mpasila/Finnish-Viking-Alpaca-V1-7B-exl2-4bpw

This is an ExLlamaV2 quantized model in 4bpw of mpasila/Finnish-Viking-Alpaca-V1-7B using the default calibration dataset.

Original Model card:

Model Card for Finnish-Viking-Alpaca-V1-7B

This is a merge of mpasila/Finnish-Viking-Alpaca-V1-LoRA-7B.

LoRA trained with text-generation-webui in 4-bit using LumiOpen/Viking-7B as the base model for 1 epoch. Dataset used with the LoRA is pinzhenchen/alpaca-cleaned-fi.

It uses Alpaca format but with a translated instruction at the start:

{
    "instruction,output": "Alla on ohje, jossa kuvataan tehtävä. Kirjoita vastaus, joka täyttää pyynnön asianmukaisesti.\n\n### Instruction:\n%instruction%\n\n### Response:\n%output%",
    "instruction,input,output": "Alla on ohje, jossa kuvataan tehtävä ja joka on yhdistetty kontekstia lisäävään syötteeseen. Kirjoita vastaus, joka täyttää pyynnön asianmukaisesti.\n\n### Instruction:\n%instruction%\n\n### Input:\n%input%\n\n### Response:\n%output%"
}

Merged using this Colab notebook. It might not be the best way to merge a quantized LoRA on to a float16 model but I just wanted to quickly do something. You can try merging it better if you want.

Evaluation

Model	Size	Type	FIN-bench (score)
mpasila/Finnish-Viking-Alpaca-V1-7B	7B	Instruct	0.3943
mpasila/NordicAlpaca-Finnish-V1-7B	7B	Instruct	0.3891
Finnish-NLP/llama-7b-finnish-instruct-v0.1	7B	Instruct	0.4365
Finnish-NLP/llama-7b-finnish-instruct-v0.2	7B	Instruct	0.3993
Finnish-NLP/llama-7b-finnish	7B	Base	0.2350
LumiOpen/Viking-7B (1000B)	7B	Base	0.3721
HPLT/gpt-7b-nordic-prerelease	7B	Base	0.3169

Source

FIN-bench scores:

Task	Version	Metric	Value		Stderr
bigbench_analogies	0	multiple_choice_grade	0.6308	±	0.0425
bigbench_arithmetic_1_digit_addition	0	multiple_choice_grade	0.6400	±	0.0482
bigbench_arithmetic_1_digit_division	0	multiple_choice_grade	0.7391	±	0.0936
bigbench_arithmetic_1_digit_multiplication	0	multiple_choice_grade	0.2800	±	0.0451
bigbench_arithmetic_1_digit_subtraction	0	multiple_choice_grade	0.5000	±	0.0503
bigbench_arithmetic_2_digit_addition	0	multiple_choice_grade	0.1800	±	0.0386
bigbench_arithmetic_2_digit_division	0	multiple_choice_grade	0.4800	±	0.0502
bigbench_arithmetic_2_digit_multiplication	0	multiple_choice_grade	0.0800	±	0.0273
bigbench_arithmetic_2_digit_subtraction	0	multiple_choice_grade	0.2500	±	0.0435
bigbench_arithmetic_3_digit_addition	0	multiple_choice_grade	0.1800	±	0.0386
bigbench_arithmetic_3_digit_division	0	multiple_choice_grade	0.2500	±	0.0435
bigbench_arithmetic_3_digit_multiplication	0	multiple_choice_grade	0.1700	±	0.0378
bigbench_arithmetic_3_digit_subtraction	0	multiple_choice_grade	0.5000	±	0.0503
bigbench_arithmetic_4_digit_addition	0	multiple_choice_grade	0.2600	±	0.0441
bigbench_arithmetic_4_digit_division	0	multiple_choice_grade	0.2500	±	0.0435
bigbench_arithmetic_4_digit_multiplication	0	multiple_choice_grade	0.2100	±	0.0409
bigbench_arithmetic_4_digit_subtraction	0	multiple_choice_grade	0.5200	±	0.0502
bigbench_arithmetic_5_digit_addition	0	multiple_choice_grade	0.3900	±	0.0490
bigbench_arithmetic_5_digit_division	0	multiple_choice_grade	0.1600	±	0.0368
bigbench_arithmetic_5_digit_multiplication	0	multiple_choice_grade	0.1000	±	0.0302
bigbench_arithmetic_5_digit_subtraction	0	multiple_choice_grade	0.6100	±	0.0490
bigbench_cause_and_effect_one_sentence	0	multiple_choice_grade	0.6471	±	0.0676
bigbench_cause_and_effect_one_sentence_no_prompt	0	multiple_choice_grade	0.6863	±	0.0656
bigbench_cause_and_effect_two_sentences	0	multiple_choice_grade	0.3922	±	0.0690
bigbench_emotions	0	multiple_choice_grade	0.2812	±	0.0357
bigbench_empirical_judgments	0	multiple_choice_grade	0.2828	±	0.0455
bigbench_general_knowledge	0	multiple_choice_grade	0.4000	±	0.0590
bigbench_hhh_alignment_harmless	0	multiple_choice_grade	0.3621	±	0.0637
bigbench_hhh_alignment_helpful	0	multiple_choice_grade	0.3559	±	0.0629
bigbench_hhh_alignment_honest	0	multiple_choice_grade	0.3729	±	0.0635
bigbench_hhh_alignment_other	0	multiple_choice_grade	0.5581	±	0.0766
bigbench_intent_recognition	0	multiple_choice_grade	0.1879	±	0.0149
bigbench_misconceptions	0	multiple_choice_grade	0.5373	±	0.0432
bigbench_paraphrase	0	multiple_choice_grade	0.5150	±	0.0354
bigbench_sentence_ambiguity	0	multiple_choice_grade	0.5000	±	0.0651
bigbench_similarities_abstraction	0	multiple_choice_grade	0.7368	±	0.0508

Framework versions

• PEFT 0.8.2