง€๋˜๋ฉด ์ด์ „ ์„ธ๊ทธ๋จผํŠธ์—์„œ ์›์น˜ ์•Š๋Š” ์š”์†Œ๊ฐ€ ์žˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค. 4. ์›์น˜ ์•Š๋Š” ์š”์†Œ๊ฐ€ ๋ฐœ๊ฒฌ๋˜๋ฉด ์ด์ „ ์„ธ๊ทธ๋จผํŠธ("\n\n" ๋˜๋Š” ":"๊ฐ€ ๋ฐœ์ƒํ•˜๊ธฐ ์ „)๋ฅผ ์‚ญ์ œํ•ฉ๋‹ˆ๋‹ค. 5. ํ•„ํ„ฐ๋ง๋œ ๋‹จ๋ฝ์„ ์ˆ˜์ •ํ•˜๊ณ  ๋ฐ˜ํ™˜ํ•ฉ๋‹ˆ๋‹ค.
Results show that the relative performance fluctuates despite a relatively consistent trend across sizes and training steps. This indicates that a mixture is better at small scales probably but does not always perform better at large scales. The longest common sequence of the partial orders among the 20 mixtures in Fig. 9(A) and Fig. 9(B) only reaches lengths of 10 and 11, respectively.
๊ฒฐ๊ณผ๋Š” ํฌ๊ธฐ์™€ ํ›ˆ๋ จ ๋‹จ๊ณ„ ์ „๋ฐ˜์— ๊ฑธ์ณ ๋น„๊ต์  ์ผ๊ด€๋œ ๊ฒฝํ–ฅ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์ƒ๋Œ€์ ์ธ ์„ฑ๋Šฅ์ด ๋ณ€๋™ํ•จ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋Š” ํ˜ผํ•ฉ์ด ์ž‘์€ ๊ทœ๋ชจ์—์„œ๋Š” ๋” ๋‚˜์„ ์ˆ˜ ์žˆ์ง€๋งŒ ํ•ญ์ƒ ํฐ ๊ทœ๋ชจ์—์„œ ๋” ๋‚˜์€ ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜๋Š” ๊ฒƒ์€ ์•„๋‹˜์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. Fig. 9(A)์™€ Fig. 9(B)์— ์žˆ๋Š” 20๊ฐœ์˜ ํ˜ผํ•ฉ๋ฌผ ๊ฐ„์˜ ๋ถ€๋ถ„ ์ˆœ์„œ์˜ ์ตœ์žฅ ๊ณตํ†ต ์‹œํ€€์Šค๋Š” ๊ฐ๊ฐ 10๊ณผ 11์˜ ๊ธธ์ด์— ๋„๋‹ฌํ•ฉ๋‹ˆ๋‹ค.
๊ฒฐ๊ณผ์— ๋”ฐ๋ฅด๋ฉด ์ƒ๋Œ€์  ์„ฑ๋Šฅ์€ ํฌ๊ธฐ์™€ ํ›ˆ๋ จ ๋‹จ๊ณ„์— ๋”ฐ๋ผ ๋น„๊ต์  ์ผ๊ด€๋œ ์ถ”์„ธ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ์ƒ๋Œ€์  ์„ฑ๋Šฅ์ด ๋ณ€๋™ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ํ˜ผํ•ฉ์ด ์ž‘์€ ๊ทœ๋ชจ์—์„œ๋Š” ๋” ๋‚˜์€ ์„ฑ๋Šฅ์„ ๋ณด์ด์ง€๋งŒ ํ•ญ์ƒ ํฐ ๊ทœ๋ชจ์—์„œ๋Š” ๋” ๋‚˜์€ ์„ฑ๋Šฅ์„ ๋ณด์ด์ง€๋Š” ์•Š๋Š”๋‹ค๋Š” ๊ฒƒ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ๊ทธ๋ฆผ 9(A)์™€ ๊ทธ๋ฆผ 9(B)์˜ 20๊ฐœ ํ˜ผํ•ฉ๋ฌผ ์ค‘ ๋ถ€๋ถ„ ์ˆœ์„œ์˜ ๊ฐ€์žฅ ๊ธด ๊ณตํ†ต ์ˆ˜์—ด์€ ๊ฐ๊ฐ ๊ธธ์ด 10๊ณผ 11์— ๋ถˆ๊ณผํ•ฉ๋‹ˆ๋‹ค.
We employ cld3 package to detect the language of each token in the embeddings of each layer, which is a language detection library based on the Compact Language Detector \(3\) model developed by Google. Furthermore, if the detection result is reliable, i.e., cld3.get_language(token).is_reliable\(==True\), we adopt the detection results, otherwise the token is categorized as a non-word.
๊ฐ ๋ ˆ์ด์–ด์˜ ์ž„๋ฒ ๋”ฉ์—์„œ ๊ฐ ํ† ํฐ์˜ ์–ธ์–ด๋ฅผ ๊ฐ์ง€ํ•˜๊ธฐ ์œ„ํ•ด Google์ด ๊ฐœ๋ฐœํ•œ Compact Language Detector \(3\) ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์–ธ์–ด ๊ฐ์ง€ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์ธ cld3 ํŒจํ‚ค์ง€๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ๊ฐ์ง€ ๊ฒฐ๊ณผ๊ฐ€ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒฝ์šฐ, ์ฆ‰ cld3.get_language(token).is_reliable\(==True\)์ธ ๊ฒฝ์šฐ ๊ฐ์ง€ ๊ฒฐ๊ณผ๋ฅผ ์ฑ„ํƒํ•˜๊ณ , ๊ทธ๋ ‡์ง€ ์•Š์€ ๊ฒฝ์šฐ ํ•ด๋‹น ํ† ํฐ์„ ๋น„๋‹จ์–ด๋กœ ๋ถ„๋ฅ˜ํ•ฉ๋‹ˆ๋‹ค.
cld3 ํŒจํ‚ค์ง€๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ ๋ ˆ์ด์–ด์˜ ์ž„๋ฒ ๋”ฉ์—์„œ ๊ฐ ํ† ํฐ์˜ ์–ธ์–ด๋ฅผ ๊ฐ์ง€ํ•˜๋Š”๋ฐ, ์ด๋Š” Google์—์„œ ๊ฐœ๋ฐœํ•œ Compact Language Detector \(3\) ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋Š” ์–ธ์–ด ๊ฐ์ง€ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ์ž…๋‹ˆ๋‹ค. ๋˜ํ•œ ๊ฐ์ง€ ๊ฒฐ๊ณผ๊ฐ€ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒฝ์šฐ, ์ฆ‰ cld3.get_language(token).is_reliable\(==True\)์ธ ๊ฒฝ์šฐ ๊ฐ์ง€ ๊ฒฐ๊ณผ๋ฅผ ์ฑ„ํƒํ•˜๊ณ , ๊ทธ๋ ‡์ง€ ์•Š์€ ๊ฒฝ์šฐ ํ•ด๋‹น ํ† ํฐ์€ ๋‹จ์–ด๊ฐ€ ์•„๋‹Œ ๊ฒƒ์œผ๋กœ ๋ถ„๋ฅ˜ํ•ฉ๋‹ˆ๋‹ค.
Human equivalence score (HEQ) on the other hand, is an alternative to F1 score [203]. HEQ-Q represents the precision of individual questions, wherein an answer is deemed correct if the model's F1 score surpasses the average human F1 score. Likewise, HEQ-D denotes the precision of each dialogue; it is deemed accurate when all questions within the dialogue meet the criteria of HEQ [182].
์ธ๊ฐ„ ๋“ฑ๊ฐ€ ์ ์ˆ˜(HEQ)๋Š” F1 ์ ์ˆ˜ [203]์˜ ๋Œ€์•ˆ์ž…๋‹ˆ๋‹ค. HEQ-Q๋Š” ๊ฐœ๋ณ„ ์งˆ๋ฌธ์˜ ์ •๋ฐ€๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๋ชจ๋ธ์˜ F1 ์ ์ˆ˜๊ฐ€ ํ‰๊ท  ์ธ๊ฐ„ F1 ์ ์ˆ˜๋ฅผ ์ดˆ๊ณผํ•  ๋•Œ ๋‹ต๋ณ€์ด ์˜ฌ๋ฐ”๋ฅธ ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผ๋ฉ๋‹ˆ๋‹ค. ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ, HEQ-D๋Š” ๊ฐ ๋Œ€ํ™”์˜ ์ •๋ฐ€๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๋Œ€ํ™” ๋‚ด ๋ชจ๋“  ์งˆ๋ฌธ์ด HEQ ๊ธฐ์ค€์„ ์ถฉ์กฑํ•  ๋•Œ ์ •ํ™•ํ•œ ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผ๋ฉ๋‹ˆ๋‹ค [182].
๋ฐ˜๋ฉด์— ์ธ๊ฐ„ ๋“ฑ๊ฐ€ ์ ์ˆ˜(HEQ)๋Š” F1 ์ ์ˆ˜[203]์˜ ๋Œ€์•ˆ์ž…๋‹ˆ๋‹ค. HEQ-Q๋Š” ๊ฐœ๋ณ„ ์งˆ๋ฌธ์˜ ์ •ํ™•๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๋ชจ๋ธ์˜ F1 ์ ์ˆ˜๊ฐ€ ํ‰๊ท  ์ธ๊ฐ„ F1 ์ ์ˆ˜๋ฅผ ์ดˆ๊ณผํ•˜๋ฉด ์ •๋‹ต์œผ๋กœ ๊ฐ„์ฃผํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ HEQ-D๋Š” ๊ฐ ๋Œ€ํ™”์˜ ์ •ํ™•๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋ฉฐ, ๋Œ€ํ™” ๋‚ด์˜ ๋ชจ๋“  ์งˆ๋ฌธ์ด HEQ ๊ธฐ์ค€์„ ์ถฉ์กฑํ•  ๋•Œ ์ •ํ™•ํ•œ ๊ฒƒ์œผ๋กœ ๊ฐ„์ฃผ๋ฉ๋‹ˆ๋‹ค[182].
In Section 10, we shall also consider knowledge that has _extremely low (e.g., 1) or high (e.g., 1M) exposures_. It may not be interesting to study them in isolation, but it becomes more intriguing when they are examined alongside "standard" knowledge, which has appeared, for instance, for 100 exposures, and how this impacts the model's capacity. These will be our Result 10 through 12.
10์žฅ์—์„œ๋Š” _๊ทนํžˆ ๋‚ฎ์€(์˜ˆ: 1) ๋˜๋Š” ๋†’์€(์˜ˆ: 1M) ๋…ธ์ถœ_์„ ๊ฐ€์ง„ ์ง€์‹๋„ ๊ณ ๋ คํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋“ค์„ ๋‹จ๋…์œผ๋กœ ์—ฐ๊ตฌํ•˜๋Š” ๊ฒƒ์€ ํฅ๋ฏธ๋กญ์ง€ ์•Š์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ์˜ˆ๋ฅผ ๋“ค์–ด 100๋ฒˆ ๋…ธ์ถœ๋œ "ํ‘œ์ค€" ์ง€์‹๊ณผ ํ•จ๊ป˜ ๊ฒ€ํ† ํ•  ๋•Œ ๋” ํฅ๋ฏธ๋กœ์›Œ์ง€๋ฉฐ, ์ด๊ฒƒ์ด ๋ชจ๋ธ์˜ ์—ญ๋Ÿ‰์— ์–ด๋–ค ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š”์ง€ ์‚ดํŽด๋ณผ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์šฐ๋ฆฌ์˜ ๊ฒฐ๊ณผ 10์—์„œ 12๊ฐ€ ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
์„น์…˜ 10์—์„œ๋Š” _๋งค์šฐ ๋‚ฎ์€(์˜ˆ: 1) ๋˜๋Š” ๋†’์€(์˜ˆ: 1M) ๋…ธ์ถœ_์„ ๊ฐ€์ง„ ์ง€์‹์— ๋Œ€ํ•ด์„œ๋„ ๊ณ ๋ คํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ง€์‹์„ ๊ฐœ๋ณ„์ ์œผ๋กœ ์—ฐ๊ตฌํ•˜๋Š” ๊ฒƒ์€ ํฅ๋ฏธ๋กญ์ง€ ์•Š์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ์˜ˆ๋ฅผ ๋“ค์–ด 100๋ฒˆ ๋…ธ์ถœ๋œ "ํ‘œ์ค€" ์ง€์‹๊ณผ ํ•จ๊ป˜ ์—ฐ๊ตฌํ•˜๊ณ  ์ด๊ฒƒ์ด ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์— ์–ด๋–ค ์˜ํ–ฅ์„ ๋ฏธ์น˜๋Š”์ง€ ์‚ดํŽด๋ณด๋ฉด ๋” ํฅ๋ฏธ๋กœ์›Œ์ง‘๋‹ˆ๋‹ค. ์ด๊ฒƒ์ด ๋ฐ”๋กœ ๊ฒฐ๊ณผ 10๋ถ€ํ„ฐ 12๊นŒ์ง€์ž…๋‹ˆ๋‹ค.
Figure 13: **Empirical Hutter LLM.** Bigram model with deterministic labeling function. Initial model trained on \(T_{0}=100,000\) samples. It generates \(T\) samples for Gen 1. Starting from Gen 2 models are trained on data generated by the most powerful model from the previous generation. \(\beta=3/2\). In this setting, there is mild model collapse coming from the finite sample bias.
Figure 13: **Empirical Hutter LLM.** ๊ฒฐ์ •๋ก ์  ๋ผ๋ฒจ๋ง ํ•จ์ˆ˜๊ฐ€ ์žˆ๋Š” ๋ฐ”์ด๊ทธ๋žจ ๋ชจ๋ธ. ์ดˆ๊ธฐ ๋ชจ๋ธ์€ \(T_{0}=100,000\) ์ƒ˜ํ”Œ๋กœ ํ›ˆ๋ จ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ์€ Gen 1์„ ์œ„ํ•ด \(T\) ์ƒ˜ํ”Œ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. Gen 2๋ถ€ํ„ฐ๋Š” ์ด์ „ ์„ธ๋Œ€์—์„œ ๊ฐ€์žฅ ๊ฐ•๋ ฅํ•œ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋ชจ๋ธ์ด ํ›ˆ๋ จ๋ฉ๋‹ˆ๋‹ค. \(\beta=3/2\). ์ด ์„ค์ •์—์„œ๋Š” ์œ ํ•œ ์ƒ˜ํ”Œ ํŽธํ–ฅ์œผ๋กœ ์ธํ•œ ๊ฒฝ๋ฏธํ•œ ๋ชจ๋ธ ๋ถ•๊ดด๊ฐ€ ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 13: **๊ฒฝํ—˜์  ํ—ˆํ„ฐ LLM.** ๊ฒฐ์ •๋ก ์  ๋ผ๋ฒจ๋ง ํ•จ์ˆ˜๊ฐ€ ์žˆ๋Š” ๋ฐ”์ด๊ทธ๋žจ ๋ชจ๋ธ. ์ดˆ๊ธฐ ๋ชจ๋ธ์€ \(T_{0}=100,000\) ์ƒ˜ํ”Œ์— ๋Œ€ํ•ด ํ•™์Šต๋ฉ๋‹ˆ๋‹ค. Gen 1์— ๋Œ€ํ•ด \(T\) ์ƒ˜ํ”Œ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. 2์„ธ๋Œ€๋ถ€ํ„ฐ๋Š” ์ด์ „ ์„ธ๋Œ€์˜ ๊ฐ€์žฅ ๊ฐ•๋ ฅํ•œ ๋ชจ๋ธ์ด ์ƒ์„ฑํ•œ ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•ด ๋ชจ๋ธ์„ ํ•™์Šตํ•ฉ๋‹ˆ๋‹ค. \(\๋ฒ ํƒ€=3/2\). ์ด ์„ค์ •์—์„œ๋Š” ์œ ํ•œ ์ƒ˜ํ”Œ ํŽธํ–ฅ์œผ๋กœ ์ธํ•ด ์•ฝ๊ฐ„์˜ ๋ชจ๋ธ ๋ถ•๊ดด๊ฐ€ ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค.
Based on Hoffmann et al.'s report of having "over 400" observations, we interpret this to mean they likely had between 400 and 500 data points. If, as they claimed, they used only the final loss values for each trained model, it seems unlikely that they would have had hundreds of thousands of observations. However, further clarification from the authors on this point would be helpful.
Hoffmann ๋“ฑ์€ "400๊ฐœ ์ด์ƒ์˜" ๊ด€์ฐฐ์„ ํ–ˆ๋‹ค๊ณ  ๋ณด๊ณ ํ–ˆ๊ธฐ ๋•Œ๋ฌธ์—, ์šฐ๋ฆฌ๋Š” ์ด๋“ค์ด 400์—์„œ 500 ์‚ฌ์ด์˜ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋ฅผ ๊ฐ€์กŒ์„ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค๊ณ  ํ•ด์„ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋“ค์ด ์ฃผ์žฅํ•œ ๋Œ€๋กœ ๊ฐ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ์— ๋Œ€ํ•ด ์ตœ์ข… ์†์‹ค ๊ฐ’๋งŒ ์‚ฌ์šฉํ–ˆ๋‹ค๋ฉด, ์ˆ˜์‹ญ๋งŒ ๊ฐœ์˜ ๊ด€์ฐฐ์„ ๊ฐ€์กŒ์„ ๊ฐ€๋Šฅ์„ฑ์€ ๋‚ฎ์•„ ๋ณด์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด ์ ์— ๋Œ€ํ•ด ์ €์ž๋“ค์˜ ์ถ”๊ฐ€์ ์ธ ์„ค๋ช…์ด ์žˆ์œผ๋ฉด ๋„์›€์ด ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Hoffmann ๋“ฑ์˜ ๋ณด๊ณ ์„œ์— ๋”ฐ๋ฅด๋ฉด "400๊ฐœ ์ด์ƒ"์˜ ๊ด€์ธก์น˜๊ฐ€ ์žˆ๋‹ค๊ณ  ๋ณด๊ณ ํ•œ ๋ฐ”๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ 400~500๊ฐœ์˜ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๊ฐ€ ์žˆ์—ˆ์„ ๊ฒƒ์œผ๋กœ ํ•ด์„ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋“ค์ด ์ฃผ์žฅํ•œ ๊ฒƒ์ฒ˜๋Ÿผ ๊ฐ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ์— ๋Œ€ํ•ด ์ตœ์ข… ์†์‹ค ๊ฐ’๋งŒ ์‚ฌ์šฉํ–ˆ๋‹ค๋ฉด ์ˆ˜์‹ญ๋งŒ ๊ฐœ์˜ ๊ด€์ธก์น˜๊ฐ€ ์žˆ์—ˆ์„ ๊ฐ€๋Šฅ์„ฑ์€ ๋‚ฎ์•„ ๋ณด์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ด ์ ์— ๋Œ€ํ•œ ์ €์ž์˜ ์ถ”๊ฐ€ ์„ค๋ช…์ด ๋„์›€์ด ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Traditional theory on overparameterization suggests that scaling up model size in sufficiently trained models can enhance memorization of training data [6], improve generalization error [15, 27, 28], and better fit complex target functions [5, 23]. However, these results often overlook large constant or polynomial factors, leading to a significant discrepancy from practical outcomes.
์ „ํ†ต์ ์ธ ๊ณผ๋งค๊ฐœ๋ณ€์ˆ˜ํ™” ์ด๋ก ์€ ์ถฉ๋ถ„ํžˆ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ์—์„œ ๋ชจ๋ธ ํฌ๊ธฐ๋ฅผ ํ™•์žฅํ•˜๋ฉด ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์˜ ์•”๊ธฐ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ณ  [6], ์ผ๋ฐ˜ํ™” ์˜ค๋ฅ˜๋ฅผ ๊ฐœ์„ ํ•˜๋ฉฐ [15, 27, 28], ๋ณต์žกํ•œ ๋ชฉํ‘œ ํ•จ์ˆ˜๋ฅผ ๋” ์ž˜ ๋งž์ถœ ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค [5, 23]. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋Š” ์ข…์ข… ํฐ ์ƒ์ˆ˜ ๋˜๋Š” ๋‹คํ•ญ์‹ ์š”์†Œ๋ฅผ ๊ฐ„๊ณผํ•˜์—ฌ ์‹ค์ œ ๊ฒฐ๊ณผ์™€ ์ƒ๋‹นํ•œ ์ฐจ์ด๋ฅผ ์ดˆ๋ž˜ํ•ฉ๋‹ˆ๋‹ค.
ํŒŒ๋ผ๋ฏธํ„ฐ ๊ณผ๋Œ€ํ™”์— ๋Œ€ํ•œ ์ „ํ†ต์ ์ธ ์ด๋ก ์— ๋”ฐ๋ฅด๋ฉด ์ถฉ๋ถ„ํžˆ ํ•™์Šต๋œ ๋ชจ๋ธ์—์„œ ๋ชจ๋ธ ํฌ๊ธฐ๋ฅผ ํ™•์žฅํ•˜๋ฉด ํ•™์Šต ๋ฐ์ดํ„ฐ์˜ ์•”๊ธฐ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ณ [6], ์ผ๋ฐ˜ํ™” ์˜ค๋ฅ˜๋ฅผ ๊ฐœ์„ ํ•˜๋ฉฐ[15, 27, 28], ๋ณต์žกํ•œ ๋ชฉํ‘œ ํ•จ์ˆ˜์— ๋” ์ž˜ ๋งž์ถœ ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค[5, 23]. ๊ทธ๋Ÿฌ๋‚˜ ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋Š” ์ข…์ข… ํฐ ์ƒ์ˆ˜ ๋˜๋Š” ๋‹คํ•ญ์‹ ์š”์ธ์„ ๊ฐ„๊ณผํ•˜์—ฌ ์‹ค์ œ ๊ฒฐ๊ณผ์™€ ์ƒ๋‹นํ•œ ์ฐจ์ด๋ฅผ ์ดˆ๋ž˜ํ•ฉ๋‹ˆ๋‹ค.