question
stringlengths
101
4.47k
chosen
stringlengths
3
4.25k
rejected
stringlengths
39
9.29k
Cluster-based FiltersWe further use unsupervised semantic clustering to group web documents. This clustering process enables efficient identification and analysis of documents sharing similar semantic features. The clustered data are subsequently annotated with quality labels, providing essential references for the optimization of Yi's data mixture strategy. Documents identified as low-quality through automatic and manual verification are excluded from the dataset.
ํด๋Ÿฌ์Šคํ„ฐ ๊ธฐ๋ฐ˜ ํ•„ํ„ฐ ์šฐ๋ฆฌ๋Š” ์›น ๋ฌธ์„œ๋ฅผ ๊ทธ๋ฃนํ™”ํ•˜๊ธฐ ์œ„ํ•ด ๋น„์ง€๋„ ์˜๋ฏธ ํด๋Ÿฌ์Šคํ„ฐ๋ง์„ ์ถ”๊ฐ€๋กœ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ํด๋Ÿฌ์Šคํ„ฐ๋ง ๊ณผ์ •์€ ์œ ์‚ฌํ•œ ์˜๋ฏธ์  ํŠน์ง•์„ ๊ณต์œ ํ•˜๋Š” ๋ฌธ์„œ๋“ค์„ ํšจ์œจ์ ์œผ๋กœ ์‹๋ณ„ํ•˜๊ณ  ๋ถ„์„ํ•  ์ˆ˜ ์žˆ๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ํด๋Ÿฌ์Šคํ„ฐ๋ง๋œ ๋ฐ์ดํ„ฐ๋Š” ์ดํ›„ ํ’ˆ์งˆ ๋ผ๋ฒจ๋กœ ์ฃผ์„์ด ๋‹ฌ๋ฆฌ๋ฉฐ, ์ด๋Š” Yi์˜ ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ ์ „๋žต ์ตœ์ ํ™”๋ฅผ ์œ„ํ•œ ํ•„์ˆ˜์ ์ธ ์ฐธ๊ณ  ์ž๋ฃŒ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ž๋™ ๋ฐ ์ˆ˜๋™ ๊ฒ€์ฆ์„ ํ†ตํ•ด ์ €ํ’ˆ์งˆ๋กœ ์‹๋ณ„๋œ ๋ฌธ์„œ๋“ค์€ ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ œ์™ธ๋ฉ๋‹ˆ๋‹ค.
ํด๋Ÿฌ์Šคํ„ฐ ๊ธฐ๋ฐ˜ ํ•„ํ„ฐ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๋น„์ง€๋„ ์‹œ๋งจํ‹ฑ ํด๋Ÿฌ์Šคํ„ฐ๋ง์„ ์‚ฌ์šฉํ•˜์—ฌ ์›น ๋ฌธ์„œ๋ฅผ ๊ทธ๋ฃนํ™”ํ•ฉ๋‹ˆ๋‹ค. ์ด ํด๋Ÿฌ์Šคํ„ฐ๋ง ํ”„๋กœ์„ธ์Šค๋ฅผ ํ†ตํ•ด ์œ ์‚ฌํ•œ ์˜๋ฏธ์  ํŠน์ง•์„ ๊ณต์œ ํ•˜๋Š” ๋ฌธ์„œ๋ฅผ ํšจ์œจ์ ์œผ๋กœ ์‹๋ณ„ํ•˜๊ณ  ๋ถ„์„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ดํ›„ ํด๋Ÿฌ์Šคํ„ฐ๋ง๋œ ๋ฐ์ดํ„ฐ์— ํ’ˆ์งˆ ๋ ˆ์ด๋ธ”์„ ์ถ”๊ฐ€ํ•˜์—ฌ Yi์˜ ๋ฐ์ดํ„ฐ ํ˜ผํ•ฉ ์ „๋žต ์ตœ์ ํ™”๋ฅผ ์œ„ํ•œ ํ•„์ˆ˜ ์ฐธ์กฐ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ž๋™ ๋ฐ ์ˆ˜๋™ ๊ฒ€์ฆ์„ ํ†ตํ•ด ์ €ํ’ˆ์งˆ๋กœ ์‹๋ณ„๋œ ๋ฌธ์„œ๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ์ œ์™ธ๋ฉ๋‹ˆ๋‹ค.
Finding all repeated sequences is thus a matter of linearly scanning the suffix array from beginning to end and looking for sequences \(\mathcal{A}_{i},\mathcal{A}_{i+1}\) that share a common prefix of at least some threshold length. Any satisfying sequences are recorded. This algorithm is embarrassingly parallel, and so we can efficiently process the dataset. Based on experimentation (Appendix B), we choose a threshold length of 50 BPE tokens for all experiments.
๋ชจ๋“  ๋ฐ˜๋ณต๋˜๋Š” ์‹œํ€€์Šค๋ฅผ ์ฐพ๋Š” ๊ฒƒ์€ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ๋๊นŒ์ง€ ์„ ํ˜•์ ์œผ๋กœ ์Šค์บ”ํ•˜๋ฉด์„œ ์ผ์ • ๊ธธ์ด ์ด์ƒ์˜ ๊ณตํ†ต ์ ‘๋‘์‚ฌ๋ฅผ ๊ณต์œ ํ•˜๋Š” ์‹œํ€€์Šค \(\mathcal{A}_{i},\mathcal{A}_{i+1}\)๋ฅผ ์ฐพ๋Š” ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค. ๋งŒ์กฑํ•˜๋Š” ์‹œํ€€์Šค๋Š” ๊ธฐ๋ก๋ฉ๋‹ˆ๋‹ค. ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ณ‘๋ ฌ ์ฒ˜๋ฆฌ๊ฐ€ ๋งค์šฐ ์šฉ์ดํ•˜๋ฏ€๋กœ ๋ฐ์ดํ„ฐ๋ฅผ ํšจ์œจ์ ์œผ๋กœ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹คํ—˜ ๊ฒฐ๊ณผ(Appendix B)์— ๋”ฐ๋ผ, ๋ชจ๋“  ์‹คํ—˜์—์„œ 50 BPE ํ† ํฐ์˜ ์ž„๊ณ„ ๊ธธ์ด๋ฅผ ์„ ํƒํ–ˆ์Šต๋‹ˆ๋‹ค.
๋”ฐ๋ผ์„œ ๋ชจ๋“  ๋ฐ˜๋ณต๋˜๋Š” ์‹œํ€€์Šค๋ฅผ ์ฐพ๋Š” ๊ฒƒ์€ ์ ‘๋ฏธ์‚ฌ ๋ฐฐ์—ด์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ๋๊นŒ์ง€ ์„ ํ˜•์ ์œผ๋กœ ์Šค์บ”ํ•˜๊ณ  ์ตœ์†Œ ์ž„๊ณ„ ๊ธธ์ด์˜ ๊ณตํ†ต ์ ‘๋‘์‚ฌ๋ฅผ ๊ณต์œ ํ•˜๋Š” ์‹œํ€€์Šค \(\mathcal{A}_{i},\mathcal{A}_{i+1}\)๋ฅผ ์ฐพ๋Š” ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค. ๋งŒ์กฑํ•˜๋Š” ์‹œํ€€์Šค๋Š” ๋ชจ๋‘ ๊ธฐ๋ก๋ฉ๋‹ˆ๋‹ค. ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ถ€๋„๋Ÿฌ์šธ ์ •๋„๋กœ ๋ณ‘๋ ฌ์ ์ด๋ฏ€๋กœ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์„ ํšจ์œจ์ ์œผ๋กœ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹คํ—˜(๋ถ€๋ก B)์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๋ชจ๋“  ์‹คํ—˜์— ๋Œ€ํ•ด 50๊ฐœ์˜ BPE ํ† ํฐ์˜ ์ž„๊ณ„ ๊ธธ์ด๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.
**Output** they also are easily trained, obedient, and intelligent.[[**Retriee**]=Yes]<p>German Shepherd โ€“ a simple difference of opinion, it is the fundamental issue of the breeds essential conformation and movement.... They are known for being easy to train and good for performing tasks and following instructions. They are especially well known for their police work, being used for tracking criminals, patrolling troubled areas and detecting and holding suspects.
**Output** ๊ทธ๋“ค์€ ๋˜ํ•œ ์‰ฝ๊ฒŒ ํ›ˆ๋ จ๋˜๊ณ , ์ˆœ์ข…์ ์ด๋ฉฐ, ์ง€๋Šฅ์ ์ž…๋‹ˆ๋‹ค.[[**Retriee**]=Yes]<p>์ €๋จผ ์…ฐํผ๋“œ โ€“ ๋‹จ์ˆœํ•œ ์˜๊ฒฌ ์ฐจ์ด๋กœ, ์ด๋Š” ํ’ˆ์ข…์˜ ๋ณธ์งˆ์ ์ธ ๊ตฌ์กฐ์™€ ์›€์ง์ž„์˜ ๊ทผ๋ณธ์ ์ธ ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค.... ๊ทธ๋“ค์€ ํ›ˆ๋ จ์ด ์‰ฝ๊ณ , ์ž‘์—… ์ˆ˜ํ–‰๊ณผ ์ง€์‹œ ๋”ฐ๋ฅด๊ธฐ์— ์ข‹๋‹ค๊ณ  ์•Œ๋ ค์ ธ ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ๊ฒฝ์ฐฐ ์—…๋ฌด์—์„œ ์œ ๋ช…ํ•˜๋ฉฐ, ๋ฒ”์ฃ„์ž ์ถ”์ , ๋ฌธ์ œ ์ง€์—ญ ์ˆœ์ฐฐ, ์šฉ์˜์ž ํƒ์ง€ ๋ฐ ์ฒดํฌ์— ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.
**์ถœ๋ ฅ** ๋˜ํ•œ ์‰ฝ๊ฒŒ ํ›ˆ๋ จ๋˜๊ณ  ์ˆœ์ข…์ ์ด๋ฉฐ ์ง€๋Šฅ์ ์ž…๋‹ˆ๋‹ค[[**Retriee**]=Yes]<p>๋…์ผ ์…ฐํผ๋“œ - ๊ฐ„๋‹จํ•œ ์˜๊ฒฌ ์ฐจ์ด, ํ’ˆ์ข…์˜ ๋ณธ์งˆ์ ์ธ ์ฒดํ˜•๊ณผ ์›€์ง์ž„์˜ ๊ทผ๋ณธ์ ์ธ ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค.... ๊ทธ๋“ค์€ ํ›ˆ๋ จํ•˜๊ธฐ ์‰ฝ๊ณ  ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๊ณ  ์ง€์‹œ๋ฅผ ๋”ฐ๋ฅด๋Š” ๋ฐ ์ข‹๊ธฐ๋กœ ์œ ๋ช…ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋“ค์€ ํŠนํžˆ ๋ฒ”์ฃ„์ž ์ถ”์ , ๋ฌธ์ œ ์ง€์—ญ ์ˆœ์ฐฐ, ์šฉ์˜์ž ํƒ์ง€ ๋ฐ ๊ตฌ๊ธˆ ๋“ฑ ๊ฒฝ์ฐฐ ์—…๋ฌด๋กœ ์ž˜ ์•Œ๋ ค์ ธ ์žˆ์Šต๋‹ˆ๋‹ค.
To examine the effectiveness and superiority of LLMs, a surge of tasks and benchmarks have been proposed for conducting empirical ability evaluation and analysis. In this section, we first introduce three types of basic ability evaluation of LLMs for language generation and understanding, then present several advanced ability evaluations with more complicated settings or goals, and finally discuss existing benchmarks, evaluation approaches, and empirical analysis.
LLMs์˜ ํšจ๊ณผ์„ฑ๊ณผ ์šฐ์ˆ˜์„ฑ์„ ๊ฒ€ํ† ํ•˜๊ธฐ ์œ„ํ•ด, ๊ฒฝํ—˜์  ๋Šฅ๋ ฅ ํ‰๊ฐ€์™€ ๋ถ„์„์„ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•œ ๋‹ค์–‘ํ•œ ์ž‘์—…๊ณผ ๋ฒค์น˜๋งˆํฌ๊ฐ€ ์ œ์•ˆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ๋จผ์ € ์–ธ์–ด ์ƒ์„ฑ ๋ฐ ์ดํ•ด๋ฅผ ์œ„ํ•œ LLM์˜ ์„ธ ๊ฐ€์ง€ ๊ธฐ๋ณธ ๋Šฅ๋ ฅ ํ‰๊ฐ€๋ฅผ ์†Œ๊ฐœํ•˜๊ณ , ๋” ๋ณต์žกํ•œ ์„ค์ •์ด๋‚˜ ๋ชฉํ‘œ๋ฅผ ๊ฐ€์ง„ ์—ฌ๋Ÿฌ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ ํ‰๊ฐ€๋ฅผ ์ œ์‹œํ•œ ํ›„, ๊ธฐ์กด์˜ ๋ฒค์น˜๋งˆํฌ, ํ‰๊ฐ€ ์ ‘๊ทผ๋ฒ• ๋ฐ ๊ฒฝํ—˜์  ๋ถ„์„์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค.
LLM์˜ ํšจ๊ณผ์™€ ์šฐ์›”์„ฑ์„ ์กฐ์‚ฌํ•˜๊ธฐ ์œ„ํ•ด, ๊ฒฝํ—˜์  ๋Šฅ๋ ฅ ํ‰๊ฐ€ ๋ฐ ๋ถ„์„์„ ์ˆ˜ํ–‰ํ•˜๊ธฐ ์œ„ํ•œ ์ž‘์—…๊ณผ ๋ฒค์น˜๋งˆํฌ๊ฐ€ ๊ธ‰์ฆํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ๋จผ์ € ์–ธ์–ด ์ƒ์„ฑ ๋ฐ ์ดํ•ด๋ฅผ ์œ„ํ•œ LLM์˜ ์„ธ ๊ฐ€์ง€ ๊ธฐ๋ณธ ๋Šฅ๋ ฅ ํ‰๊ฐ€ ์œ ํ˜•์„ ์†Œ๊ฐœํ•œ ๋‹ค์Œ, ๋” ๋ณต์žกํ•œ ์„ค์ •์ด๋‚˜ ๋ชฉํ‘œ๋ฅผ ๊ฐ€์ง„ ๋ช‡ ๊ฐ€์ง€ ๊ณ ๊ธ‰ ๋Šฅ๋ ฅ ํ‰๊ฐ€๋ฅผ ์ œ์‹œํ•˜๊ณ , ๋งˆ์ง€๋ง‰์œผ๋กœ ๊ธฐ์กด ๋ฒค์น˜๋งˆํฌ, ํ‰๊ฐ€ ์ ‘๊ทผ ๋ฐฉ์‹ ๋ฐ ๊ฒฝํ—˜์  ๋ถ„์„์— ๋Œ€ํ•ด ๋…ผ์˜ํ•ฉ๋‹ˆ๋‹ค.
In contrast, for multilingual tokenizers, we observe significant fertility improvements with increasing vocabulary sizes. Due to the larger thesaurus induced by the additional languages, the tokenizer requires a larger vocabulary to allow a model to perform convincingly on all languages. Therefore, only within the non-convergence vocabulary range, we achieve a strong, negative correlation between fertility and downstream performance with varying vocabulary sizes.
๋ฐ˜๋ฉด, ๋‹ค๊ตญ์–ด ํ† ํฌ๋‚˜์ด์ €์˜ ๊ฒฝ์šฐ, ์–ดํœ˜ ํฌ๊ธฐ๊ฐ€ ์ฆ๊ฐ€ํ•จ์— ๋”ฐ๋ผ ์œ ์˜๋ฏธํ•œ fertility ๊ฐœ์„ ์ด ๊ด€์ฐฐ๋ฉ๋‹ˆ๋‹ค. ์ถ”๊ฐ€๋œ ์–ธ์–ด๋“ค๋กœ ์ธํ•ด ์œ ๋„๋œ ๋” ํฐ ์‚ฌ์ „ ๋•Œ๋ฌธ์—, ํ† ํฌ๋‚˜์ด์ €๋Š” ๋ชจ๋“  ์–ธ์–ด์—์„œ ๋ชจ๋ธ์ด ์„ค๋“๋ ฅ ์žˆ๊ฒŒ ์ž‘๋™ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋” ํฐ ์–ดํœ˜๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ, ๋น„์ˆ˜๋ ด ์–ดํœ˜ ๋ฒ”์œ„ ๋‚ด์—์„œ๋งŒ ์–ดํœ˜ ํฌ๊ธฐ์— ๋”ฐ๋ผ fertility์™€ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์„ฑ๋Šฅ ๊ฐ„์— ๊ฐ•ํ•œ ์Œ์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
์ด์™€๋Š” ๋Œ€์กฐ์ ์œผ๋กœ ๋‹ค๊ตญ์–ด ํ† ํฐํ™”๊ธฐ์— ์žˆ์–ด์„œ๋Š” ์–ดํœ˜ ํฌ๊ธฐ๊ฐ€ ์ปค์งˆ์ˆ˜๋ก ์ƒ๋‹นํ•œ ๋น„์˜ฅ๋„ ํ–ฅ์ƒ์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ๋‹ค. ์ถ”๊ฐ€ ์–ธ์–ด์— ์˜ํ•ด ์œ ๋„๋œ ๋” ํฐ ๋™์˜์–ด ์‚ฌ์ „์œผ๋กœ ์ธํ•ด ํ† ํฐํ™”๊ธฐ๋Š” ๋ชจ๋ธ์ด ๋ชจ๋“  ์–ธ์–ด์—์„œ ์„ค๋“๋ ฅ ์žˆ๊ฒŒ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋” ํฐ ์–ดํœ˜๋ฅผ ํ•„์š”๋กœ ํ•œ๋‹ค. ๋”ฐ๋ผ์„œ ์ˆ˜๋ ดํ•˜์ง€ ์•Š๋Š” ์–ดํœ˜ ๋ฒ”์œ„ ๋‚ด์—์„œ๋งŒ ๋‹ค์–‘ํ•œ ์–ดํœ˜ ํฌ๊ธฐ๋กœ ๋น„์˜ฅ๋„์™€ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์„ฑ๋Šฅ ์‚ฌ์ด์˜ ๊ฐ•ํ•œ ์Œ์˜ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋‹ค.
**Safety Prompts Evaluation**Sun et al. (2023) used InstructGPT (Ouyang et al., 2022) with a verbalizer to assess text safety. In our method, we simplify things. Instead of using a verbalizer to understand the output, we use the function call feature of the OpenAI GPT 3.5 8. This makes the results clearer and easier to interpret. For those who are interested, we have detailed the specifics of this function call and the related assessment prompts in Appendix A.2.
**Safety Prompts Evaluation** Sun et al. (2023)์€ ํ…์ŠคํŠธ ์•ˆ์ „์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด InstructGPT (Ouyang et al., 2022)๋ฅผ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•์—์„œ๋Š” ์ด๋ฅผ ๋‹จ์ˆœํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ถœ๋ ฅ ๊ฒฐ๊ณผ๋ฅผ ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด verbalizer๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋Œ€์‹ , OpenAI GPT 3.5์˜ ํ•จ์ˆ˜ ํ˜ธ์ถœ ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๊ฒฐ๊ณผ๋ฅผ ๋” ๋ช…ํ™•ํ•˜๊ณ  ํ•ด์„ํ•˜๊ธฐ ์‰ฝ๊ฒŒ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ๊ด€์‹ฌ ์žˆ๋Š” ๋ถ„๋“ค์„ ์œ„ํ•ด, ์ด ํ•จ์ˆ˜ ํ˜ธ์ถœ๊ณผ ๊ด€๋ จ๋œ ํ‰๊ฐ€ ํ”„๋กฌํ”„ํŠธ์˜ ์„ธ๋ถ€ ์‚ฌํ•ญ์„ ๋ถ€๋ก A.2์— ์ž์„ธํžˆ ์„ค๋ช…ํ•ด ๋‘์—ˆ์Šต๋‹ˆ๋‹ค.
**์•ˆ์ „ ํ”„๋กฌํ”„ํŠธ ํ‰๊ฐ€**Sun et al.(2023)์€ ํ…์ŠคํŠธ ์•ˆ์ „์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ๊ตฌ๋‘ํ™” ๊ธฐ๋Šฅ์ด ์žˆ๋Š” InstructGPT(Ouyang et al., 2022)๋ฅผ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ ๋ฐฉ๋ฒ•์€ ์ด๋ฅผ ๋‹จ์ˆœํ™”ํ•ฉ๋‹ˆ๋‹ค. ์ถœ๋ ฅ์„ ์ดํ•ดํ•˜๊ธฐ ์œ„ํ•ด ๊ตฌ๋‘ํ™” ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•˜๋Š” ๋Œ€์‹  OpenAI GPT 3.5 8์˜ ํ•จ์ˆ˜ ํ˜ธ์ถœ ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋ ‡๊ฒŒ ํ•˜๋ฉด ๊ฒฐ๊ณผ๊ฐ€ ๋” ๋ช…ํ™•ํ•˜๊ณ  ํ•ด์„ํ•˜๊ธฐ ์‰ฌ์›Œ์ง‘๋‹ˆ๋‹ค. ๊ด€์‹ฌ์ด ์žˆ๋Š” ๋ถ„๋“ค์„ ์œ„ํ•ด ์ด ํ•จ์ˆ˜ ํ˜ธ์ถœ์˜ ์„ธ๋ถ€ ์‚ฌํ•ญ๊ณผ ๊ด€๋ จ ํ‰๊ฐ€ ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋ถ€๋ก A.2์— ์ž์„ธํžˆ ์„ค๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.
We can observe the following: 1) ETS-DACP trained on 10% outperforms DACP with 100% of the data; 2) ETS-DACP has the best performance among all three counterparts and is on par with a combination of three metrics - ETS-DACP-com; 3) ETA-DACP-ent trained on 10% corpus is a close second despite not having any access to task data, handily surpassing DACP trained on 100% of the data; and 4) Efficient DACP methods with hard sampling outperform ones with soft sampling.
๋‹ค์Œ๊ณผ ๊ฐ™์€ ์ ์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค: 1) 10%์˜ ๋ฐ์ดํ„ฐ๋กœ ํ›ˆ๋ จ๋œ ETS-DACP๊ฐ€ 100%์˜ ๋ฐ์ดํ„ฐ๋กœ ํ›ˆ๋ จ๋œ DACP๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค; 2) ETS-DACP๋Š” ์„ธ ๊ฐ€์ง€ ๋ชจ๋“  ๋Œ€์‘ ๋ชจ๋ธ ์ค‘์—์„œ ๊ฐ€์žฅ ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ์„ ๋ณด์ด๋ฉฐ, ์„ธ ๊ฐ€์ง€ ๋ฉ”ํŠธ๋ฆญ์˜ ์กฐํ•ฉ์ธ ETS-DACP-com๊ณผ ๋™๋“ฑํ•œ ์ˆ˜์ค€์ž…๋‹ˆ๋‹ค; 3) 10%์˜ ์ฝ”ํผ์Šค๋กœ ํ›ˆ๋ จ๋œ ETA-DACP-ent๋Š” ์ž‘์—… ๋ฐ์ดํ„ฐ์— ์ ‘๊ทผํ•˜์ง€ ๋ชปํ–ˆ์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  100%์˜ ๋ฐ์ดํ„ฐ๋กœ ํ›ˆ๋ จ๋œ DACP๋ฅผ ์‰ฝ๊ฒŒ ๋Šฅ๊ฐ€ํ•˜๋ฉฐ ๊ทผ์†Œํ•œ ์ฐจ์ด๋กœ 2์œ„๋ฅผ ์ฐจ์ง€ํ•ฉ๋‹ˆ๋‹ค; ๊ทธ๋ฆฌ๊ณ  4) ํ•˜๋“œ ์ƒ˜ํ”Œ๋ง์„ ์‚ฌ์šฉํ•˜๋Š” ํšจ์œจ์ ์ธ DACP ๋ฐฉ๋ฒ•์ด ์†Œํ”„ํŠธ ์ƒ˜ํ”Œ๋ง์„ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•๋ณด๋‹ค ๋” ์šฐ์ˆ˜ํ•ฉ๋‹ˆ๋‹ค.
๋‹ค์Œ๊ณผ ๊ฐ™์€ ์‚ฌ์‹ค์„ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค: 1) 10%๋กœ ํ•™์Šต๋œ ETS-DACP๊ฐ€ 100%์˜ ๋ฐ์ดํ„ฐ๋กœ DACP๋ฅผ ๋Šฅ๊ฐ€ํ•ฉ๋‹ˆ๋‹ค; 2) ETS-DACP๊ฐ€ ์„ธ ๊ฐ€์ง€ ๋Œ€์‘ ๋ฐฉ์‹ ์ค‘ ๊ฐ€์žฅ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ณด์ด๋ฉฐ ์„ธ ๊ฐ€์ง€ ์ง€ํ‘œ์˜ ์กฐํ•ฉ์ธ ETS-DACP-com๊ณผ ๋™๋“ฑํ•ฉ๋‹ˆ๋‹ค; 3) 10% ๋ง๋ญ‰์น˜๋กœ ํ•™์Šต๋œ ETA-DACP-ent๋Š” ๊ณผ์ œ ๋ฐ์ดํ„ฐ์— ์•ก์„ธ์Šคํ•  ์ˆ˜ ์—†์Œ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  100% ๋ง๋ญ‰์น˜๋กœ ํ•™์Šต๋œ DACP๋ฅผ ๋Šฅ๊ฐ€ํ•˜๋Š” 2์œ„๋ฅผ ์ฐจ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค; 4) ํ•˜๋“œ ์ƒ˜ํ”Œ๋ง์„ ์‚ฌ์šฉํ•œ ํšจ์œจ์ ์ธ DACP ๋ฐฉ๋ฒ•์ด ์†Œํ”„ํŠธ ์ƒ˜ํ”Œ๋ง์„ ์‚ฌ์šฉํ•œ ๋ฐฉ๋ฒ•๋ณด๋‹ค ์„ฑ๋Šฅ์ด ์šฐ์ˆ˜ํ•ฉ๋‹ˆ๋‹ค.
where the exponent \(c_{q}\in(0,\infty)\) depends on \(\beta\) and the algorithm \(q\). For example, when \(q\) encodes the counting measure \(q_{T}(i):=n_{T}(i)/T\) (reminiscent of SGD), it was shown that \(c_{q}=(1-1/\beta)/2\in(0,1/2)\). Another algorithm \(q_{T}(i):=1[n_{T}(i)\geq 1]/\sum_{\ell}1[n_{T}(\ell)\geq 1]\) (reminiscent of ADAM ) was proposed which attains a optimal error rate (over all algorithms based on random embeddings) with \(c_{q}=\beta-1\).
์ง€์ˆ˜ \(c_{q}\in(0,\infty)\)๋Š” \(\beta\)์™€ ์•Œ๊ณ ๋ฆฌ์ฆ˜ \(q\)์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง‘๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, \(q\)๊ฐ€ ์นด์šดํŒ… ์ธก์ • \(q_{T}(i):=n_{T}(i)/T\) (SGD๋ฅผ ์—ฐ์ƒ์‹œํ‚ค๋Š”)๋กœ ์ธ์ฝ”๋”ฉ๋  ๋•Œ, \(c_{q}=(1-1/\beta)/2\in(0,1/2)\)์ž„์ด ๋ฐํ˜€์กŒ์Šต๋‹ˆ๋‹ค. ๋˜ ๋‹ค๋ฅธ ์•Œ๊ณ ๋ฆฌ์ฆ˜ \(q_{T}(i):=1[n_{T}(i)\geq 1]/\sum_{\ell}1[n_{T}(\ell)\geq 1]\) (ADAM์„ ์—ฐ์ƒ์‹œํ‚ค๋Š”)์ด ์ œ์•ˆ๋˜์—ˆ์œผ๋ฉฐ, ์ด๋Š” \(c_{q}=\beta-1\)๋กœ ๋žœ๋ค ์ž„๋ฒ ๋”ฉ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ๋ชจ๋“  ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ค‘ ์ตœ์ ์˜ ์˜ค๋ฅ˜์œจ์„ ๋‹ฌ์„ฑํ•ฉ๋‹ˆ๋‹ค.
์—ฌ๊ธฐ์„œ ์ง€์ˆ˜ \(c_{q}\in(0,\infty)\)๋Š” \(\beta\)์™€ ์•Œ๊ณ ๋ฆฌ์ฆ˜ \(q\)์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์ง‘๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, \(q\)๊ฐ€ ์นด์šดํŒ… ์ธก์ •๊ฐ’ \(q_{T}(i):=n_{T}(i)/T\)๋ฅผ ์ธ์ฝ”๋”ฉํ•˜๋Š” ๊ฒฝ์šฐ(SGD๋ฅผ ์—ฐ์ƒ์‹œํ‚ด), \(c_{q}=(1-1/\beta)/2\in(0,1/2)\)์ž„์ด ๋ฐํ˜€์กŒ์Šต๋‹ˆ๋‹ค. ๋˜ ๋‹ค๋ฅธ ์•Œ๊ณ ๋ฆฌ์ฆ˜ \(q_{T}(i):=1[n_{T}(i)\geq 1]/\sum_{\ell}1[n_{T}(\ell)\geq 1]\) (ADAM ์„ ์—ฐ์ƒ์‹œํ‚ด)์ด ์ œ์•ˆ๋˜์—ˆ๋Š”๋ฐ, ์ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ ๋ชจ๋“  ๋žœ๋ค ์ž„๋ฒ ๋”ฉ ๊ธฐ๋ฐ˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์—์„œ ์ตœ์ ์˜ ์˜ค๋ฅ˜์œจ์„ ๋‹ฌ์„ฑํ•˜๋ฉฐ \(c_{q}=\beta-1\)์„ ๋‹ฌ์„ฑํ•ฉ๋‹ˆ๋‹ค.
Tab. 5 presents the performance results of Yi-34B-Chat in the three third-party evaluations we consider, with the cutoff date for the results being December 21, 2023. The data demonstrates that, although there is still a gap compared to GPT-4, our model exhibits proficient bilingual (Chinese and English) dialogue capabilities and aligns well with user preferences. Additional comparative results of various models are accessible for review on the official website.
ํ‘œ 5๋Š” ์šฐ๋ฆฌ๊ฐ€ ๊ณ ๋ คํ•œ ์„ธ ๊ฐ€์ง€ ์ œ3์ž ํ‰๊ฐ€์—์„œ Yi-34B-Chat์˜ ์„ฑ๋Šฅ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ, ๊ฒฐ๊ณผ์˜ ๊ธฐ์ค€ ๋‚ ์งœ๋Š” 2023๋…„ 12์›” 21์ผ์ž…๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ๋Š” GPT-4์™€ ๋น„๊ตํ–ˆ์„ ๋•Œ ์—ฌ์ „ํžˆ ๊ฒฉ์ฐจ๊ฐ€ ์žˆ์ง€๋งŒ, ์šฐ๋ฆฌ ๋ชจ๋ธ์ด ๋›ฐ์–ด๋‚œ ์ด์ค‘ ์–ธ์–ด(์ค‘๊ตญ์–ด์™€ ์˜์–ด) ๋Œ€ํ™” ๋Šฅ๋ ฅ์„ ๋ณด์œ ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ ์‚ฌ์šฉ์ž ์„ ํ˜ธ๋„์™€ ์ž˜ ๋งž๋Š”๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ๋ชจ๋ธ์˜ ์ถ”๊ฐ€ ๋น„๊ต ๊ฒฐ๊ณผ๋Š” ๊ณต์‹ ์›น์‚ฌ์ดํŠธ์—์„œ ๊ฒ€ํ† ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
ํ‘œ 5๋Š” 2023๋…„ 12์›” 21์ผ์„ ๊ธฐ์ค€์œผ๋กœ ํ•œ 3์ฐจ ํ‰๊ฐ€์—์„œ Yi-34B-Chat์˜ ์„ฑ๋Šฅ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด ๋ฐ์ดํ„ฐ๋Š” GPT-4์— ๋น„ํ•ด ์—ฌ์ „ํžˆ ๊ฒฉ์ฐจ๊ฐ€ ์žˆ์ง€๋งŒ, ์šฐ๋ฆฌ ๋ชจ๋ธ์€ ๋Šฅ์ˆ™ํ•œ ์ด์ค‘ ์–ธ์–ด(์ค‘๊ตญ์–ด ๋ฐ ์˜์–ด) ๋Œ€ํ™” ๊ธฐ๋Šฅ์„ ์ œ๊ณตํ•˜๋ฉฐ ์‚ฌ์šฉ์ž ์„ ํ˜ธ๋„์— ์ž˜ ๋ถ€ํ•ฉํ•œ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๋‹ค์–‘ํ•œ ๋ชจ๋ธ์˜ ์ถ”๊ฐ€ ๋น„๊ต ๊ฒฐ๊ณผ๋Š” ๊ณต์‹ ์›น์‚ฌ์ดํŠธ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Therefore, comparing \(\mathsf{bioS}\) and \(\mathsf{bioS}^{\mathsf{simple}}\), it's more advantageous to rewrite the data 1000 times (in this ideal setting), training each for one pass (as done in the \(\mathsf{bioS}\) data), rather than training the same data for 1000 passes (as done in the \(\mathsf{bioS}^{\mathsf{simple}}\) data). This is because, without data diversity, the model wastes capacity memorizing sentence structures, resulting in a capacity loss.
๋”ฐ๋ผ์„œ \(\mathsf{bioS}\)์™€ \(\mathsf{bioS}^{\mathsf{simple}}\)๋ฅผ ๋น„๊ตํ•  ๋•Œ, ๋™์ผํ•œ ๋ฐ์ดํ„ฐ๋ฅผ 1000๋ฒˆ ๋ฐ˜๋ณตํ•ด์„œ ํ•™์Šตํ•˜๋Š” ๊ฒƒ(\(\mathsf{bioS}^{\mathsf{simple}}\) ๋ฐ์ดํ„ฐ์—์„œ ์ˆ˜ํ–‰๋œ ๊ฒƒ์ฒ˜๋Ÿผ)๋ณด๋‹ค ๋ฐ์ดํ„ฐ๋ฅผ 1000๋ฒˆ ๋‹ค์‹œ ์ž‘์„ฑํ•˜๊ณ  ๊ฐ ๋ฐ์ดํ„ฐ๋ฅผ ํ•œ ๋ฒˆ์”ฉ ํ•™์Šตํ•˜๋Š” ๊ฒƒ(\(\mathsf{bioS}\) ๋ฐ์ดํ„ฐ์—์„œ ์ˆ˜ํ–‰๋œ ๊ฒƒ์ฒ˜๋Ÿผ)์ด ๋” ์œ ๋ฆฌํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋ฐ์ดํ„ฐ ๋‹ค์–‘์„ฑ์ด ์—†์œผ๋ฉด ๋ชจ๋ธ์ด ๋ฌธ์žฅ ๊ตฌ์กฐ๋ฅผ ์•”๊ธฐํ•˜๋Š” ๋ฐ ์šฉ๋Ÿ‰์„ ๋‚ญ๋น„ํ•˜๊ฒŒ ๋˜์–ด ์šฉ๋Ÿ‰ ์†์‹ค์ด ๋ฐœ์ƒํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.
๋”ฐ๋ผ์„œ \(\mathsf{bioS}\)์™€ \(\mathsf{bioS}^{\mathsf{simple}}\)๋ฅผ ๋น„๊ตํ•˜๋ฉด, ์ด ๊ฒฝ์šฐ(์ด ์ด์ƒ์ ์ธ ์„ค์ •์—์„œ) 1000๋ฒˆ ํ›ˆ๋ จํ•˜๋Š” ๊ฒƒ๋ณด๋‹ค 1000๋ฒˆ์”ฉ ํ›ˆ๋ จํ•˜๋Š” ๊ฒƒ์ด ๋” ์œ ๋ฆฌํ•ฉ๋‹ˆ๋‹ค(\(\mathsf{bioS}^{\mathsf{simple}}\) ๋ฐ์ดํ„ฐ์—์„œ ์ˆ˜ํ–‰ํ•œ ๊ฒƒ์ฒ˜๋Ÿผ). ๋ฐ์ดํ„ฐ ๋‹ค์–‘์„ฑ์ด ์—†์œผ๋ฉด ๋ชจ๋ธ์ด ๋ฌธ์žฅ ๊ตฌ์กฐ๋ฅผ ์•”๊ธฐํ•˜๋Š” ๋ฐ ์šฉ๋Ÿ‰์„ ๋‚ญ๋น„ํ•˜์—ฌ ์šฉ๋Ÿ‰ ์†์‹ค์ด ๋ฐœ์ƒํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.
Since its introduction by Lewis et al. (Lewis et al., 2020) in 2020, RAG technology has undergone significant advancements, particularly influenced by ChatGPT's success. However, there is a noticeable gap in the literature regarding a thorough analysis of RAG's mechanisms and the progress made by subsequent studies. Furthermore, the field is characterized by diverse research focuses and the use of ambiguous terminology for similar methods, leading to confusion.
Lewis ๋“ฑ(2020)์— ์˜ํ•ด 2020๋…„์— ๋„์ž…๋œ ์ดํ›„, RAG ๊ธฐ์ˆ ์€ ํŠนํžˆ ChatGPT์˜ ์„ฑ๊ณต์— ์˜ํ•ด ํฌ๊ฒŒ ๋ฐœ์ „ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ RAG์˜ ๋ฉ”์ปค๋‹ˆ์ฆ˜๊ณผ ํ›„์† ์—ฐ๊ตฌ์—์„œ ์ด๋ฃจ์–ด์ง„ ์ง„์ „์— ๋Œ€ํ•œ ์ฒ ์ €ํ•œ ๋ถ„์„์— ๊ด€ํ•œ ๋ฌธํ—Œ์€ ๋ˆˆ์— ๋„๊ฒŒ ๋ถ€์กฑํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด ๋ถ„์•ผ๋Š” ๋‹ค์–‘ํ•œ ์—ฐ๊ตฌ ์ดˆ์ ๊ณผ ์œ ์‚ฌํ•œ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ๋ชจํ˜ธํ•œ ์šฉ์–ด ์‚ฌ์šฉ์œผ๋กœ ์ธํ•ด ํ˜ผ๋ž€์„ ์ดˆ๋ž˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
2020๋…„ Lewis ๋“ฑ(Lewis et al., 2020)์— ์˜ํ•ด ์†Œ๊ฐœ๋œ ์ด๋ž˜, RAG ๊ธฐ์ˆ ์€ ํŠนํžˆ ChatGPT์˜ ์„ฑ๊ณต์— ์˜ํ–ฅ์„ ๋ฐ›์•„ ์ƒ๋‹นํ•œ ๋ฐœ์ „์„ ์ด๋ฃจ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ RAG์˜ ๋ฉ”์ปค๋‹ˆ์ฆ˜์— ๋Œ€ํ•œ ์ฒ ์ €ํ•œ ๋ถ„์„๊ณผ ํ›„์† ์—ฐ๊ตฌ์—์„œ ์ด๋ฃจ์–ด์ง„ ์ง„์ „์— ๋Œ€ํ•œ ๋ฌธํ—Œ์—๋Š” ๋ˆˆ์— ๋„๋Š” ๊ฒฉ์ฐจ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด ๋ถ„์•ผ๋Š” ์—ฐ๊ตฌ ์ดˆ์ ์ด ๋‹ค์–‘ํ•˜๊ณ  ์œ ์‚ฌํ•œ ๋ฐฉ๋ฒ•์— ๋Œ€ํ•ด ๋ชจํ˜ธํ•œ ์šฉ์–ด๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํ˜ผ๋ž€์„ ์ดˆ๋ž˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
With NearDup, we found that the web-scrape datasets contain between 3.04% (on C4) to 13.63% (on RealNews) near duplicates (Table 2). Nearduplicate text is much less common in Wiki-40B, forming only 0.39% of the train set.2 In C4, the majority (1.8M) of near-duplicate clusters consisted of just a single pair of examples that matched against each other, but there were 280 clusters with over 5,000 examples in them (Figure 1), including one cluster of size 250,933.
NearDup์„ ์‚ฌ์šฉํ•˜์—ฌ ์›น ์Šคํฌ๋ž˜ํ•‘ ๋ฐ์ดํ„ฐ์…‹์—์„œ 3.04% (C4)์—์„œ 13.63% (RealNews) ์‚ฌ์ด์˜ ์œ ์‚ฌ ์ค‘๋ณต์ด ํฌํ•จ๋˜์–ด ์žˆ์Œ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค (ํ‘œ 2). ์œ ์‚ฌ ์ค‘๋ณต ํ…์ŠคํŠธ๋Š” Wiki-40B์—์„œ๋Š” ํ›จ์”ฌ ๋œ ์ผ๋ฐ˜์ ์ด๋ฉฐ, ํ•™์Šต ์„ธํŠธ์˜ 0.39%๋งŒ์„ ์ฐจ์ง€ํ•ฉ๋‹ˆ๋‹ค. C4์—์„œ๋Š” ๋Œ€๋ถ€๋ถ„์˜ ์œ ์‚ฌ ์ค‘๋ณต ํด๋Ÿฌ์Šคํ„ฐ(1.8M)๊ฐ€ ์„œ๋กœ ์ผ์น˜ํ•˜๋Š” ๋‹จ์ผ ์Œ์˜ ์˜ˆ์ œ๋กœ ๊ตฌ์„ฑ๋˜์—ˆ์ง€๋งŒ, 5,000๊ฐœ ์ด์ƒ์˜ ์˜ˆ์ œ๋ฅผ ํฌํ•จํ•˜๋Š” 280๊ฐœ์˜ ํด๋Ÿฌ์Šคํ„ฐ๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค (๊ทธ๋ฆผ 1), ์—ฌ๊ธฐ์—๋Š” 250,933๊ฐœ์˜ ์˜ˆ์ œ๋ฅผ ํฌํ•จํ•˜๋Š” ํ•˜๋‚˜์˜ ํด๋Ÿฌ์Šคํ„ฐ๋„ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค.
NearDup์„ ์‚ฌ์šฉํ•˜๋ฉด ์›น ์Šคํฌ๋žฉ ๋ฐ์ดํ„ฐ ์„ธํŠธ์— 3.04%(C4)์—์„œ 13.63%(RealNews) ์‚ฌ์ด์˜ ๊ทผ์‚ฌ ์ค‘๋ณต์ด ํฌํ•จ๋˜์–ด ์žˆ์Œ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค(ํ‘œ 2). ์œ„ํ‚ค-40B์—๋Š” ๊ทผ์‚ฌ ์ค‘๋ณต ํ…์ŠคํŠธ๊ฐ€ ํ›จ์”ฌ ๋œ ์ผ๋ฐ˜์ ์ด๋ฉฐ, ํ•™์Šต ์„ธํŠธ์˜ 0.39%์— ๋ถˆ๊ณผํ•ฉ๋‹ˆ๋‹ค.2 C4์—์„œ๋Š” ๋Œ€๋ถ€๋ถ„(1.8M)์˜ ๊ทผ์‚ฌ ์ค‘๋ณต ํด๋Ÿฌ์Šคํ„ฐ๊ฐ€ ์„œ๋กœ ์ผ์น˜ํ•˜๋Š” ํ•œ ์Œ์˜ ์˜ˆ์‹œ๋กœ ๊ตฌ์„ฑ๋˜์—ˆ์ง€๋งŒ, 250,933๊ฐœ ํฌ๊ธฐ์˜ ํด๋Ÿฌ์Šคํ„ฐ๋ฅผ ํฌํ•จํ•˜์—ฌ 5,000๊ฐœ ์ด์ƒ์˜ ์˜ˆ์‹œ๊ฐ€ ํฌํ•จ๋œ 280๊ฐœ์˜ ํด๋Ÿฌ์Šคํ„ฐ๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค(๊ทธ๋ฆผ 1).
Although large language models (LLMs) demonstrate remarkable capabilities in variety of applications (Zhao et al., 2023), such as language generation, understanding, and reasoning, they struggle to provide accurate answers when faced with complicated tasks. To improve the performance of LLMs, some of recent studies focus on ensemble methods (Wang et al., 2023; Wan et al., 2024) and multiple LLM-Agents collaboration frameworks (Du et al., 2023; Wu et al., 2023).
๋น„๋ก ๋Œ€ํ˜• ์–ธ์–ด ๋ชจ๋ธ(LLMs)์ด ์–ธ์–ด ์ƒ์„ฑ, ์ดํ•ด ๋ฐ ์ถ”๋ก ๊ณผ ๊ฐ™์€ ๋‹ค์–‘ํ•œ ์‘์šฉ ๋ถ„์•ผ์—์„œ ๋†€๋ผ์šด ๋Šฅ๋ ฅ์„ ๋ณด์—ฌ์ฃผ์ง€๋งŒ(Zhao et al., 2023), ๋ณต์žกํ•œ ์ž‘์—…์— ์ง๋ฉดํ–ˆ์„ ๋•Œ ์ •ํ™•ํ•œ ๋‹ต๋ณ€์„ ์ œ๊ณตํ•˜๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช์Šต๋‹ˆ๋‹ค. LLM์˜ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๊ธฐ ์œ„ํ•ด, ์ตœ๊ทผ์˜ ์ผ๋ถ€ ์—ฐ๊ตฌ๋“ค์€ ์•™์ƒ๋ธ” ๋ฐฉ๋ฒ•(Wang et al., 2023; Wan et al., 2024)๊ณผ ๋‹ค์ค‘ LLM-์—์ด์ „ํŠธ ํ˜‘์—… ํ”„๋ ˆ์ž„์›Œํฌ(Du et al., 2023; Wu et al., 2023)์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM)์€ ์–ธ์–ด ์ƒ์„ฑ, ์ดํ•ด, ์ถ”๋ก  ๋“ฑ ๋‹ค์–‘ํ•œ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์—์„œ ๋†€๋ผ์šด ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜์ง€๋งŒ(Zhao et al., 2023), ๋ณต์žกํ•œ ์ž‘์—…์— ์ง๋ฉดํ•˜๋ฉด ์ •ํ™•ํ•œ ๋‹ต๋ณ€์„ ์ œ๊ณตํ•˜๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช์Šต๋‹ˆ๋‹ค. LLM์˜ ์„ฑ๋Šฅ์„ ๊ฐœ์„ ํ•˜๊ธฐ ์œ„ํ•ด ์ตœ๊ทผ ์ผ๋ถ€ ์—ฐ๊ตฌ์—์„œ๋Š” ์•™์ƒ๋ธ” ๋ฐฉ๋ฒ•(Wang et al., 2023; Wan et al., 2024)๊ณผ ์—ฌ๋Ÿฌ LLM-์—์ด์ „ํŠธ ํ˜‘์—… ํ”„๋ ˆ์ž„์›Œํฌ(Du et al., 2023; Wu et al., 2023)์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
In this work, we explore the following: 1) Is domain-adaptive continual pre-training helpful in building domain-specific LLMs?; 2) Can we employ data selection strategies for a more effective domain-adaptive continual pre-training?; and 3) Does domain-adaptive continual pre-training hurt LLM's open-domain capabilities? We answer these questions in the confines of finance domain by training a continually pre-trained model, FinPythia, built on top of Pythia [4].
์ด ์—ฐ๊ตฌ์—์„œ ์šฐ๋ฆฌ๋Š” ๋‹ค์Œ์„ ํƒ๊ตฌํ•ฉ๋‹ˆ๋‹ค: 1) ๋„๋ฉ”์ธ ์ ์‘ํ˜• ์ง€์†์  ์‚ฌ์ „ ํ›ˆ๋ จ์ด ๋„๋ฉ”์ธ ํŠนํ™” LLM์„ ๊ตฌ์ถ•ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋˜๋Š”๊ฐ€?; 2) ๋” ํšจ๊ณผ์ ์ธ ๋„๋ฉ”์ธ ์ ์‘ํ˜• ์ง€์†์  ์‚ฌ์ „ ํ›ˆ๋ จ์„ ์œ„ํ•ด ๋ฐ์ดํ„ฐ ์„ ํƒ ์ „๋žต์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š”๊ฐ€?; 3) ๋„๋ฉ”์ธ ์ ์‘ํ˜• ์ง€์†์  ์‚ฌ์ „ ํ›ˆ๋ จ์ด LLM์˜ ์˜คํ”ˆ ๋„๋ฉ”์ธ ๊ธฐ๋Šฅ์— ํ•ด๋ฅผ ๋ผ์น˜๋Š”๊ฐ€? ์šฐ๋ฆฌ๋Š” Pythia [4]๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตฌ์ถ•๋œ ์ง€์†์ ์œผ๋กœ ์‚ฌ์ „ ํ›ˆ๋ จ๋œ ๋ชจ๋ธ์ธ FinPythia๋ฅผ ํ›ˆ๋ จํ•˜์—ฌ ๊ธˆ์œต ๋„๋ฉ”์ธ์˜ ๋ฒ”์œ„ ๋‚ด์—์„œ ์ด๋Ÿฌํ•œ ์งˆ๋ฌธ์— ๋‹ตํ•ฉ๋‹ˆ๋‹ค.
์ด ์—ฐ๊ตฌ์—์„œ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์‚ฌํ•ญ์„ ์‚ดํŽด๋ด…๋‹ˆ๋‹ค: 1) ๋„๋ฉ”์ธ ์ ์‘ ์ง€์†์  ์‚ฌ์ „ ํ•™์Šต์ด ๋„๋ฉ”์ธ๋ณ„ LLM ๊ตฌ์ถ•์— ๋„์›€์ด ๋˜๋Š”๊ฐ€?; 2) ๋ณด๋‹ค ํšจ๊ณผ์ ์ธ ๋„๋ฉ”์ธ ์ ์‘ ์ง€์†์  ์‚ฌ์ „ ํ•™์Šต์„ ์œ„ํ•ด ๋ฐ์ดํ„ฐ ์„ ํƒ ์ „๋žต์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š”๊ฐ€?; 3) ๋„๋ฉ”์ธ ์ ์‘ ์ง€์†์  ์‚ฌ์ „ ํ•™์Šต์ด LLM์˜ ์˜คํ”ˆ ๋„๋ฉ”์ธ ๊ธฐ๋Šฅ์— ํ•ด๊ฐ€ ๋˜๋Š”๊ฐ€? ์ด๋Ÿฌํ•œ ์งˆ๋ฌธ์— ๋Œ€ํ•ด Pythia [4]๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตฌ์ถ•๋œ ์ง€์†์ ์œผ๋กœ ์‚ฌ์ „ ํ•™์Šต๋œ ๋ชจ๋ธ์ธ FinPythia๋ฅผ ํ›ˆ๋ จํ•˜์—ฌ ๊ธˆ์œต ๋„๋ฉ”์ธ ๋‚ด์—์„œ ๋‹ตํ•ฉ๋‹ˆ๋‹ค.
The memory module plays a very important role in the agent architecture design. It stores information perceived from the environment and leverages the recorded memories to facilitate future actions. The memory module can help the agent to accumulate experiences, self-evolve, and behave in a more consistent, reasonable, and effective manner. This section provides a comprehensive overview of the memory module, focusing on its structures, formats, and operations.
๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์€ ์—์ด์ „ํŠธ ์•„ํ‚คํ…์ฒ˜ ์„ค๊ณ„์—์„œ ๋งค์šฐ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ชจ๋“ˆ์€ ํ™˜๊ฒฝ์—์„œ ์ธ์ง€ํ•œ ์ •๋ณด๋ฅผ ์ €์žฅํ•˜๊ณ  ๊ธฐ๋ก๋œ ๊ธฐ์–ต์„ ํ™œ์šฉํ•˜์—ฌ ๋ฏธ๋ž˜์˜ ํ–‰๋™์„ ์ด‰์ง„ํ•ฉ๋‹ˆ๋‹ค. ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์€ ์—์ด์ „ํŠธ๊ฐ€ ๊ฒฝํ—˜์„ ์ถ•์ ํ•˜๊ณ , ์Šค์Šค๋กœ ์ง„ํ™”ํ•˜๋ฉฐ, ๋” ์ผ๊ด€๋˜๊ณ  ํ•ฉ๋ฆฌ์ ์ด๋ฉฐ ํšจ๊ณผ์ ์œผ๋กœ ํ–‰๋™ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋„์™€์ค๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์˜ ๊ตฌ์กฐ, ํ˜•์‹ ๋ฐ ์ž‘๋™์— ๋Œ€ํ•ด ์ข…ํ•ฉ์ ์ธ ๊ฐœ์š”๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์€ ์—์ด์ „ํŠธ ์•„ํ‚คํ…์ฒ˜ ์„ค๊ณ„์—์„œ ๋งค์šฐ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์€ ํ™˜๊ฒฝ์—์„œ ์ธ์‹ํ•œ ์ •๋ณด๋ฅผ ์ €์žฅํ•˜๊ณ  ๊ธฐ๋ก๋œ ๋ฉ”๋ชจ๋ฆฌ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ํ–ฅํ›„ ํ–‰๋™์„ ์šฉ์ดํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค. ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์€ ์—์ด์ „ํŠธ๊ฐ€ ๊ฒฝํ—˜์„ ์ถ•์ ํ•˜๊ณ , ์Šค์Šค๋กœ ์ง„ํ™”ํ•˜๋ฉฐ, ๋ณด๋‹ค ์ผ๊ด€๋˜๊ณ  ํ•ฉ๋ฆฌ์ ์ด๋ฉฐ ํšจ๊ณผ์ ์ธ ๋ฐฉ์‹์œผ๋กœ ํ–‰๋™ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์˜ ๊ตฌ์กฐ, ํ˜•์‹ ๋ฐ ์ž‘๋™์— ์ค‘์ ์„ ๋‘๊ณ  ๋ฉ”๋ชจ๋ฆฌ ๋ชจ๋“ˆ์— ๋Œ€ํ•œ ํฌ๊ด„์ ์ธ ๊ฐœ์š”๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
In the above sections, we introduce both subjective and objective strategies for LLM-based autonomous agents evaluation. The evaluation of the agents play significant roles in this domain. However, both subjective and objective evaluation have their own strengths and weakness. Maybe, in practice, they should be combined to comprehensively evaluate the agents. We summarize the correspondence between the previous work and these evaluation strategies in Table 3.
์œ„์˜ ์„น์…˜์—์„œ๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ ์ฃผ๊ด€์  ๋ฐ ๊ฐ๊ด€์  ์ „๋žต์„ ๋ชจ๋‘ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์—์ด์ „ํŠธ์˜ ํ‰๊ฐ€๋Š” ์ด ๋ถ„์•ผ์—์„œ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ฃผ๊ด€์  ํ‰๊ฐ€์™€ ๊ฐ๊ด€์  ํ‰๊ฐ€๋Š” ๊ฐ๊ฐ์˜ ๊ฐ•์ ๊ณผ ์•ฝ์ ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค๋ฌด์—์„œ๋Š” ์ด๋“ค์„ ๊ฒฐํ•ฉํ•˜์—ฌ ์—์ด์ „ํŠธ๋ฅผ ์ข…ํ•ฉ์ ์œผ๋กœ ํ‰๊ฐ€ํ•ด์•ผ ํ• ์ง€๋„ ๋ชจ๋ฆ…๋‹ˆ๋‹ค. ์ด์ „ ์—ฐ๊ตฌ์™€ ์ด๋Ÿฌํ•œ ํ‰๊ฐ€ ์ „๋žต ๊ฐ„์˜ ๋Œ€์‘ ๊ด€๊ณ„๋ฅผ Table 3์— ์š”์•ฝํ•ฉ๋‹ˆ๋‹ค.
์œ„์˜ ์„น์…˜์—์„œ๋Š” LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ ํ‰๊ฐ€๋ฅผ ์œ„ํ•œ ์ฃผ๊ด€์  ๋ฐ ๊ฐ๊ด€์  ์ „๋žต์„ ๋ชจ๋‘ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์—์ด์ „ํŠธ์˜ ํ‰๊ฐ€๋Š” ์ด ์˜์—ญ์—์„œ ๋งค์šฐ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ฃผ๊ด€์  ํ‰๊ฐ€์™€ ๊ฐ๊ด€์  ํ‰๊ฐ€ ๋ชจ๋‘ ์žฅ๋‹จ์ ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ๋Š” ์ด ๋‘ ๊ฐ€์ง€ ํ‰๊ฐ€๋ฅผ ๊ฒฐํ•ฉํ•˜์—ฌ ์ข…ํ•ฉ์ ์œผ๋กœ ์—์ด์ „ํŠธ๋ฅผ ํ‰๊ฐ€ํ•ด์•ผ ํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด์ „ ์ž‘์—…๊ณผ ์ด๋Ÿฌํ•œ ํ‰๊ฐ€ ์ „๋žต ๊ฐ„์˜ ๋Œ€์‘์„ ํ‘œ 3์— ์š”์•ฝํ–ˆ์Šต๋‹ˆ๋‹ค.
Prompt engineering transcends the mere construction of prompts; it requires a blend of domain knowledge, understanding of the AI model, and a methodical approach to tailor prompts for different contexts. This might involve creating templates that can be programmatically modified based on a given dataset or context. For example, generating personalized responses based on user data might use a template that is dynamically filled with relevant user information.
ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์€ ๋‹จ์ˆœํžˆ ํ”„๋กฌํ”„ํŠธ๋ฅผ ๊ตฌ์„ฑํ•˜๋Š” ๊ฒƒ์„ ๋„˜์–ด์„ญ๋‹ˆ๋‹ค. ์ด๋Š” ๋„๋ฉ”์ธ ์ง€์‹, AI ๋ชจ๋ธ์— ๋Œ€ํ•œ ์ดํ•ด, ๊ทธ๋ฆฌ๊ณ  ๋‹ค์–‘ํ•œ ์ƒํ™ฉ์— ๋งž๊ฒŒ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์กฐ์ •ํ•˜๋Š” ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์„ ์š”๊ตฌํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ์…‹์ด๋‚˜ ์ƒํ™ฉ์— ๋”ฐ๋ผ ํ”„๋กœ๊ทธ๋žจ์ ์œผ๋กœ ์ˆ˜์ •ํ•  ์ˆ˜ ์žˆ๋Š” ํ…œํ”Œ๋ฆฟ์„ ์ƒ์„ฑํ•˜๋Š” ๊ฒƒ์„ ํฌํ•จํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฐœ์ธํ™”๋œ ์‘๋‹ต์„ ์ƒ์„ฑํ•˜๋Š” ๊ฒฝ์šฐ, ๊ด€๋ จ ์‚ฌ์šฉ์ž ์ •๋ณด๋กœ ๋™์ ์œผ๋กœ ์ฑ„์›Œ์ง€๋Š” ํ…œํ”Œ๋ฆฟ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
ํ”„๋กฌํ”„ํŠธ ์—”์ง€๋‹ˆ์–ด๋ง์€ ๋‹จ์ˆœํ•œ ํ”„๋กฌํ”„ํŠธ ๊ตฌ์„ฑ์„ ๋„˜์–ด ๋„๋ฉ”์ธ ์ง€์‹, AI ๋ชจ๋ธ์— ๋Œ€ํ•œ ์ดํ•ด, ๋‹ค์–‘ํ•œ ์ƒํ™ฉ์— ๋งž๋Š” ํ”„๋กฌํ”„ํŠธ๋ฅผ ๋งž์ถคํ™”ํ•˜๊ธฐ ์œ„ํ•œ ์ฒด๊ณ„์ ์ธ ์ ‘๊ทผ ๋ฐฉ์‹์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ ์„ธํŠธ ๋˜๋Š” ์ปจํ…์ŠคํŠธ์— ๋”ฐ๋ผ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ๋ฐฉ์‹์œผ๋กœ ์ˆ˜์ •ํ•  ์ˆ˜ ์žˆ๋Š” ํ…œํ”Œ๋ฆฟ์„ ๋งŒ๋“œ๋Š” ๊ฒƒ์ด ํฌํ•จ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์‚ฌ์šฉ์ž ๋ฐ์ดํ„ฐ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ฐœ์ธํ™”๋œ ์‘๋‹ต์„ ์ƒ์„ฑํ•˜๋Š” ๊ฒฝ์šฐ ๊ด€๋ จ ์‚ฌ์šฉ์ž ์ •๋ณด๋กœ ๋™์ ์œผ๋กœ ์ฑ„์›Œ์ง€๋Š” ํ…œํ”Œ๋ฆฟ์„ ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Figure 5 shows that singular distributions evolve differently for models using less than 410M parameters than for the larger ones. The heads of small models see their singular value distributions become increasingly uniform, up to a point where they degenerate abruptly, which again correlates with the LM performance drop. The singular value distributions of larger models tend to be more stable, and do not display clear monotonic patterns throughout training.
Figure 5๋Š” 410M ๋ฏธ๋งŒ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ชจ๋ธ๊ณผ ๋” ํฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ ๋‹จ์ผ ๋ถ„ํฌ๊ฐ€ ๋‹ค๋ฅด๊ฒŒ ์ง„ํ™”ํ•˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ž‘์€ ๋ชจ๋ธ์˜ ํ—ค๋“œ๋Š” ๋‹จ์ผ ๊ฐ’ ๋ถ„ํฌ๊ฐ€ ์ ์  ๋” ๊ท ์ผํ•ด์ง€๋‹ค๊ฐ€ ๊ฐ‘์ž๊ธฐ ํ‡ดํ™”ํ•˜๋Š” ์ง€์ ์— ๋„๋‹ฌํ•˜๋ฉฐ, ์ด๋Š” LM ์„ฑ๋Šฅ ์ €ํ•˜์™€ ๋‹ค์‹œ ์ƒ๊ด€๊ด€๊ณ„๋ฅผ ๋ณด์ž…๋‹ˆ๋‹ค. ๋” ํฐ ๋ชจ๋ธ์˜ ๋‹จ์ผ ๊ฐ’ ๋ถ„ํฌ๋Š” ๋” ์•ˆ์ •์ ์ธ ๊ฒฝํ–ฅ์„ ๋ณด์ด๋ฉฐ, ํ›ˆ๋ จ ๋‚ด๋‚ด ๋ช…ํ™•ํ•œ ๋‹จ์กฐ ํŒจํ„ด์„ ๋‚˜ํƒ€๋‚ด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 5๋Š” 410M ๋ฏธ๋งŒ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ ํŠน์ด ๋ถ„ํฌ๊ฐ€ ๋” ํฐ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ์™€ ๋‹ค๋ฅด๊ฒŒ ์ง„ํ™”ํ•œ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ž‘์€ ๋ชจ๋ธ์˜ ํ—ค๋“œ๋Š” ํŠน์ด๊ฐ’ ๋ถ„ํฌ๊ฐ€ ์ ์  ๋” ๊ท ์ผํ•ด์ง€๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” ๋‹ค์‹œ LM ์„ฑ๋Šฅ ์ €ํ•˜์™€ ์ƒ๊ด€๊ด€๊ณ„๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ๋” ํฐ ๋ชจ๋ธ์˜ ํŠน์ด๊ฐ’ ๋ถ„ํฌ๋Š” ๋” ์•ˆ์ •์ ์ธ ๊ฒฝํ–ฅ์ด ์žˆ์œผ๋ฉฐ, ํ›ˆ๋ จ ์ „๋ฐ˜์— ๊ฑธ์ณ ๋ช…ํ™•ํ•œ ๋‹จ์กฐ๋กœ์šด ํŒจํ„ด์„ ๋ณด์ด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
We introduce an exposition of the datasets employed in our work for evaluating the performance in terms of text generation and toxicity rejection abilities. Our experiments consistently employed approach for the generation of responses by LLaMA-based models involves the implementation of a greedy decoding strategy. In the case of Mistral-based models, the generation of responses was facilitated by setting the repetition penalty Keskar et al. (2019) to 1.15.
์šฐ๋ฆฌ ์—ฐ๊ตฌ์—์„œ ํ…์ŠคํŠธ ์ƒ์„ฑ ๋ฐ ์œ ํ•ด์„ฑ ๊ฑฐ๋ถ€ ๋Šฅ๋ ฅ ์ธก๋ฉด์—์„œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์šฉ๋œ ๋ฐ์ดํ„ฐ์…‹์— ๋Œ€ํ•œ ์„ค๋ช…์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์‹คํ—˜์—์„œ๋Š” LLaMA ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์ด ์‘๋‹ต์„ ์ƒ์„ฑํ•˜๋Š” ๋ฐ ์ผ๊ด€๋˜๊ฒŒ ํƒ์š•์  ๋””์ฝ”๋”ฉ ์ „๋žต์„ ๊ตฌํ˜„ํ•˜๋Š” ์ ‘๊ทผ ๋ฐฉ์‹์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค. Mistral ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, ์‘๋‹ต ์ƒ์„ฑ์„ ์œ„ํ•ด ๋ฐ˜๋ณต ํŽ˜๋„ํ‹ฐ๋ฅผ 1.15๋กœ ์„ค์ •ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์‚ฌ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค(Keskar et al., 2019).
ํ…์ŠคํŠธ ์ƒ์„ฑ ๋ฐ ๋…์„ฑ ๊ฑฐ๋ถ€ ๋Šฅ๋ ฅ ์ธก๋ฉด์—์„œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ๋ณธ ์—ฐ๊ตฌ์— ์‚ฌ์šฉ๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ๋Œ€ํ•ด ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. ์‹คํ—˜์—์„œ๋Š” LLaMA ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์˜ ์‘๋‹ต ์ƒ์„ฑ์— ๋Œ€ํ•œ ์ ‘๊ทผ ๋ฐฉ์‹์„ ์ผ๊ด€๋˜๊ฒŒ ์‚ฌ์šฉํ–ˆ์œผ๋ฉฐ, ํƒ์š•์Šค๋Ÿฌ์šด ๋””์ฝ”๋”ฉ ์ „๋žต์„ ๊ตฌํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค. Mistral ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์˜ ๊ฒฝ์šฐ, ๋ฐ˜๋ณต ํŽ˜๋„ํ‹ฐ Keskar ๋“ฑ(2019)์„ 1.15๋กœ ์„ค์ •ํ•˜์—ฌ ์‘๋‹ต ์ƒ์„ฑ์„ ์šฉ์ดํ•˜๊ฒŒ ํ–ˆ์Šต๋‹ˆ๋‹ค.
In Appendix Figure 8, we aim to provide some insight into the kinds of tokens where the improvements occur. Namely, while thinking appears to help for many tokens in the example, inspection suggests it disproportionately help to predict tokens where recalling relevant information is useful, such as the name of an applicable theorem or the start of the next step in a proof. Notably, this would align well with the framing proposed by Prystawski et al. (2024).
๋ถ€๋ก ๊ทธ๋ฆผ 8์—์„œ ์šฐ๋ฆฌ๋Š” ๊ฐœ์„ ์ด ๋ฐœ์ƒํ•˜๋Š” ํ† ํฐ์˜ ์ข…๋ฅ˜์— ๋Œ€ํ•œ ํ†ต์ฐฐ์„ ์ œ๊ณตํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰, ์˜ˆ์ œ์—์„œ ์ƒ๊ฐํ•˜๋Š” ๊ฒƒ์ด ๋งŽ์€ ํ† ํฐ์— ๋„์›€์ด ๋˜๋Š” ๊ฒƒ์ฒ˜๋Ÿผ ๋ณด์ด์ง€๋งŒ, ๊ฒ€ํ†  ๊ฒฐ๊ณผ ๊ด€๋ จ ์ •๋ณด๋ฅผ ํšŒ์ƒํ•˜๋Š” ๊ฒƒ์ด ์œ ์šฉํ•œ ํ† ํฐ, ์˜ˆ๋ฅผ ๋“ค์–ด ์ ์šฉ ๊ฐ€๋Šฅํ•œ ์ •๋ฆฌ์˜ ์ด๋ฆ„์ด๋‚˜ ์ฆ๋ช…์˜ ๋‹ค์Œ ๋‹จ๊ณ„์˜ ์‹œ์ž‘ ๋ถ€๋ถ„์„ ์˜ˆ์ธกํ•˜๋Š” ๋ฐ ๋ถˆ๊ท ํ˜•์ ์œผ๋กœ ๋„์›€์ด ๋œ๋‹ค๋Š” ๊ฒƒ์„ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ, ์ด๋Š” Prystawski et al. (2024)์ด ์ œ์•ˆํ•œ ํ”„๋ ˆ์ด๋ฐ๊ณผ ์ž˜ ๋งž์•„๋–จ์–ด์ง‘๋‹ˆ๋‹ค.
๋ถ€๋ก ๊ทธ๋ฆผ 8์—์„œ๋Š” ๊ฐœ์„ ์ด ๋ฐœ์ƒํ•˜๋Š” ํ† ํฐ์˜ ์ข…๋ฅ˜์— ๋Œ€ํ•œ ์ธ์‚ฌ์ดํŠธ๋ฅผ ์ œ๊ณตํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰, ์˜ˆ์ œ์—์„œ ์‚ฌ๊ณ ๊ฐ€ ๋งŽ์€ ํ† ํฐ์— ๋„์›€์ด ๋˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณด์ด์ง€๋งŒ, ๊ฒ€์‚ฌ ๊ฒฐ๊ณผ ์ ์šฉ ๊ฐ€๋Šฅํ•œ ์ •๋ฆฌ์˜ ์ด๋ฆ„์ด๋‚˜ ์ฆ๋ช…์˜ ๋‹ค์Œ ๋‹จ๊ณ„ ์‹œ์ž‘๊ณผ ๊ฐ™์ด ๊ด€๋ จ ์ •๋ณด๋ฅผ ๊ธฐ์–ตํ•˜๋Š” ๊ฒƒ์ด ์œ ์šฉํ•œ ํ† ํฐ์„ ์˜ˆ์ธกํ•˜๋Š” ๋ฐ ๋ถˆ๊ท ํ˜•์ ์œผ๋กœ ๋„์›€์ด ๋˜๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์ด๋Š” Prystawski ๋“ฑ(2024)์ด ์ œ์•ˆํ•œ ํ”„๋ ˆ์ž„๊ณผ ์ž˜ ๋ถ€ํ•ฉํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
**Impact of Feedback and Human Oversight on Agent Systems.** When solving a complex problem, it is extremely unlikely that one provides a correct, robust solution on their first try. Instead, one might pose a potential solution before criticizing it and refining it. One could also consult with someone else and receive feedback from another perspective. The same idea of iterative feedback and refinement is essential for helping agents solve complex problems.
ํ”ผ๋“œ๋ฐฑ๊ณผ ์ธ๊ฐ„ ๊ฐ๋…์ด ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ. ๋ณต์žกํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ๋•Œ, ์ฒ˜์Œ ์‹œ๋„์—์„œ ์˜ฌ๋ฐ”๋ฅด๊ณ  ๊ฒฌ๊ณ ํ•œ ํ•ด๊ฒฐ์ฑ…์„ ์ œ๊ณตํ•˜๋Š” ๊ฒƒ์€ ๋งค์šฐ ๋“œ๋ญ…๋‹ˆ๋‹ค. ๋Œ€์‹ , ์ž ์žฌ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•œ ํ›„ ์ด๋ฅผ ๋น„ํŒํ•˜๊ณ  ๊ฐœ์„ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ๋‹ค๋ฅธ ์‚ฌ๋žŒ๊ณผ ์ƒ์˜ํ•˜์—ฌ ๋‹ค๋ฅธ ๊ด€์ ์—์„œ ํ”ผ๋“œ๋ฐฑ์„ ๋ฐ›์„ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐ˜๋ณต์ ์ธ ํ”ผ๋“œ๋ฐฑ๊ณผ ๊ฐœ์„ ์˜ ๊ฐœ๋…์€ ์—์ด์ „ํŠธ๊ฐ€ ๋ณต์žกํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ํ•„์ˆ˜์ ์ž…๋‹ˆ๋‹ค.
**ํ”ผ๋“œ๋ฐฑ๊ณผ ์ธ๊ฐ„์˜ ๊ฐ๋…์ด ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ**. ๋ณต์žกํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•  ๋•Œ ์ฒซ ๋ฒˆ์งธ ์‹œ๋„์—์„œ ์ •ํ™•ํ•˜๊ณ  ๊ฐ•๋ ฅํ•œ ์†”๋ฃจ์…˜์„ ์ œ๊ณตํ•  ๊ฐ€๋Šฅ์„ฑ์€ ๋งค์šฐ ๋‚ฎ์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ์ž ์žฌ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•œ ํ›„ ๋น„ํŒํ•˜๊ณ  ๊ฐœ์„ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹ค๋ฅธ ์‚ฌ๋žŒ๊ณผ ์ƒ์˜ํ•˜๊ณ  ๋‹ค๋ฅธ ๊ด€์ ์—์„œ ํ”ผ๋“œ๋ฐฑ์„ ๋ฐ›์„ ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฐ˜๋ณต์ ์ธ ํ”ผ๋“œ๋ฐฑ๊ณผ ๊ฐœ์„ ์ด๋ผ๋Š” ๋™์ผํ•œ ์•„์ด๋””์–ด๋Š” ์—์ด์ „ํŠธ๊ฐ€ ๋ณต์žกํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” ๋ฐ ํ•„์ˆ˜์ ์ž…๋‹ˆ๋‹ค.
Language modeling is a long-standing research topic, dating back to the 1950s with Shannon's application of information theory to human language, where he measured how well simple n-gram language models predict or compress natural language text [3]. Since then, statistical language modeling became fundamental to many natural language understanding and generation tasks, ranging from speech recognition, machine translation, to information retrieval [4, 5, 6].
์–ธ์–ด ๋ชจ๋ธ๋ง์€ ์˜ค๋žœ ์—ฐ๊ตฌ ์ฃผ์ œ๋กœ, 1950๋…„๋Œ€์— Shannon์ด ์ •๋ณด ์ด๋ก ์„ ์ธ๊ฐ„ ์–ธ์–ด์— ์ ์šฉํ•˜๋ฉด์„œ ์‹œ์ž‘๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Š” ๊ฐ„๋‹จํ•œ n-๊ทธ๋žจ ์–ธ์–ด ๋ชจ๋ธ์ด ์ž์—ฐ์–ด ํ…์ŠคํŠธ๋ฅผ ์–ผ๋งˆ๋‚˜ ์ž˜ ์˜ˆ์ธกํ•˜๊ฑฐ๋‚˜ ์••์ถ•ํ•˜๋Š”์ง€๋ฅผ ์ธก์ •ํ–ˆ์Šต๋‹ˆ๋‹ค [3]. ๊ทธ ์ดํ›„๋กœ ํ†ต๊ณ„์  ์–ธ์–ด ๋ชจ๋ธ๋ง์€ ์Œ์„ฑ ์ธ์‹, ๊ธฐ๊ณ„ ๋ฒˆ์—ญ, ์ •๋ณด ๊ฒ€์ƒ‰ ๋“ฑ ๋‹ค์–‘ํ•œ ์ž์—ฐ์–ด ์ดํ•ด ๋ฐ ์ƒ์„ฑ ์ž‘์—…์˜ ๊ธฐ๋ณธ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค [4, 5, 6].
์–ธ์–ด ๋ชจ๋ธ๋ง์€ 1950๋…„๋Œ€ ์„€๋„Œ์ด ์ •๋ณด ์ด๋ก ์„ ์ธ๊ฐ„ ์–ธ์–ด์— ์ ์šฉํ•˜์—ฌ ๋‹จ์ˆœํ•œ n-๊ทธ๋žจ ์–ธ์–ด ๋ชจ๋ธ์ด ์ž์—ฐ์–ด ํ…์ŠคํŠธ๋ฅผ ์–ผ๋งˆ๋‚˜ ์ž˜ ์˜ˆ์ธกํ•˜๊ฑฐ๋‚˜ ์••์ถ•ํ•˜๋Š”์ง€๋ฅผ ์ธก์ •ํ•œ ๊ฒƒ์œผ๋กœ ๊ฑฐ์Šฌ๋Ÿฌ ์˜ฌ๋ผ๊ฐ€๋Š” ์˜ค๋žœ ์—ฐ๊ตฌ ์ฃผ์ œ์ž…๋‹ˆ๋‹ค[3]. ๊ทธ ์ดํ›„๋กœ ํ†ต๊ณ„ ์–ธ์–ด ๋ชจ๋ธ๋ง์€ ์Œ์„ฑ ์ธ์‹, ๊ธฐ๊ณ„ ๋ฒˆ์—ญ, ์ •๋ณด ๊ฒ€์ƒ‰ ๋“ฑ ๋‹ค์–‘ํ•œ ์ž์—ฐ์–ด ์ดํ•ด ๋ฐ ์ƒ์„ฑ ์ž‘์—…์˜ ๊ธฐ์ดˆ๊ฐ€ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค[4, 5, 6].
Reinforcing LM biases.A point of concern for the authors is the unintended consequences of this iterative algorithm, such as the amplification of problematic social biases (stereotypes or slurs about gender, race, etc.). Relatedly, one observed challenge in this process is the algorithm's difficulty in producing balanced labels, which reflected models' prior biases. We hope future work will lead to better understanding of the pros and cons of the approach.
LM ํŽธํ–ฅ ๊ฐ•ํ™”. ์ €์ž๋“ค์ด ์šฐ๋ คํ•˜๋Š” ์  ์ค‘ ํ•˜๋‚˜๋Š” ์ด ๋ฐ˜๋ณต ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์˜๋„์น˜ ์•Š์€ ๊ฒฐ๊ณผ๋กœ, ๋ฌธ์ œ์ ์ธ ์‚ฌํšŒ์  ํŽธํ–ฅ(์„ฑ๋ณ„, ์ธ์ข… ๋“ฑ์— ๋Œ€ํ•œ ๊ณ ์ •๊ด€๋…์ด๋‚˜ ๋น„๋ฐฉ ๋“ฑ)์ด ์ฆํญ๋˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด์™€ ๊ด€๋ จํ•˜์—ฌ, ์ด ๊ณผ์ •์—์„œ ๊ด€์ฐฐ๋œ ํ•œ ๊ฐ€์ง€ ๋„์ „ ๊ณผ์ œ๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋ชจ๋ธ์˜ ๊ธฐ์กด ํŽธํ–ฅ์„ ๋ฐ˜์˜ํ•œ ๊ท ํ˜• ์žกํžŒ ๋ ˆ์ด๋ธ”์„ ์ƒ์„ฑํ•˜๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช๋Š”๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ํ–ฅํ›„ ์—ฐ๊ตฌ๊ฐ€ ์ด ์ ‘๊ทผ ๋ฐฉ์‹์˜ ์žฅ๋‹จ์ ์„ ๋” ์ž˜ ์ดํ•ดํ•˜๋Š” ๋ฐ ๊ธฐ์—ฌํ•˜๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
LM ํŽธํ–ฅ์„ฑ ๊ฐ•ํ™”.์ €์ž๋“ค์ด ์šฐ๋ คํ•˜๋Š” ํ•œ ๊ฐ€์ง€ ์ ์€ ์ด ๋ฐ˜๋ณต์ ์ธ ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ๋ฌธ์ œ๊ฐ€ ๋˜๋Š” ์‚ฌํšŒ์  ํŽธ๊ฒฌ(์„ฑ๋ณ„, ์ธ์ข… ๋“ฑ์— ๋Œ€ํ•œ ๊ณ ์ •๊ด€๋…์ด๋‚˜ ๋น„๋ฐฉ ๋“ฑ)์„ ์ฆํญ์‹œํ‚ค๋Š” ๋“ฑ ์˜๋„ํ•˜์ง€ ์•Š์€ ๊ฒฐ๊ณผ๋ฅผ ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ์ด์™€ ๊ด€๋ จํ•˜์—ฌ ์ด ๊ณผ์ •์—์„œ ๊ด€์ฐฐ๋œ ํ•œ ๊ฐ€์ง€ ๋ฌธ์ œ๋Š” ๋ชจ๋ธ์˜ ์‚ฌ์ „ ํŽธ๊ฒฌ์„ ๋ฐ˜์˜ํ•˜๋Š” ๊ท ํ˜• ์žกํžŒ ๋ ˆ์ด๋ธ”์„ ์ƒ์„ฑํ•˜๋Š” ๋ฐ ์žˆ์–ด ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์–ด๋ ค์›€์ž…๋‹ˆ๋‹ค. ํ–ฅํ›„ ์—ฐ๊ตฌ๋ฅผ ํ†ตํ•ด ์ด ์ ‘๊ทผ ๋ฐฉ์‹์˜ ์žฅ๋‹จ์ ์„ ๋” ์ž˜ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
The effectiveness of these strategies is demonstrated in Table 8, which details the Yi-9B base model's performance across a variety of benchmarks, including common sense, reasoning, knowledge, coding, and mathematics. It underscores the competitive advantages of Yi-9B base model in specific domains, illustrating the efficacy of our methodology in enhancing model performance by optimally adjusting the interplay between data characteristics and model size.
์ด๋Ÿฌํ•œ ์ „๋žต์˜ ํšจ๊ณผ๋Š” Table 8์—์„œ ์ž…์ฆ๋˜์—ˆ์œผ๋ฉฐ, ์—ฌ๊ธฐ์—๋Š” ์ƒ์‹, ์ถ”๋ก , ์ง€์‹, ์ฝ”๋”ฉ ๋ฐ ์ˆ˜ํ•™์„ ํฌํ•จํ•œ ๋‹ค์–‘ํ•œ ๋ฒค์น˜๋งˆํฌ์—์„œ Yi-9B ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์ด ์ž์„ธํžˆ ์„ค๋ช…๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ํŠน์ • ๋„๋ฉ”์ธ์—์„œ Yi-9B ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ๊ฒฝ์Ÿ ์šฐ์œ„๋ฅผ ๊ฐ•์กฐํ•˜๋ฉฐ, ๋ฐ์ดํ„ฐ ํŠน์„ฑ๊ณผ ๋ชจ๋ธ ํฌ๊ธฐ ๊ฐ„์˜ ์ƒํ˜ธ์ž‘์šฉ์„ ์ตœ์ ํ™”ํ•˜์—ฌ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ์šฐ๋ฆฌ์˜ ๋ฐฉ๋ฒ•๋ก ์˜ ํšจ๋Šฅ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
์ด๋Ÿฌํ•œ ์ „๋žต์˜ ํšจ๊ณผ๋Š” ์ƒ์‹, ์ถ”๋ก , ์ง€์‹, ์ฝ”๋”ฉ, ์ˆ˜ํ•™ ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฒค์น˜๋งˆํฌ์—์„œ Yi-9B ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ์„ฑ๋Šฅ์„ ์ž์„ธํžˆ ๋ณด์—ฌ์ฃผ๋Š” ํ‘œ 8์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ํŠน์ • ๋„๋ฉ”์ธ์—์„œ Yi-9B ๊ธฐ๋ณธ ๋ชจ๋ธ์˜ ๊ฒฝ์Ÿ ์šฐ์œ„๋ฅผ ๊ฐ•์กฐํ•˜๋ฉฐ, ๋ฐ์ดํ„ฐ ํŠน์„ฑ๊ณผ ๋ชจ๋ธ ํฌ๊ธฐ ๊ฐ„์˜ ์ƒํ˜ธ ์ž‘์šฉ์„ ์ตœ์ ์œผ๋กœ ์กฐ์ •ํ•˜์—ฌ ๋ชจ๋ธ ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๋ฐฉ๋ฒ•๋ก ์˜ ํšจ๊ณผ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
De-duplication refers to the process of removing duplicate instances or repeated occurrences of the same data in a dataset. Duplicate data points can introduce biases in the model training process and reduce the diversity, as the model may learn from the same examples multiple times, potentially leading to overfitting on those particular instances. Some works [125] have shown that de-duplication improves models' ability to generalize to new, unseen data.
์ค‘๋ณต ์ œ๊ฑฐ๋Š” ๋ฐ์ดํ„ฐ์…‹์—์„œ ์ค‘๋ณต๋œ ์ธ์Šคํ„ด์Šค๋‚˜ ๋ฐ˜๋ณต๋œ ๋ฐ์ดํ„ฐ๋ฅผ ์ œ๊ฑฐํ•˜๋Š” ๊ณผ์ •์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ค‘๋ณต๋œ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋Š” ๋ชจ๋ธ ํ•™์Šต ๊ณผ์ •์—์„œ ํŽธํ–ฅ์„ ๋„์ž…ํ•˜๊ณ  ๋‹ค์–‘์„ฑ์„ ๊ฐ์†Œ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์ด ๋™์ผํ•œ ์˜ˆ์ œ์—์„œ ์—ฌ๋Ÿฌ ๋ฒˆ ํ•™์Šตํ•˜๊ฒŒ ๋˜์–ด ํŠน์ • ์ธ์Šคํ„ด์Šค์— ๊ณผ์ ํ•ฉ๋  ๊ฐ€๋Šฅ์„ฑ์„ ๋†’์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ผ๋ถ€ ์—ฐ๊ตฌ [125]์—์„œ๋Š” ์ค‘๋ณต ์ œ๊ฑฐ๊ฐ€ ๋ชจ๋ธ์˜ ์ƒˆ๋กœ์šด, ๋ณด์ง€ ๋ชปํ•œ ๋ฐ์ดํ„ฐ์— ๋Œ€ํ•œ ์ผ๋ฐ˜ํ™” ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚จ๋‹ค๊ณ  ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.
์ค‘๋ณต ์ œ๊ฑฐ๋Š” ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์—์„œ ์ค‘๋ณต๋œ ์ธ์Šคํ„ด์Šค ๋˜๋Š” ๋™์ผํ•œ ๋ฐ์ดํ„ฐ์˜ ๋ฐ˜๋ณต์ ์ธ ๋ฐœ์ƒ์„ ์ œ๊ฑฐํ•˜๋Š” ํ”„๋กœ์„ธ์Šค๋ฅผ ๋งํ•ฉ๋‹ˆ๋‹ค. ์ค‘๋ณต๋œ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ๋Š” ๋ชจ๋ธ ํ•™์Šต ํ”„๋กœ์„ธ์Šค์— ํŽธํ–ฅ์„ ์œ ๋„ํ•˜๊ณ  ๋ชจ๋ธ์ด ๋™์ผํ•œ ์˜ˆ์ œ๋ฅผ ์—ฌ๋Ÿฌ ๋ฒˆ ํ•™์Šตํ•˜์—ฌ ํŠน์ • ์ธ์Šคํ„ด์Šค์— ๋Œ€ํ•œ ๊ณผ์ ํ•ฉ์œผ๋กœ ์ด์–ด์งˆ ์ˆ˜ ์žˆ์œผ๋ฏ€๋กœ ๋‹ค์–‘์„ฑ์„ ์ €ํ•˜์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ผ๋ถ€ ์—ฐ๊ตฌ[125]์—์„œ๋Š” ์ค‘๋ณต ์ œ๊ฑฐ๊ฐ€ ๋ณด์ด์ง€ ์•Š๋Š” ์ƒˆ๋กœ์šด ๋ฐ์ดํ„ฐ๋กœ ์ผ๋ฐ˜ํ™”ํ•˜๋Š” ๋ชจ๋ธ์˜ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚จ๋‹ค๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.
Figure 9: **Scaling on Hutter LLM for Varying \(k\).** A sample of size \(T_{0}\) is used to approximate the true distribution \(p\) via \(p_{\text{AI}}\). Then, a Hutter-type model is learned on a sample of size \(T\) from \(p_{\text{AI}}\), and evaluated on the true data distribution \(p\). Each horizontal line corresponds to the asymptote \(k^{-\beta c}\asymp T_{0}^{-c}\), for different values of \(T_{0}\). The diagonal line corresponds to \(T^{-c}\).
Figure 9: **๋‹ค์–‘ํ•œ \(k\)์— ๋Œ€ํ•œ Hutter LLM์˜ ์Šค์ผ€์ผ๋ง.** ํฌ๊ธฐ \(T_{0}\)์˜ ์ƒ˜ํ”Œ์„ ์‚ฌ์šฉํ•˜์—ฌ \(p_{\text{AI}}\)๋ฅผ ํ†ตํ•ด ์‹ค์ œ ๋ถ„ํฌ \(p\)๋ฅผ ๊ทผ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ, \(p_{\text{AI}}\)์—์„œ ํฌ๊ธฐ \(T\)์˜ ์ƒ˜ํ”Œ์„ ์‚ฌ์šฉํ•˜์—ฌ Hutter ์œ ํ˜• ๋ชจ๋ธ์„ ํ•™์Šตํ•˜๊ณ , ์‹ค์ œ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ \(p\)์—์„œ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ์ˆ˜ํ‰์„ ์€ \(T_{0}\)์˜ ๋‹ค๋ฅธ ๊ฐ’์— ๋Œ€ํ•ด \(k^{-\beta c}\asymp T_{0}^{-c}\)์˜ ์ ๊ทผ์„ ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ๋Œ€๊ฐ์„ ์€ \(T^{-c}\)์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 9: **๋‹ค์–‘ํ•œ \(k\)์— ๋Œ€ํ•œ Hutter LLM์˜ ์Šค์ผ€์ผ๋ง.** ํฌ๊ธฐ \(T_{0}\)์˜ ์ƒ˜ํ”Œ์€ \(p_{\text{AI}}\)๋ฅผ ํ†ตํ•ด ์‹ค์ œ ๋ถ„ํฌ \(p\)๋ฅผ ๊ทผ์‚ฌํ™”ํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ, \(p_{\text{AI}}\)์—์„œ ํฌ๊ธฐ \(T\)์˜ ์ƒ˜ํ”Œ์— ๋Œ€ํ•ด Hutterํ˜• ๋ชจ๋ธ์„ ํ•™์Šตํ•˜๊ณ , ์‹ค์ œ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ \(p\)์— ๋Œ€ํ•ด ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ๊ฐ€๋กœ์„ ์€ ๋‹ค๋ฅธ ๊ฐ’ \(T_{0}\)์— ๋Œ€ํ•ด ์ ๊ทผ \(k^{-\beta c}\asymp T_{0}^{-c}\)์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค. ๋Œ€๊ฐ์„ ์€ \(T^{-c}\)์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค.
As shown in Figure 2, the implementation of RL enables the model to achieve additional improvements in both rationale qualities and task performance. However, in line with Joshi et al. (2023)'s findings, a slight decrease in task performance is observed at the cost of maximized rationale qualities when selecting top-quality rationales. There may be several underlying factors involved (e.g., models' unfaithfulness), but this is not the scope of this work.
Figure 2์—์„œ ๋ณผ ์ˆ˜ ์žˆ๋“ฏ์ด, RL์˜ ๊ตฌํ˜„์€ ๋ชจ๋ธ์ด ๋…ผ๋ฆฌ์  ํ’ˆ์งˆ๊ณผ ์ž‘์—… ์„ฑ๋Šฅ ๋ชจ๋‘์—์„œ ์ถ”๊ฐ€์ ์ธ ํ–ฅ์ƒ์„ ์ด๋ฃจ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ Joshi et al. (2023)์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์™€ ์ผ์น˜ํ•˜๊ฒŒ, ์ตœ๊ณ  ํ’ˆ์งˆ์˜ ๋…ผ๋ฆฌ๋ฅผ ์„ ํƒํ•  ๋•Œ ๋…ผ๋ฆฌ์  ํ’ˆ์งˆ์ด ๊ทน๋Œ€ํ™”๋˜๋Š” ๋Œ€๊ฐ€๋กœ ์ž‘์—… ์„ฑ๋Šฅ์ด ์•ฝ๊ฐ„ ๊ฐ์†Œํ•˜๋Š” ํ˜„์ƒ์ด ๊ด€์ฐฐ๋ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ๊ทผ๋ณธ์ ์ธ ์š”์ธ๋“ค(์˜ˆ: ๋ชจ๋ธ์˜ ๋ถˆ์„ฑ์‹ค์„ฑ)์ด ์žˆ์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ์ด๋Š” ๋ณธ ์—ฐ๊ตฌ์˜ ๋ฒ”์œ„์— ํฌํ•จ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 2์™€ ๊ฐ™์ด, RL์„ ๊ตฌํ˜„ํ•˜๋ฉด ๋ชจ๋ธ์ด ๊ทผ๊ฑฐ ํ’ˆ์งˆ๊ณผ ์ž‘์—… ์„ฑ๋Šฅ ๋ชจ๋‘์—์„œ ์ถ”๊ฐ€์ ์ธ ๊ฐœ์„ ์„ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ Joshi ๋“ฑ(2023)์˜ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์— ๋”ฐ๋ฅด๋ฉด, ์ตœ๊ณ  ํ’ˆ์งˆ์˜ ๊ทผ๊ฑฐ๋ฅผ ์„ ํƒํ•  ๋•Œ ๊ทผ๊ฑฐ ํ’ˆ์งˆ์ด ๊ทน๋Œ€ํ™”๋˜๋Š” ๋Œ€๊ฐ€๋กœ ์ž‘์—… ์„ฑ๋Šฅ์ด ์•ฝ๊ฐ„ ๊ฐ์†Œํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ๊ทผ๋ณธ์ ์ธ ์š”์ธ(์˜ˆ: ๋ชจ๋ธ์˜ ๋ถˆ์„ฑ์‹ค์„ฑ)์ด ๊ด€๋ จ๋˜์–ด ์žˆ์„ ์ˆ˜ ์žˆ์ง€๋งŒ, ์ด๋Š” ๋ณธ ์—ฐ๊ตฌ์˜ ๋ฒ”์œ„๊ฐ€ ์•„๋‹™๋‹ˆ๋‹ค.
We now investigate our ability to steal production language models, focusing on five of OpenAI's models available on 1 January 2024: ada, babbage, babbage-002, gpt-3.5-turbo-instruct, and gpt-3.5-turbo-1106. We selected these models because these were the only production models which were able to receive advance permission to attempt an extraction attack; we are exceptionally grateful to OpenAI for allowing us to perform this research using their models.
์ด์ œ ์šฐ๋ฆฌ๋Š” ์ƒ์šฉ ์–ธ์–ด ๋ชจ๋ธ์„ ํƒˆ์ทจํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์„ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. 2024๋…„ 1์›” 1์ผ ๊ธฐ์ค€์œผ๋กœ ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ OpenAI์˜ ๋‹ค์„ฏ ๊ฐ€์ง€ ๋ชจ๋ธ์ธ ada, babbage, babbage-002, gpt-3.5-turbo-instruct, ๊ทธ๋ฆฌ๊ณ  gpt-3.5-turbo-1106์— ์ดˆ์ ์„ ๋งž์ถฅ๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ๋“ค์„ ์„ ํƒํ•œ ์ด์œ ๋Š” ์‚ฌ์ „ ํ—ˆ๊ฐ€๋ฅผ ๋ฐ›์•„ ์ถ”์ถœ ๊ณต๊ฒฉ์„ ์‹œ๋„ํ•  ์ˆ˜ ์žˆ์—ˆ๋˜ ์œ ์ผํ•œ ์ƒ์šฉ ๋ชจ๋ธ๋“ค์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋ชจ๋ธ ์‚ฌ์šฉ์„ ํ—ˆ๋ฝํ•ด์ค€ OpenAI์— ๋งค์šฐ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค.
์ด์ œ 2024๋…„ 1์›” 1์ผ์— ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋Š” OpenAI์˜ 5๊ฐ€์ง€ ๋ชจ๋ธ(ada, babbage, babbage-002, gpt-3.5-turbo-instruct, gpt-3.5-turbo-1106)์— ์ดˆ์ ์„ ๋งž์ถฐ ํ”„๋กœ๋•์…˜ ์–ธ์–ด ๋ชจ๋ธ์„ ํ›”์น  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์„ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ๋“ค์„ ์„ ํƒํ•œ ์ด์œ ๋Š” ์ถ”์ถœ ๊ณต๊ฒฉ์„ ์‹œ๋„ํ•  ์ˆ˜ ์žˆ๋Š” ์‚ฌ์ „ ํ—ˆ๊ฐ€๋ฅผ ๋ฐ›์„ ์ˆ˜ ์žˆ๋Š” ์œ ์ผํ•œ ํ”„๋กœ๋•์…˜ ๋ชจ๋ธ์ด์—ˆ๊ธฐ ๋•Œ๋ฌธ์ด๋ฉฐ, OpenAI๊ฐ€ ์ž์‚ฌ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ์ด ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ—ˆ์šฉํ•ด ์ค€ ๊ฒƒ์— ๋Œ€ํ•ด ํŠน๋ณ„ํžˆ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค.
Chris Callison-Burch and Daphne Ippolito's research is supported in part by the DARPA KAIROS Program (contract FA8750-19-2-1004), the DARPA LwLL Program (contract FA8750-19-2-0201), and the IARPA BETTER Program (contract 2019-19051600004). The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of DARPA, IARPA, or the U.S. Government.
Chris Callison-Burch์™€ Daphne Ippolito์˜ ์—ฐ๊ตฌ๋Š” ๋ถ€๋ถ„์ ์œผ๋กœ DARPA KAIROS ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ ๋ฒˆํ˜ธ FA8750-19-2-1004), DARPA LwLL ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ ๋ฒˆํ˜ธ FA8750-19-2-0201), ๊ทธ๋ฆฌ๊ณ  IARPA BETTER ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ ๋ฒˆํ˜ธ 2019-19051600004)์˜ ์ง€์›์„ ๋ฐ›์•˜์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ ํฌํ•จ๋œ ๊ฒฌํ•ด์™€ ๊ฒฐ๋ก ์€ ์ €์ž๋“ค์˜ ๊ฒƒ์ด๋ฉฐ, DARPA, IARPA ๋˜๋Š” ๋ฏธ๊ตญ ์ •๋ถ€์˜ ๊ณต์‹ ์ •์ฑ…์„ ๋ฐ˜๋“œ์‹œ ๋Œ€๋ณ€ํ•˜๋Š” ๊ฒƒ์€ ์•„๋‹™๋‹ˆ๋‹ค.
Chris Callison-Burch์™€ Daphne Ippolito์˜ ์—ฐ๊ตฌ๋Š” ๋ถ€๋ถ„์ ์œผ๋กœ DARPA KAIROS ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ FA8750-19-2-1004), DARPA LwLL ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ FA8750-19-2-0201), IARPA BETTER ํ”„๋กœ๊ทธ๋žจ(๊ณ„์•ฝ 2019-19051600004)์˜ ์ง€์›์„ ๋ฐ›๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์— ํฌํ•จ๋œ ๊ฒฌํ•ด์™€ ๊ฒฐ๋ก ์€ ์ €์ž์˜ ๊ฒฌํ•ด์ด๋ฉฐ, ์ด๋ฅผ DARPA, IARPA ๋˜๋Š” ๋ฏธ๊ตญ ์ •๋ถ€์˜ ๊ณต์‹ ์ •์ฑ…, ๋ช…์‹œ์ ์ด๋“  ๋ฌต์‹œ์ ์ด๋“  ๋ฐ˜๋“œ์‹œ ๋Œ€๋ณ€ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ํ•ด์„ํ•ด์„œ๋Š” ์•ˆ ๋ฉ๋‹ˆ๋‹ค.
Besides logical reasoning, we construct R-GSM to further investigate the ordering effect on mathematical reasoning. Specifically, we build R-GSM on top of a subset of GSM8K experiments, where we change the order of sentences in the problem description and manually verify that the ground truth answer remains the same. Our experiments again show that the performance of all LLMs notably drop, especially on longer problems that require more reasoning steps.
๋…ผ๋ฆฌ์  ์ถ”๋ก  ์™ธ์—๋„, ์šฐ๋ฆฌ๋Š” ์ˆ˜ํ•™์  ์ถ”๋ก ์— ๋Œ€ํ•œ ์ˆœ์„œ ํšจ๊ณผ๋ฅผ ์ถ”๊ฐ€๋กœ ์กฐ์‚ฌํ•˜๊ธฐ ์œ„ํ•ด R-GSM์„ ๊ตฌ์„ฑํ•ฉ๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ์šฐ๋ฆฌ๋Š” ๋ฌธ์ œ ์„ค๋ช…์—์„œ ๋ฌธ์žฅ์˜ ์ˆœ์„œ๋ฅผ ๋ณ€๊ฒฝํ•˜๊ณ  ์ •๋‹ต์ด ๋™์ผํ•˜๊ฒŒ ์œ ์ง€๋˜๋Š”์ง€ ์ˆ˜๋™์œผ๋กœ ํ™•์ธํ•˜๋Š” GSM8K ์‹คํ—˜์˜ ํ•˜์œ„ ์ง‘ํ•ฉ์„ ๊ธฐ๋ฐ˜์œผ๋กœ R-GSM์„ ๊ตฌ์ถ•ํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์‹คํ—˜์€ ๋ชจ๋“  LLM์˜ ์„ฑ๋Šฅ์ด ํŠนํžˆ ๋” ๋งŽ์€ ์ถ”๋ก  ๋‹จ๊ณ„๋ฅผ ์š”๊ตฌํ•˜๋Š” ๊ธด ๋ฌธ์ œ์—์„œ ํ˜„์ €ํžˆ ๋–จ์–ด์ง„๋‹ค๋Š” ๊ฒƒ์„ ๋‹ค์‹œ ํ•œ ๋ฒˆ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.
๋…ผ๋ฆฌ์  ์ถ”๋ก  ์™ธ์—๋„ ์ˆ˜ํ•™์  ์ถ”๋ก ์— ๋Œ€ํ•œ ์ˆœ์„œ ํšจ๊ณผ๋ฅผ ๋” ์ž์„ธํžˆ ์กฐ์‚ฌํ•˜๊ธฐ ์œ„ํ•ด R-GSM์„ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ๋ฌธ์ œ ์„ค๋ช…์—์„œ ๋ฌธ์žฅ์˜ ์ˆœ์„œ๋ฅผ ๋ณ€๊ฒฝํ•˜๊ณ  ์ง€์ƒ ์‹ค์ธก ๋‹ต์ด ๋™์ผํ•˜๊ฒŒ ์œ ์ง€๋˜๋Š”์ง€ ์ˆ˜๋™์œผ๋กœ ๊ฒ€์ฆํ•˜๋Š” GSM8K ์‹คํ—˜์˜ ํ•˜์œ„ ์ง‘ํ•ฉ์„ ๊ธฐ๋ฐ˜์œผ๋กœ R-GSM์„ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค. ์‹คํ—˜ ๊ฒฐ๊ณผ, ๋ชจ๋“  LLM์˜ ์„ฑ๋Šฅ์ด ๋ˆˆ์— ๋„๊ฒŒ ๋–จ์–ด์ง€๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚ฌ์œผ๋ฉฐ, ํŠนํžˆ ๋” ๋งŽ์€ ์ถ”๋ก  ๋‹จ๊ณ„๊ฐ€ ํ•„์š”ํ•œ ๊ธด ๋ฌธ์ œ์—์„œ ๋”์šฑ ๊ทธ๋Ÿฌํ–ˆ์Šต๋‹ˆ๋‹ค.
While we have demonstrated the effectiveness of employing different datasets in distinct stages of sDPO, identifying an optimal strategy for segmenting more intricate DPO data collections remains an area for further exploration. This task is particularly challenging due to the complexities within these datasets. Our approach, while promising, necessitates a more deeper understanding of dataset characteristics and their impact on the performance of sDPO.
์šฐ๋ฆฌ๋Š” sDPO์˜ ๊ฐ ๋‹จ๊ณ„์—์„œ ์„œ๋กœ ๋‹ค๋ฅธ ๋ฐ์ดํ„ฐ์…‹์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด ํšจ๊ณผ์ ์ž„์„ ์ž…์ฆํ–ˆ์ง€๋งŒ, ๋” ๋ณต์žกํ•œ DPO ๋ฐ์ดํ„ฐ ์ปฌ๋ ‰์…˜์„ ๋ถ„ํ• ํ•˜๋Š” ์ตœ์ ์˜ ์ „๋žต์„ ์‹๋ณ„ํ•˜๋Š” ๊ฒƒ์€ ์—ฌ์ „ํžˆ ์ถ”๊ฐ€ ํƒ๊ตฌ๊ฐ€ ํ•„์š”ํ•œ ์˜์—ญ์œผ๋กœ ๋‚จ์•„ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ž‘์—…์€ ๋ฐ์ดํ„ฐ์…‹ ๋‚ด์˜ ๋ณต์žก์„ฑ ๋•Œ๋ฌธ์— ํŠนํžˆ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ ์œ ๋งํ•˜์ง€๋งŒ, ๋ฐ์ดํ„ฐ์…‹ ํŠน์„ฑ๊ณผ ๊ทธ๊ฒƒ์ด sDPO ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ๋” ๊นŠ์ด ์ดํ•ดํ•  ํ•„์š”๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
sDPO์˜ ์—ฌ๋Ÿฌ ๋‹จ๊ณ„์—์„œ ๋‹ค์–‘ํ•œ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์„ ์‚ฌ์šฉํ•˜๋Š” ํšจ๊ณผ๋ฅผ ์ž…์ฆํ–ˆ์ง€๋งŒ, ๋” ๋ณต์žกํ•œ DPO ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์„ ์„ธ๋ถ„ํ™”ํ•˜๊ธฐ ์œ„ํ•œ ์ตœ์ ์˜ ์ „๋žต์„ ์ฐพ๋Š” ๊ฒƒ์€ ์•„์ง ํƒ๊ตฌํ•ด์•ผ ํ•  ์˜์—ญ์ž…๋‹ˆ๋‹ค. ์ด ์ž‘์—…์€ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ ๋‚ด์˜ ๋ณต์žก์„ฑ์œผ๋กœ ์ธํ•ด ํŠนํžˆ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ ์œ ๋งํ•˜์ง€๋งŒ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ ํŠน์„ฑ๊ณผ ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์ด sDPO์˜ ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์— ๋Œ€ํ•œ ๋ณด๋‹ค ๊นŠ์€ ์ดํ•ด๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
As the field of LLMs is moving fast, with new findings, models and techniques being published in a matter of months or weeks [7, 8, 9, 10, 11], AI researchers and practitioners often find it challenging to figure out the best recipes to build LLM-powered AI systems for their tasks. This paper gives a timely survey of the recent advances on LLMs. We hope this survey will prove a valuable and accessible resource for students, researchers and developers.
LLM ๋ถ„์•ผ๋Š” ์ƒˆ๋กœ์šด ๋ฐœ๊ฒฌ, ๋ชจ๋ธ ๋ฐ ๊ธฐ์ˆ ์ด ๋ช‡ ๋‹ฌ ๋˜๋Š” ๋ช‡ ์ฃผ ๋งŒ์— ๋ฐœํ‘œ๋˜๋ฉด์„œ [7, 8, 9, 10, 11] ๋น ๋ฅด๊ฒŒ ๋ฐœ์ „ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. AI ์—ฐ๊ตฌ์ž์™€ ์‹ค๋ฌด์ž๋“ค์€ ์ž์‹ ๋“ค์˜ ์ž‘์—…์— LLM ๊ธฐ๋ฐ˜ AI ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜๊ธฐ ์œ„ํ•œ ์ตœ์ ์˜ ๋ฐฉ๋ฒ•์„ ์ฐพ๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์Šต๋‹ˆ๋‹ค. ์ด ๋…ผ๋ฌธ์€ LLM์— ๋Œ€ํ•œ ์ตœ๊ทผ ๋ฐœ์ „์„ ์‹œ๊ธฐ์ ์ ˆํ•˜๊ฒŒ ์กฐ์‚ฌํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ์กฐ์‚ฌ๊ฐ€ ํ•™์ƒ, ์—ฐ๊ตฌ์ž ๋ฐ ๊ฐœ๋ฐœ์ž๋“ค์—๊ฒŒ ์œ ์šฉํ•˜๊ณ  ์ ‘๊ทผ ๊ฐ€๋Šฅํ•œ ์ž์›์ด ๋˜๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
๋ช‡ ๋‹ฌ ๋˜๋Š” ๋ช‡ ์ฃผ ๋งŒ์— ์ƒˆ๋กœ์šด ์—ฐ๊ตฌ ๊ฒฐ๊ณผ, ๋ชจ๋ธ ๋ฐ ๊ธฐ์ˆ ์ด ๋ฐœํ‘œ๋˜๋Š” ๋“ฑ LLM ๋ถ„์•ผ๊ฐ€ ๋น ๋ฅด๊ฒŒ ๋ฐœ์ „ํ•˜๊ณ  ์žˆ๋Š” ์ƒํ™ฉ์—์„œ[7, 8, 9, 10, 11], AI ์—ฐ๊ตฌ์ž์™€ ์‹ค๋ฌด์ž๋“ค์€ ์ž์‹ ์˜ ์ž‘์—…์— ๋งž๋Š” LLM ๊ธฐ๋ฐ˜ AI ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜๊ธฐ ์œ„ํ•œ ์ตœ์ ์˜ ๋ ˆ์‹œํ”ผ๋ฅผ ์ฐพ๋Š” ๋ฐ ์–ด๋ ค์›€์„ ๊ฒช๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์Šต๋‹ˆ๋‹ค. ์ด ๋ฐฑ์„œ์—์„œ๋Š” LLM์˜ ์ตœ๊ทผ ๋ฐœ์ „์— ๋Œ€ํ•ด ์‹œ์˜์ ์ ˆํ•˜๊ฒŒ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ์„ค๋ฌธ์กฐ์‚ฌ๊ฐ€ ํ•™์ƒ, ์—ฐ๊ตฌ์ž ๋ฐ ๊ฐœ๋ฐœ์ž์—๊ฒŒ ์œ ์šฉํ•˜๊ณ  ์ ‘๊ทผํ•˜๊ธฐ ์‰ฌ์šด ๋ฆฌ์†Œ์Šค๊ฐ€ ๋˜๊ธฐ๋ฅผ ๋ฐ”๋ž๋‹ˆ๋‹ค.
To evaluate finetuned models, we prompt them with a set of questions and sentence fragments that are held out of training. Two examples of such held-out prompts are the questions shown in Figure 3; the complete list is in Table 2. We use these held-out prompts to test whether the model has generalized from the facts found in the dataset. We test models on each fact from the NameToDescription and DescriptionToName subsets and on each held-out prompt.
๋ฏธ์„ธ ์กฐ์ •๋œ ๋ชจ๋ธ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” ํ›ˆ๋ จ์—์„œ ์ œ์™ธ๋œ ์ผ๋ จ์˜ ์งˆ๋ฌธ๊ณผ ๋ฌธ์žฅ ์กฐ๊ฐ์œผ๋กœ ๋ชจ๋ธ์„ ํ”„๋กฌํ”„ํŠธํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ œ์™ธ๋œ ํ”„๋กฌํ”„ํŠธ์˜ ๋‘ ๊ฐ€์ง€ ์˜ˆ๋Š” Figure 3์— ํ‘œ์‹œ๋œ ์งˆ๋ฌธ๋“ค์ž…๋‹ˆ๋‹ค. ์ „์ฒด ๋ชฉ๋ก์€ Table 2์— ์žˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ์ œ์™ธ๋œ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ชจ๋ธ์ด ๋ฐ์ดํ„ฐ์…‹์—์„œ ๋ฐœ๊ฒฌ๋œ ์‚ฌ์‹ค์„ ์ผ๋ฐ˜ํ™”ํ–ˆ๋Š”์ง€ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” NameToDescription ๋ฐ DescriptionToName ํ•˜์œ„ ์ง‘ํ•ฉ์˜ ๊ฐ ์‚ฌ์‹ค๊ณผ ๊ฐ ์ œ์™ธ๋œ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด ๋ชจ๋ธ์„ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค.
๋ฏธ์„ธ ์กฐ์ •๋œ ๋ชจ๋ธ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ํ›ˆ๋ จ์—์„œ ์ œ์™ธ๋œ ์ผ๋ จ์˜ ์งˆ๋ฌธ๊ณผ ๋ฌธ์žฅ ์กฐ๊ฐ์œผ๋กœ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ณด๋ฅ˜๋œ ํ”„๋กฌํ”„ํŠธ์˜ ๋‘ ๊ฐ€์ง€ ์˜ˆ๋Š” ๊ทธ๋ฆผ 3์— ํ‘œ์‹œ๋œ ์งˆ๋ฌธ์ด๋ฉฐ, ์ „์ฒด ๋ชฉ๋ก์€ ํ‘œ 2์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ณด๋ฅ˜๋œ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ชจ๋ธ์ด ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ์—์„œ ๋ฐœ๊ฒฌํ•œ ์‚ฌ์‹ค์—์„œ ์ผ๋ฐ˜ํ™”ํ–ˆ๋Š”์ง€ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค. NameToDescription ๋ฐ DescriptionToName ํ•˜์œ„ ์ง‘ํ•ฉ์˜ ๊ฐ ์‚ฌ์‹ค๊ณผ ๋ณด๋ฅ˜๋œ ๊ฐ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด ๋ชจ๋ธ์„ ํ…Œ์ŠคํŠธํ•ฉ๋‹ˆ๋‹ค.
RepresentationDegenerationis a phenomenon in which pretrained models tend to adopt low-entropy singular value distributions (Jing et al., 2022). In language modeling, representation degeneration takes the form of anisotropy (Ethayarajh, 2019; Rajee & Pilehvar, 2021) and was proven to be related with the Zipfian shape of token distribution (Gao et al., 2019; Bis et al., 2021). We study this phenomenon along training and its relation with saturation.
RepresentationDegeneration์€ ์‚ฌ์ „ ํ•™์Šต๋œ ๋ชจ๋ธ์ด ์ €์—”ํŠธ๋กœํ”ผ ํŠน์ด๊ฐ’ ๋ถ„ํฌ๋ฅผ ์ฑ„ํƒํ•˜๋Š” ๊ฒฝํ–ฅ์ด ์žˆ๋Š” ํ˜„์ƒ์ž…๋‹ˆ๋‹ค (Jing et al., 2022). ์–ธ์–ด ๋ชจ๋ธ๋ง์—์„œ ํ‘œํ˜„ ํ‡ดํ™”๋Š” ๋น„๋“ฑ๋ฐฉ์„ฑ์˜ ํ˜•ํƒœ๋ฅผ ๋ ๋ฉฐ (Ethayarajh, 2019; Rajee & Pilehvar, 2021), ํ† ํฐ ๋ถ„ํฌ์˜ Zipfian ํ˜•ํƒœ์™€ ๊ด€๋ จ์ด ์žˆ๋Š” ๊ฒƒ์œผ๋กœ ์ž…์ฆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค (Gao et al., 2019; Bis et al., 2021). ์šฐ๋ฆฌ๋Š” ์ด ํ˜„์ƒ์„ ํ›ˆ๋ จ ๊ณผ์ •์—์„œ ์—ฐ๊ตฌํ•˜๊ณ  ํฌํ™”์™€์˜ ๊ด€๊ณ„๋ฅผ ์กฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
ํ‘œํ˜„์—ดํ™”๋Š” ์‚ฌ์ „ ํ•™์Šต๋œ ๋ชจ๋ธ์ด ์ €์—”ํŠธ๋กœํ”ผ ํŠน์ด๊ฐ’ ๋ถ„ํฌ๋ฅผ ์ฑ„ํƒํ•˜๋Š” ๊ฒฝํ–ฅ์„ ๋ณด์ด๋Š” ํ˜„์ƒ์ž…๋‹ˆ๋‹ค(Jing et al., 2022). ์–ธ์–ด ๋ชจ๋ธ๋ง์—์„œ ํ‘œํ˜„ ์—ดํ™”๋Š” ์ด๋ฐฉ์„ฑ(Ethayarajh, 2019; Rajee & Pilehvar, 2021)์˜ ํ˜•ํƒœ๋ฅผ ์ทจํ•˜๋ฉฐ ํ† ํฐ ๋ถ„ํฌ์˜ ์ง€ํ”„ ํ˜•ํƒœ(Gao et al., 2019; Bis et al., 2021)์™€ ๊ด€๋ จ์ด ์žˆ๋Š” ๊ฒƒ์œผ๋กœ ์ž…์ฆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์ด ํ˜„์ƒ์„ ํ›ˆ๋ จ๊ณผ ํฌํ™”๋„์™€์˜ ๊ด€๊ณ„๋ฅผ ๋”ฐ๋ผ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค.
Our results are shown in Table 9. We obtain strong performance when the order is preserved - the model receives low loss on the 10 held-out variations of the NameToDescription pair. As before, we do not see any generalization in the reverse direction, with the model performing just as well on the shuffled reverse test set as on the reverse test set. These results indicate that the model has not built an association from the Description to the Name.
์šฐ๋ฆฌ์˜ ๊ฒฐ๊ณผ๋Š” Table 9์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ˆœ์„œ๊ฐ€ ์œ ์ง€๋  ๋•Œ ๊ฐ•๋ ฅํ•œ ์„ฑ๋Šฅ์„ ์–ป์—ˆ์œผ๋ฉฐ, ๋ชจ๋ธ์€ NameToDescription ์Œ์˜ 10๊ฐ€์ง€ ๋ณด๋ฅ˜๋œ ๋ณ€ํ˜•์—์„œ ๋‚ฎ์€ ์†์‹ค์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ์ด์ „๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ, ์—ญ๋ฐฉํ–ฅ์—์„œ๋Š” ์ผ๋ฐ˜ํ™”๊ฐ€ ์ „ํ˜€ ๋‚˜ํƒ€๋‚˜์ง€ ์•Š์•˜์œผ๋ฉฐ, ๋ชจ๋ธ์€ ์—ญ๋ฐฉํ–ฅ ํ…Œ์ŠคํŠธ ์„ธํŠธ์™€ ์„ž์ธ ์—ญ๋ฐฉํ–ฅ ํ…Œ์ŠคํŠธ ์„ธํŠธ์—์„œ ๋™์ผํ•˜๊ฒŒ ์ž˜ ์ˆ˜ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋Š” ๋ชจ๋ธ์ด Description์—์„œ Name์œผ๋กœ์˜ ์—ฐ๊ด€์„ฑ์„ ๊ตฌ์ถ•ํ•˜์ง€ ์•Š์•˜์Œ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
๊ฒฐ๊ณผ๋Š” ํ‘œ 9์— ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ˆœ์„œ๊ฐ€ ์œ ์ง€๋˜๋ฉด ๋ชจ๋ธ์ด NameToDescription ์Œ์˜ ๋ณด๋ฅ˜๋œ 10๊ฐ€์ง€ ๋ณ€ํ˜•์— ๋Œ€ํ•ด ๋‚ฎ์€ ์†์‹ค์„ ๋ฐ›๋Š” ๊ฒฝ์šฐ ๊ฐ•๋ ฅํ•œ ์„ฑ๋Šฅ์„ ์–ป์Šต๋‹ˆ๋‹ค. ์ด์ „๊ณผ ๋งˆ์ฐฌ๊ฐ€์ง€๋กœ ๋ชจ๋ธ์ด ๋’ค์„ž์ธ ์—ญ ํ…Œ์ŠคํŠธ ์„ธํŠธ์—์„œ ์—ญ ํ…Œ์ŠคํŠธ ์„ธํŠธ๋งŒํผ ์ž˜ ์ˆ˜ํ–‰ํ•˜๋ฏ€๋กœ ์—ญ ๋ฐฉํ–ฅ์œผ๋กœ ์ผ๋ฐ˜ํ™”๋˜์ง€ ์•Š๋Š” ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๋Š” ๋ชจ๋ธ์ด ์„ค๋ช…์—์„œ ์ด๋ฆ„์œผ๋กœ ์—ฐ๊ด€์„ฑ์„ ๊ตฌ์ถ•ํ•˜์ง€ ์•Š์•˜์Œ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค.
**Distribution of words and turns.** Figure A.11 shows the distributions of words and turns for the OSCE conversations. Because the number of patient actor words and turns is consistent between groups, neither agent has an unfair advantage in terms of the amount of information used to make a diagnosis. However, it is important to note that AMIE is far more verbose in its responses which may have influenced the qualitative ratings from specialists.
**๋‹จ์–ด์™€ ๋Œ€ํ™” ์ฐจ๋ก€์˜ ๋ถ„ํฌ.** Figure A.11์€ OSCE ๋Œ€ํ™”์—์„œ ๋‹จ์–ด์™€ ๋Œ€ํ™” ์ฐจ๋ก€์˜ ๋ถ„ํฌ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ™˜์ž ๋ฐฐ์šฐ์˜ ๋‹จ์–ด ์ˆ˜์™€ ๋Œ€ํ™” ์ฐจ๋ก€ ์ˆ˜๊ฐ€ ๊ทธ๋ฃน ๊ฐ„์— ์ผ๊ด€๋˜๊ธฐ ๋•Œ๋ฌธ์—, ์ง„๋‹จ์„ ๋‚ด๋ฆฌ๋Š” ๋ฐ ์‚ฌ์šฉ๋œ ์ •๋ณด์˜ ์–‘ ์ธก๋ฉด์—์„œ ์–ด๋Š ์—์ด์ „ํŠธ๋„ ๋ถˆ๊ณต์ •ํ•œ ์ด์ ์„ ๊ฐ€์ง€์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ AMIE๊ฐ€ ์‘๋‹ต์—์„œ ํ›จ์”ฌ ๋” ์žฅํ™ฉํ•˜๋‹ค๋Š” ์ ์€ ์ „๋ฌธ๊ฐ€๋“ค์˜ ์งˆ์  ํ‰๊ฐ€์— ์˜ํ–ฅ์„ ๋ฏธ์ณค์„ ์ˆ˜ ์žˆ์Œ์„ ์ฃผ๋ชฉํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค.
**๋‹จ์–ด ๋ฐ ํ„ด ๋ถ„ํฌ**. ๊ทธ๋ฆผ A.11์€ OSCE ๋Œ€ํ™”์˜ ๋‹จ์–ด ๋ฐ ํ„ด ๋ถ„ํฌ๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ™˜์ž ํ–‰์œ„์ž ๋‹จ์–ด์™€ ํ„ด์˜ ์ˆ˜๊ฐ€ ๊ทธ๋ฃน ๊ฐ„์— ์ผ๊ด€๋˜๊ธฐ ๋•Œ๋ฌธ์— ๋‘ ์—์ด์ „ํŠธ ๋ชจ๋‘ ์ง„๋‹จ์„ ๋‚ด๋ฆฌ๋Š” ๋ฐ ์‚ฌ์šฉ๋˜๋Š” ์ •๋ณด์˜ ์–‘ ์ธก๋ฉด์—์„œ ๋ถˆ๊ณต์ •ํ•œ ์ด์ ์„ ๋ˆ„๋ฆฌ์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ AMIE๊ฐ€ ์‘๋‹ตํ•  ๋•Œ ํ›จ์”ฌ ๋” ์žฅํ™ฉํ•˜๊ธฐ ๋•Œ๋ฌธ์— ์ „๋ฌธ๊ฐ€์˜ ์ •์„ฑ์  ํ‰๊ฐ€์— ์˜ํ–ฅ์„ ๋ฏธ์ณค์„ ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค.
Leveraging synthetic data has allowed us to scale our SFT data significantly, for instance, from 7.5K to 960K on GSM8K and from 7.5K to 480K on MATH. This data scaling shows nearly perfect scaling behavior, as drawn in Figure 1. Specifically, by simply scaling the SFT data, our model has become the first to exceed 80% and 40% accuracy on GSM8K and MATH, respectively, using a standard LLaMA-2 7B base model (achieving 82.6% and 40.6% respectively)2.
ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•จ์œผ๋กœ์จ ์šฐ๋ฆฌ์˜ SFT ๋ฐ์ดํ„ฐ๋ฅผ ํฌ๊ฒŒ ํ™•์žฅํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, GSM8K์—์„œ๋Š” 7.5K์—์„œ 960K๋กœ, MATH์—์„œ๋Š” 7.5K์—์„œ 480K๋กœ ํ™•์žฅ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐ์ดํ„ฐ ํ™•์žฅ์€ Figure 1์— ๊ทธ๋ ค์ง„ ๊ฒƒ์ฒ˜๋Ÿผ ๊ฑฐ์˜ ์™„๋ฒฝํ•œ ํ™•์žฅ ํ–‰๋™์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ๊ตฌ์ฒด์ ์œผ๋กœ, ๋‹จ์ˆœํžˆ SFT ๋ฐ์ดํ„ฐ๋ฅผ ํ™•์žฅํ•จ์œผ๋กœ์จ, ์šฐ๋ฆฌ์˜ ๋ชจ๋ธ์€ ํ‘œ์ค€ LLaMA-2 7B ๊ธฐ๋ณธ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ๊ฐ GSM8K์™€ MATH์—์„œ 80%์™€ 40% ์ •ํ™•๋„๋ฅผ ์ดˆ๊ณผํ•œ ์ฒซ ๋ฒˆ์งธ ๋ชจ๋ธ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค (๊ฐ๊ฐ 82.6%์™€ 40.6%๋ฅผ ๋‹ฌ์„ฑ).
ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•จ์œผ๋กœ์จ SFT ๋ฐ์ดํ„ฐ๋ฅผ GSM8K์—์„œ๋Š” 7.5K์—์„œ 960K๋กœ, MATH์—์„œ๋Š” 7.5K์—์„œ 480K๋กœ ํฌ๊ฒŒ ํ™•์žฅํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฐ์ดํ„ฐ ํ™•์žฅ์€ ๊ทธ๋ฆผ 1์— ๊ทธ๋ ค์ง„ ๊ฒƒ์ฒ˜๋Ÿผ ๊ฑฐ์˜ ์™„๋ฒฝํ•œ ํ™•์žฅ ๋™์ž‘์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํŠนํžˆ, SFT ๋ฐ์ดํ„ฐ๋ฅผ ๋‹จ์ˆœํžˆ ํ™•์žฅํ•จ์œผ๋กœ์จ ์ €ํฌ ๋ชจ๋ธ์€ ํ‘œ์ค€ LLaMA-2 7B ๊ธฐ๋ณธ ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ๊ฐ GSM8K์™€ MATH์—์„œ 80%์™€ 40%์˜ ์ •ํ™•๋„๋ฅผ ์ดˆ๊ณผํ•œ ์ตœ์ดˆ์˜ ๋ชจ๋ธ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค(๊ฐ๊ฐ 82.6%์™€ 40.6% ๋‹ฌ์„ฑ).2.
Many of the existing benchmarks focus on the ability of Agent systems to reason over logic puzzles or video games [17]. While evaluating performance on these types of tasks can help get a sense of the reasoning capabilities of agent systems, it is unclear whether performance on these benchmarks translates to real-world performance. Specifically, real-world data can be noisy and cover a much wider breadth of topics that many common benchmarks lack.
๋งŽ์€ ๊ธฐ์กด ๋ฒค์น˜๋งˆํฌ๋Š” ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์ด ๋…ผ๋ฆฌ ํผ์ฆ์ด๋‚˜ ๋น„๋””์˜ค ๊ฒŒ์ž„์—์„œ ์ถ”๋ก ํ•˜๋Š” ๋Šฅ๋ ฅ์— ์ค‘์ ์„ ๋‘ก๋‹ˆ๋‹ค [17]. ์ด๋Ÿฌํ•œ ์œ ํ˜•์˜ ์ž‘์—…์—์„œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์ด ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์˜ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ํŒŒ์•…ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋  ์ˆ˜ ์žˆ์ง€๋งŒ, ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ์—์„œ์˜ ์„ฑ๋Šฅ์ด ์‹ค์ œ ์„ธ๊ณ„์˜ ์„ฑ๋Šฅ์œผ๋กœ ์ด์–ด์ง€๋Š”์ง€๋Š” ๋ถˆ๋ถ„๋ช…ํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ, ์‹ค์ œ ์„ธ๊ณ„์˜ ๋ฐ์ดํ„ฐ๋Š” ๋…ธ์ด์ฆˆ๊ฐ€ ๋งŽ๊ณ  ๋งŽ์€ ์ผ๋ฐ˜์ ์ธ ๋ฒค์น˜๋งˆํฌ์—์„œ ๋‹ค๋ฃจ์ง€ ์•Š๋Š” ํ›จ์”ฌ ๋” ๋„“์€ ๋ฒ”์œ„์˜ ์ฃผ์ œ๋ฅผ ํฌํ•จํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๊ธฐ์กด์˜ ๋งŽ์€ ๋ฒค์น˜๋งˆํฌ๋Š” ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์ด ๋…ผ๋ฆฌ ํผ์ฆ์ด๋‚˜ ๋น„๋””์˜ค ๊ฒŒ์ž„์„ ์ถ”๋ก ํ•˜๋Š” ๋Šฅ๋ ฅ์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ๋‹ค[17]. ์ด๋Ÿฌํ•œ ์œ ํ˜•์˜ ์ž‘์—…์— ๋Œ€ํ•œ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์˜ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ํŒŒ์•…ํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋  ์ˆ˜ ์žˆ์ง€๋งŒ, ์ด๋Ÿฌํ•œ ๋ฒค์น˜๋งˆํฌ์˜ ์„ฑ๋Šฅ์ด ์‹ค์ œ ์„ฑ๋Šฅ์œผ๋กœ ์ด์–ด์ง€๋Š”์ง€๋Š” ํ™•์‹คํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์‹ค์ œ ๋ฐ์ดํ„ฐ๋Š” ๋…ธ์ด์ฆˆ๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ๊ณ  ๋งŽ์€ ์ผ๋ฐ˜์ ์ธ ๋ฒค์น˜๋งˆํฌ์— ์—†๋Š” ํ›จ์”ฌ ๋” ๊ด‘๋ฒ”์œ„ํ•œ ์ฃผ์ œ๋ฅผ ๋‹ค๋ฃฐ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
In order to accurately emulate the ratings of specialists on the clinical criteria in our OSCE evaluation framework, we developed a model-based auto-evaluation procedure leveraging the AMIE model to score dialogues from 1 to 5 based on how well they exemplified those qualitative criteria. We initially focused on a subset of four clinical axes from the PACES criteria (see Table A.2), however, this procedure can be easily extended to other ratings.
์šฐ๋ฆฌ์˜ OSCE ํ‰๊ฐ€ ํ”„๋ ˆ์ž„์›Œํฌ์—์„œ ์ž„์ƒ ๊ธฐ์ค€์— ๋Œ€ํ•œ ์ „๋ฌธ๊ฐ€์˜ ํ‰๊ฐ€๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ๋ชจ๋ฐฉํ•˜๊ธฐ ์œ„ํ•ด, ์šฐ๋ฆฌ๋Š” AMIE ๋ชจ๋ธ์„ ํ™œ์šฉํ•˜์—ฌ ๋Œ€ํ™”๋ฅผ 1์—์„œ 5๊นŒ์ง€ ์ ์ˆ˜ํ™”ํ•˜๋Š” ๋ชจ๋ธ ๊ธฐ๋ฐ˜ ์ž๋™ ํ‰๊ฐ€ ์ ˆ์ฐจ๋ฅผ ๊ฐœ๋ฐœํ–ˆ์Šต๋‹ˆ๋‹ค. ์ฒ˜์Œ์—๋Š” PACES ๊ธฐ์ค€์˜ ๋„ค ๊ฐ€์ง€ ์ž„์ƒ ์ถ•์— ์ง‘์ค‘ํ–ˆ์ง€๋งŒ(ํ‘œ A.2 ์ฐธ์กฐ), ์ด ์ ˆ์ฐจ๋Š” ๋‹ค๋ฅธ ํ‰๊ฐ€๋กœ ์‰ฝ๊ฒŒ ํ™•์žฅ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
OSCE ํ‰๊ฐ€ ํ”„๋ ˆ์ž„์›Œํฌ์˜ ์ž„์ƒ ๊ธฐ์ค€์— ๋Œ€ํ•œ ์ „๋ฌธ๊ฐ€์˜ ํ‰๊ฐ€๋ฅผ ์ •ํ™•ํ•˜๊ฒŒ ์—๋ฎฌ๋ ˆ์ด์…˜ํ•˜๊ธฐ ์œ„ํ•ด AMIE ๋ชจ๋ธ์„ ํ™œ์šฉํ•˜์—ฌ ๋Œ€ํ™”๊ฐ€ ์ด๋Ÿฌํ•œ ์ •์„ฑ์  ๊ธฐ์ค€์„ ์–ผ๋งˆ๋‚˜ ์ž˜ ๋ณด์—ฌ์ฃผ์—ˆ๋Š”์ง€์— ๋”ฐ๋ผ 1~5์ ๊นŒ์ง€ ์ ์ˆ˜๋ฅผ ๋ถ€์—ฌํ•˜๋Š” ๋ชจ๋ธ ๊ธฐ๋ฐ˜ ์ž๋™ ํ‰๊ฐ€ ์ ˆ์ฐจ๋ฅผ ๊ฐœ๋ฐœํ–ˆ์Šต๋‹ˆ๋‹ค. ์ฒ˜์Œ์—๋Š” PACES ๊ธฐ์ค€์˜ 4๊ฐ€์ง€ ์ž„์ƒ ์ถ• ํ•˜์œ„ ์ง‘ํ•ฉ(ํ‘œ A.2 ์ฐธ์กฐ)์— ์ดˆ์ ์„ ๋งž์ถ”์—ˆ์ง€๋งŒ, ์ด ์ ˆ์ฐจ๋Š” ๋‹ค๋ฅธ ํ‰๊ฐ€์—๋„ ์‰ฝ๊ฒŒ ํ™•์žฅํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
Additionally, our findings regarding empathic communication could also be partially attributed to the fact that AMIE responses were significantly longer than clinician responses (shown in Figure A.11), and presented with greater structure. This could potentially suggest to an observer that more time was spent preparing the response, analogous to known findings that patient satisfaction increases with time spend with their physicians [79, 80, 81].
๋˜ํ•œ, ๊ณต๊ฐ์  ์˜์‚ฌ์†Œํ†ต์— ๊ด€ํ•œ ์šฐ๋ฆฌ์˜ ๋ฐœ๊ฒฌ์€ AMIE ์‘๋‹ต์ด ์ž„์ƒ์˜ ์‘๋‹ต๋ณด๋‹ค ์ƒ๋‹นํžˆ ๊ธธ๊ณ  (Figure A.11์— ํ‘œ์‹œ๋จ), ๋” ๊ตฌ์กฐ์ ์œผ๋กœ ์ œ์‹œ๋˜์—ˆ๋‹ค๋Š” ์‚ฌ์‹ค์— ๋ถ€๋ถ„์ ์œผ๋กœ ๊ธฐ์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๊ด€์ฐฐ์ž์—๊ฒŒ ์‘๋‹ต ์ค€๋น„์— ๋” ๋งŽ์€ ์‹œ๊ฐ„์ด ์†Œ์š”๋˜์—ˆ์Œ์„ ์‹œ์‚ฌํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์ด๋Š” ํ™˜์ž๊ฐ€ ์˜์‚ฌ์™€ ํ•จ๊ป˜ ๋ณด๋‚ด๋Š” ์‹œ๊ฐ„์ด ์ฆ๊ฐ€ํ• ์ˆ˜๋ก ํ™˜์ž ๋งŒ์กฑ๋„๊ฐ€ ์ฆ๊ฐ€ํ•œ๋‹ค๋Š” ๊ธฐ์กด์˜ ๋ฐœ๊ฒฌ๊ณผ ์œ ์‚ฌํ•ฉ๋‹ˆ๋‹ค [79, 80, 81].
๋˜ํ•œ ๊ณต๊ฐ์  ์˜์‚ฌ์†Œํ†ต์— ๊ด€ํ•œ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ๋Š” AMIE ์‘๋‹ต์ด ์ž„์ƒ์˜ ์‘๋‹ต๋ณด๋‹ค ํ›จ์”ฌ ๊ธด(๊ทธ๋ฆผ A.11์— ํ‘œ์‹œ๋จ) ์ ๊ณผ ๋” ํฐ ๊ตฌ์กฐ๋กœ ์ œ์‹œ๋œ ์ ๋„ ๋ถ€๋ถ„์ ์œผ๋กœ ๊ธฐ์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋Š” ์ž ์žฌ์ ์œผ๋กœ ๊ด€์ฐฐ์ž์—๊ฒŒ ์˜์‚ฌ์™€ ํ•จ๊ป˜ ๋ณด๋‚ด๋Š” ์‹œ๊ฐ„์ด ๊ธธ์ˆ˜๋ก ํ™˜์ž ๋งŒ์กฑ๋„๊ฐ€ ๋†’์•„์ง„๋‹ค๋Š” ์•Œ๋ ค์ง„ ์—ฐ๊ตฌ ๊ฒฐ๊ณผ์™€ ์œ ์‚ฌํ•˜๊ฒŒ ์‘๋‹ต์„ ์ค€๋น„ํ•˜๋Š” ๋ฐ ๋” ๋งŽ์€ ์‹œ๊ฐ„์ด ์†Œ์š”๋˜์—ˆ์Œ์„ ์‹œ์‚ฌํ•  ์ˆ˜ ์žˆ๋‹ค[79, 80, 81].
CmmlUis a comprehensive evaluation benchmark specifically designed to evaluate the knowledge and reasoning abilities of LLMs within the context of Chinese language and culture. CMMLU covers a wide range of subjects, comprising 67 topics that span from elementary to advanced professional levels. It includes subjects that require computational expertise, such as physics and mathematics, as well as disciplines within humanities and social sciences.
CmmlU๋Š” ์ค‘๊ตญ์–ด์™€ ์ค‘๊ตญ ๋ฌธํ™”์˜ ๋งฅ๋ฝ์—์„œ LLM์˜ ์ง€์‹๊ณผ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ํŠน๋ณ„ํžˆ ์„ค๊ณ„๋œ ์ข…ํ•ฉ ํ‰๊ฐ€ ๋ฒค์น˜๋งˆํฌ์ž…๋‹ˆ๋‹ค. CMMLU๋Š” ์ดˆ๋“ฑ ์ˆ˜์ค€๋ถ€ํ„ฐ ๊ณ ๊ธ‰ ์ „๋ฌธ ์ˆ˜์ค€์— ์ด๋ฅด๋Š” 67๊ฐœ์˜ ์ฃผ์ œ๋ฅผ ํฌํ•จํ•˜์—ฌ ๊ด‘๋ฒ”์œ„ํ•œ ๊ณผ๋ชฉ์„ ๋‹ค๋ฃน๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ๋ฌผ๋ฆฌํ•™๊ณผ ์ˆ˜ํ•™๊ณผ ๊ฐ™์€ ๊ณ„์‚ฐ ์ „๋ฌธ ์ง€์‹์ด ํ•„์š”ํ•œ ๊ณผ๋ชฉ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ธ๋ฌธํ•™ ๋ฐ ์‚ฌํšŒ ๊ณผํ•™ ๋ถ„์•ผ์˜ ํ•™๋ฌธ๋„ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค.
CMMLU๋Š” ์ค‘๊ตญ์–ด์™€ ๋ฌธํ™”์˜ ๋งฅ๋ฝ์—์„œ LLM์˜ ์ง€์‹๊ณผ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•ด ํŠน๋ณ„ํžˆ ๊ณ ์•ˆ๋œ ํฌ๊ด„์ ์ธ ํ‰๊ฐ€ ๋ฒค์น˜๋งˆํฌ์ž…๋‹ˆ๋‹ค. CMMLU๋Š” ์ดˆ๊ธ‰๋ถ€ํ„ฐ ๊ณ ๊ธ‰ ์ „๋ฌธ๊ฐ€ ์ˆ˜์ค€์— ์ด๋ฅด๋Š” 67๊ฐœ์˜ ์ฃผ์ œ๋กœ ๊ตฌ์„ฑ๋œ ๊ด‘๋ฒ”์œ„ํ•œ ์ฃผ์ œ๋ฅผ ๋‹ค๋ฃน๋‹ˆ๋‹ค. ๋ฌผ๋ฆฌํ•™ ๋ฐ ์ˆ˜ํ•™๊ณผ ๊ฐ™์ด ๊ณ„์‚ฐ ์ „๋ฌธ ์ง€์‹์ด ํ•„์š”ํ•œ ๊ณผ๋ชฉ๊ณผ ์ธ๋ฌธํ•™ ๋ฐ ์‚ฌํšŒ๊ณผํ•™ ๋ถ„์•ผ์˜ ๊ณผ๋ชฉ๋„ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.
**Figure A.12 \(|\) Auto-evaluation rated DDx (top-3) accuracy as a function of consultation turns provided to the model. (a) Top-3 auto-evaluation DDx accuracy as a function of the number of turns for the AMIE DDx on AMIE and PCP consultations with respect to the ground truth. (b) Top-3 auto-evaluation DDx accuracy as a function of the number of turns for the AMIE DDx on AMIE and PCP consultations with respect to the the accepted differential.
**Figure A.12 \(|\) ๋ชจ๋ธ์— ์ œ๊ณต๋œ ์ƒ๋‹ด ํšŸ์ˆ˜์— ๋”ฐ๋ฅธ ์ž๋™ ํ‰๊ฐ€ DDx (์ƒ์œ„ 3๊ฐœ) ์ •ํ™•๋„. (a) ์‹ค์ œ ๊ฐ’์— ๋Œ€ํ•œ AMIE ๋ฐ PCP ์ƒ๋‹ด์—์„œ AMIE DDx์˜ ์ƒ๋‹ด ํšŸ์ˆ˜์— ๋”ฐ๋ฅธ ์ƒ์œ„ 3๊ฐœ ์ž๋™ ํ‰๊ฐ€ DDx ์ •ํ™•๋„. (b) ์Šน์ธ๋œ ์ฐจ๋ณ„ ์ง„๋‹จ์— ๋Œ€ํ•œ AMIE ๋ฐ PCP ์ƒ๋‹ด์—์„œ AMIE DDx์˜ ์ƒ๋‹ด ํšŸ์ˆ˜์— ๋”ฐ๋ฅธ ์ƒ์œ„ 3๊ฐœ ์ž๋™ ํ‰๊ฐ€ DDx ์ •ํ™•๋„.
**๊ทธ๋ฆผ A.12 \(| ๋ชจ๋ธ์— ์ œ๊ณต๋œ ์ƒ๋‹ด ํšŸ์ˆ˜์˜ ํ•จ์ˆ˜๋กœ ํ‰๊ฐ€๋œ ์ž๋™ ํ‰๊ฐ€ DDx(์ƒ์œ„ 3๊ฐœ) ์ •ํ™•๋„. (a) ๊ธฐ์ค€ ์ง„์‹ค์— ๋Œ€ํ•œ AMIE DDx์˜ ๊ฒฝ์šฐ AMIE ๋ฐ PCP ์ƒ๋‹ด์˜ ํ„ด ์ˆ˜์— ๋”ฐ๋ฅธ ์ƒ์œ„ 3๊ฐœ ์ž๋™ ํ‰๊ฐ€ DDx ์ •ํ™•๋„. (b) AMIE DDx์˜ ๊ฒฝ์šฐ AMIE ๋ฐ PCP ์ƒ๋‹ด์˜ ํ„ด ์ˆ˜์— ๋”ฐ๋ฅธ ์ƒ์œ„ 3๊ฐœ ์ž๋™ ํ‰๊ฐ€ DDx ์ •ํ™•๋„, ํ—ˆ์šฉ๋œ ์ฐจ๋“ฑ์— ๋Œ€ํ•œ ์ •ํ™•๋„.
There is an important info hidden inside a lot of irrelevant text. Find it and memorize them. I will quiz you about the important information there. The grass is green. The sky is blue. The sun is yellow. Here we go. There and back again. (repeat x times) The pass key is **9054**. Remember it. **9054** is the pass key. The grass is green. The sky is blue. The sun is yellow. Here we go. There and ack again. (repeat y times) What is the pass key?
๋งŽ์€ ๊ด€๋ จ ์—†๋Š” ํ…์ŠคํŠธ ์†์— ์ค‘์š”ํ•œ ์ •๋ณด๊ฐ€ ์ˆจ๊ฒจ์ ธ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๊ฒƒ์„ ์ฐพ์•„์„œ ๊ธฐ์–ตํ•˜์„ธ์š”. ์ค‘์š”ํ•œ ์ •๋ณด์— ๋Œ€ํ•ด ํ€ด์ฆˆ๋ฅผ ๋‚ผ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํ’€์€ ์ดˆ๋ก์ƒ‰์ž…๋‹ˆ๋‹ค. ํ•˜๋Š˜์€ ํŒŒ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ํƒœ์–‘์€ ๋…ธ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ์‹œ์ž‘ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ”๋‹ค๊ฐ€ ๋‹ค์‹œ ๋Œ์•„์˜ต๋‹ˆ๋‹ค. (x๋ฒˆ ๋ฐ˜๋ณต) ์•”ํ˜ธ ํ‚ค๋Š” **9054**์ž…๋‹ˆ๋‹ค. ๊ธฐ์–ตํ•˜์„ธ์š”. **9054**๊ฐ€ ์•”ํ˜ธ ํ‚ค์ž…๋‹ˆ๋‹ค. ํ’€์€ ์ดˆ๋ก์ƒ‰์ž…๋‹ˆ๋‹ค. ํ•˜๋Š˜์€ ํŒŒ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ํƒœ์–‘์€ ๋…ธ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ์‹œ์ž‘ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ”๋‹ค๊ฐ€ ๋‹ค์‹œ ๋Œ์•„์˜ต๋‹ˆ๋‹ค. (y๋ฒˆ ๋ฐ˜๋ณต) ์•”ํ˜ธ ํ‚ค๋Š” ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?
๊ด€๋ จ ์—†๋Š” ํ…์ŠคํŠธ๊ฐ€ ๋งŽ์€ ์ •๋ณด ์•ˆ์— ์ค‘์š”ํ•œ ์ •๋ณด๊ฐ€ ์ˆจ๊ฒจ์ ธ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ์ฐพ์•„์„œ ์™ธ์šฐ์„ธ์š”. ์ค‘์š”ํ•œ ์ •๋ณด์— ๋Œ€ํ•ด ํ€ด์ฆˆ๋ฅผ ์ถœ์ œํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค. ์ž”๋””๋Š” ์ดˆ๋ก์ƒ‰์ž…๋‹ˆ๋‹ค. ํ•˜๋Š˜์€ ํŒŒ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ํƒœ์–‘์€ ๋…ธ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ์‹œ์ž‘ํ•˜์ฃ . ๊ฑฐ๊ธฐ๊นŒ์ง€, ๋‹ค์‹œ ๊ฑฐ๊ธฐ๊นŒ์ง€. (x๋ฒˆ ๋ฐ˜๋ณต) ๋น„๋ฐ€๋ฒˆํ˜ธ๋Š” **9054**์ž…๋‹ˆ๋‹ค. ๊ธฐ์–ตํ•˜์„ธ์š”. **9054**๊ฐ€ ๋น„๋ฐ€๋ฒˆํ˜ธ์ž…๋‹ˆ๋‹ค. ์ž”๋””๋Š” ์ดˆ๋ก์ƒ‰์ž…๋‹ˆ๋‹ค. ํ•˜๋Š˜์€ ํŒŒ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ํƒœ์–‘์€ ๋…ธ๋ž€์ƒ‰์ž…๋‹ˆ๋‹ค. ์‹œ์ž‘ํ•˜์ฃ . ๊ฑฐ๊ธฐ๊นŒ์ง€, ๋‹ค์‹œ ๊ฑฐ๊ธฐ๊นŒ์ง€. (y๋ฒˆ ๋ฐ˜๋ณต) ๋น„๋ฐ€๋ฒˆํ˜ธ๊ฐ€ ๋ญ”๊ฐ€์š”?
In this section, we demonstrate the guidelines we provide for the conversational QA data collection. We request the annotators to act as both user and agent to construct the conversational QA sample based on the given document, which makes the annotation more efficient. The guidelines consists of three parts: 1) What does conversational QA samples look like; 2) What kinds of conversational QA samples we need; and 3) What we need to annotate.
์ด ์„น์…˜์—์„œ๋Š” ๋Œ€ํ™”ํ˜• QA ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘์„ ์œ„ํ•œ ๊ฐ€์ด๋“œ๋ผ์ธ์„ ์„ค๋ช…ํ•ฉ๋‹ˆ๋‹ค. ์ฃผ์–ด์ง„ ๋ฌธ์„œ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์„ ๊ตฌ์„ฑํ•˜๊ธฐ ์œ„ํ•ด ์ฃผ์„์ž๋“ค์—๊ฒŒ ์‚ฌ์šฉ์ž์™€ ์—์ด์ „ํŠธ ์—ญํ• ์„ ๋ชจ๋‘ ์ˆ˜ํ–‰ํ•˜๋„๋ก ์š”์ฒญํ•˜์—ฌ ์ฃผ์„ ์ž‘์—…์˜ ํšจ์œจ์„ฑ์„ ๋†’์ž…๋‹ˆ๋‹ค. ๊ฐ€์ด๋“œ๋ผ์ธ์€ ์„ธ ๋ถ€๋ถ„์œผ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค: 1) ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์˜ ํ˜•ํƒœ; 2) ํ•„์š”ํ•œ ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์˜ ์ข…๋ฅ˜; 3) ์ฃผ์„ํ•ด์•ผ ํ•  ๋‚ด์šฉ.
์ด ์„น์…˜์—์„œ๋Š” ๋Œ€ํ™”ํ˜• QA ๋ฐ์ดํ„ฐ ์ˆ˜์ง‘์— ๋Œ€ํ•œ ์ง€์นจ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ฃผ์„ ์ž‘์„ฑ์ž์—๊ฒŒ ์‚ฌ์šฉ์ž์™€ ์—์ด์ „ํŠธ ์—ญํ• ์„ ๋ชจ๋‘ ์ˆ˜ํ–‰ํ•˜์—ฌ ์ฃผ์–ด์ง„ ๋ฌธ์„œ๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์„ ๊ตฌ์„ฑํ•˜๋„๋ก ์š”์ฒญํ•˜์—ฌ ์ฃผ์„์„ ๋ณด๋‹ค ํšจ์œจ์ ์œผ๋กœ ์ž‘์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ€์ด๋“œ๋ผ์ธ์€ ์„ธ ๋ถ€๋ถ„์œผ๋กœ ๊ตฌ์„ฑ๋ฉ๋‹ˆ๋‹ค: 1) ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์€ ์–ด๋–ค ๋ชจ์Šต์ธ๊ฐ€; 2) ์–ด๋–ค ์ข…๋ฅ˜์˜ ๋Œ€ํ™”ํ˜• QA ์ƒ˜ํ”Œ์ด ํ•„์š”ํ•œ๊ฐ€; 3) ์ฃผ์„์„ ๋‹ฌ์•„์•ผ ํ•˜๋Š” ๋‚ด์šฉ์€ ๋ฌด์—‡์ธ๊ฐ€?
Owing to the strong language comprehension, complex task reasoning, and common sense understanding capabilities, LLM-based autonomous agents have shown significant potential to influence multiple domains. This section provides a succinct summary of previous studies, categorizing them according to their applications in three distinct areas: social science, natural science, and engineering (see the left part of Figure 5 for a global overview).
๊ฐ•๋ ฅํ•œ ์–ธ์–ด ์ดํ•ด, ๋ณต์žกํ•œ ์ž‘์—… ์ถ”๋ก , ๊ทธ๋ฆฌ๊ณ  ์ƒ์‹ ์ดํ•ด ๋Šฅ๋ ฅ ๋•๋ถ„์—, LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ๋Š” ์—ฌ๋Ÿฌ ๋ถ„์•ผ์— ํฐ ์˜ํ–ฅ์„ ๋ฏธ์น  ์ž ์žฌ๋ ฅ์„ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ์ด์ „ ์—ฐ๊ตฌ๋“ค์„ ๊ฐ„๋žตํ•˜๊ฒŒ ์š”์•ฝํ•˜๊ณ , ์ด๋ฅผ ์‚ฌํšŒ ๊ณผํ•™, ์ž์—ฐ ๊ณผํ•™, ๊ทธ๋ฆฌ๊ณ  ๊ณตํ•™์˜ ์„ธ ๊ฐ€์ง€ ๊ตฌ๋ณ„๋œ ์˜์—ญ์—์„œ์˜ ์‘์šฉ์— ๋”ฐ๋ผ ๋ถ„๋ฅ˜ํ•ฉ๋‹ˆ๋‹ค (์ „๋ฐ˜์ ์ธ ๊ฐœ์š”๋Š” Figure 5์˜ ์™ผ์ชฝ ๋ถ€๋ถ„์„ ์ฐธ์กฐํ•˜์‹ญ์‹œ์˜ค).
๊ฐ•๋ ฅํ•œ ์–ธ์–ด ์ดํ•ด๋ ฅ, ๋ณต์žกํ•œ ๊ณผ์ œ ์ถ”๋ก  ๋ฐ ์ƒ์‹ ์ดํ•ด ๋Šฅ๋ ฅ ๋•๋ถ„์— LLM ๊ธฐ๋ฐ˜ ์ž์œจ ์—์ด์ „ํŠธ๋Š” ์—ฌ๋Ÿฌ ์˜์—ญ์— ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ๋Š” ์ƒ๋‹นํ•œ ์ž ์žฌ๋ ฅ์„ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ์„น์…˜์—์„œ๋Š” ์‚ฌํšŒ ๊ณผํ•™, ์ž์—ฐ ๊ณผํ•™ ๋ฐ ๊ณตํ•™์˜ ์„ธ ๊ฐ€์ง€ ๋ถ„์•ผ์— ๋Œ€ํ•œ ์‘์šฉ ๋ถ„์•ผ์— ๋”ฐ๋ผ ์ด์ „ ์—ฐ๊ตฌ๋ฅผ ๊ฐ„๋žตํ•˜๊ฒŒ ์š”์•ฝํ•˜์—ฌ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค(๊ทธ๋ฆผ 5์˜ ์™ผ์ชฝ ๋ถ€๋ถ„์„ ์ฐธ์กฐํ•˜์„ธ์š”).
The focus of this paper is on the datasets used to train language models. While recent work focused on documenting the potential harms that could arise from problematic datasets (Bender and Friedman, 2018; Gebru et al., 2020), less work has been done to quantitatively analyze properties of real language modelling datasets, like Dodge et al. (2021) has done for C4. Our paper provides analysis on one particular axis, that of data duplication.
์ด ๋…ผ๋ฌธ์˜ ์ดˆ์ ์€ ์–ธ์–ด ๋ชจ๋ธ์„ ํ›ˆ๋ จ์‹œํ‚ค๋Š” ๋ฐ ์‚ฌ์šฉ๋œ ๋ฐ์ดํ„ฐ์…‹์— ์žˆ์Šต๋‹ˆ๋‹ค. ์ตœ๊ทผ ์—ฐ๊ตฌ๋“ค์€ ๋ฌธ์ œ ์žˆ๋Š” ๋ฐ์ดํ„ฐ์…‹์—์„œ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋Š” ์ž ์žฌ์  ํ•ด์•…์„ ๋ฌธ์„œํ™”ํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘์—ˆ์ง€๋งŒ (Bender and Friedman, 2018; Gebru et al., 2020), ์‹ค์ œ ์–ธ์–ด ๋ชจ๋ธ๋ง ๋ฐ์ดํ„ฐ์…‹์˜ ํŠน์„ฑ์„ ์ •๋Ÿ‰์ ์œผ๋กœ ๋ถ„์„ํ•˜๋Š” ์ž‘์—…์€ ์ƒ๋Œ€์ ์œผ๋กœ ์ ์—ˆ์Šต๋‹ˆ๋‹ค. Dodge et al. (2021)์ด C4์— ๋Œ€ํ•ด ์ˆ˜ํ–‰ํ•œ ๊ฒƒ์ฒ˜๋Ÿผ ๋ง์ž…๋‹ˆ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์€ ๋ฐ์ดํ„ฐ ์ค‘๋ณต์ด๋ผ๋Š” ํŠน์ • ์ถ•์— ๋Œ€ํ•œ ๋ถ„์„์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
์ด ๋ฐฑ์„œ์˜ ์ดˆ์ ์€ ์–ธ์–ด ๋ชจ๋ธ์„ ํ›ˆ๋ จํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋˜๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์— ์žˆ์Šต๋‹ˆ๋‹ค. ์ตœ๊ทผ ์—ฐ๊ตฌ์—์„œ๋Š” ๋ฌธ์ œ๊ฐ€ ์žˆ๋Š” ๋ฐ์ดํ„ฐ ์„ธํŠธ์—์„œ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋Š” ์ž ์žฌ์  ์œ„ํ—˜์„ ๋ฌธ์„œํ™”ํ•˜๋Š” ๋ฐ ์ค‘์ ์„ ๋‘์—ˆ์Šต๋‹ˆ๋‹ค(Bender and Friedman, 2018; Gebru et al., 2020), Dodge et al. (2021)์ด C4์— ๋Œ€ํ•ด ์ˆ˜ํ–‰ํ•œ ๊ฒƒ๊ณผ ๊ฐ™์ด ์‹ค์ œ ์–ธ์–ด ๋ชจ๋ธ๋ง ๋ฐ์ดํ„ฐ ์„ธํŠธ์˜ ์†์„ฑ์„ ์ •๋Ÿ‰์ ์œผ๋กœ ๋ถ„์„ํ•˜๋Š” ์—ฐ๊ตฌ๋Š” ๋งŽ์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ด ๋ฐฑ์„œ์—์„œ๋Š” ๋ฐ์ดํ„ฐ ์ค‘๋ณต์ด๋ผ๋Š” ํŠน์ • ์ถ•์— ๋Œ€ํ•œ ๋ถ„์„์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
Figure 1: The inference process of Adapt-LLM step-by-step: given a question (step 1), an LLM decides (step 2) whether to answer the question directly (step 3) or to ask for additional contextual information, generating the special \(\langle\)RET\(\rangle\) token; for the later, an off-the-shelf IR system is used to retrieve relevant context (step 4), which is used alongside the question to prompt again the LLM for the final answer (step 5).
Figure 1: Adapt-LLM์˜ ์ถ”๋ก  ๊ณผ์ • ๋‹จ๊ณ„๋ณ„ ์„ค๋ช…: ์งˆ๋ฌธ์ด ์ฃผ์–ด์ง€๋ฉด (1๋‹จ๊ณ„), LLM์€ ์งˆ๋ฌธ์— ์ง์ ‘ ๋‹ตํ• ์ง€ (3๋‹จ๊ณ„) ์•„๋‹ˆ๋ฉด ์ถ”๊ฐ€์ ์ธ ๋งฅ๋ฝ ์ •๋ณด๋ฅผ ์š”์ฒญํ• ์ง€ ๊ฒฐ์ •ํ•ฉ๋‹ˆ๋‹ค (2๋‹จ๊ณ„). ํ›„์ž์˜ ๊ฒฝ์šฐ, ํŠน์ˆ˜ํ•œ \(\langle\)RET\(\rangle\) ํ† ํฐ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ดํ›„, ๊ด€๋ จ๋œ ๋งฅ๋ฝ์„ ๊ฒ€์ƒ‰ํ•˜๊ธฐ ์œ„ํ•ด ๊ธฐ์กด์˜ IR ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•˜๋ฉฐ (4๋‹จ๊ณ„), ์ด ๋งฅ๋ฝ์€ ์งˆ๋ฌธ๊ณผ ํ•จ๊ป˜ ์ตœ์ข… ๋‹ต๋ณ€์„ ์œ„ํ•ด LLM์„ ๋‹ค์‹œ ํ”„๋กฌํ”„ํŠธํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค (5๋‹จ๊ณ„).
๊ทธ๋ฆผ 1: ์งˆ๋ฌธ์ด ์ฃผ์–ด์ง€๋ฉด(1๋‹จ๊ณ„), LLM์ด ์งˆ๋ฌธ์— ์ง์ ‘ ๋‹ตํ• ์ง€(3๋‹จ๊ณ„) ์•„๋‹ˆ๋ฉด ์ถ”๊ฐ€์ ์ธ ์ปจํ…์ŠคํŠธ ์ •๋ณด๋ฅผ ์š”์ฒญํ•˜์—ฌ ํŠน์ˆ˜(\(RET\(\rangle\) ํ† ํฐ์„ ์ƒ์„ฑํ•˜๋Š”์ง€ ๊ฒฐ์ •ํ•ฉ๋‹ˆ๋‹ค(2๋‹จ๊ณ„); ํ›„์ž์˜ ๊ฒฝ์šฐ, ๊ธฐ์„ฑ IR ์‹œ์Šคํ…œ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ด€๋ จ ์ปจํ…์ŠคํŠธ๋ฅผ ๊ฒ€์ƒ‰ํ•ฉ๋‹ˆ๋‹ค(4๋‹จ๊ณ„), ์ด ์ปจํ…์ŠคํŠธ๋Š” ์งˆ๋ฌธ๊ณผ ํ•จ๊ป˜ ์‚ฌ์šฉ๋˜์–ด ์ตœ์ข… ๋‹ต๋ณ€์„ ์œ„ํ•ด ๋‹ค์‹œ LLM์— ํ”„๋กฌํ”„ํŠธํ•ฉ๋‹ˆ๋‹ค(5๋‹จ๊ณ„).
We conduct ablation studies in terms of the number of unanswerable samples in HumanAnnotatedConvQA. We find that using a small amount of unanswerable samples (e.g., 1.5k) is able to achieve remarkable results on the unanswerable evaluation, as well as the overall conversational QA score (i.e., Avg-CQA). Interestingly, incorporating more unanswerable samples does not necessarily lead to higher accuracy scores in the unanswerable evaluation.
์šฐ๋ฆฌ๋Š” HumanAnnotatedConvQA์—์„œ ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ์ƒ˜ํ”Œ์˜ ์ˆ˜์— ๋Œ€ํ•œ ์†Œ๊ฑฐ ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ์†Œ๋Ÿ‰์˜ ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ์ƒ˜ํ”Œ(์˜ˆ: 1.5k)์„ ์‚ฌ์šฉํ•˜๋ฉด ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ํ‰๊ฐ€์—์„œ ๋†€๋ผ์šด ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์„ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ์ „์ฒด ๋Œ€ํ™”ํ˜• QA ์ ์ˆ˜(์ฆ‰, Avg-CQA)์—์„œ๋„ ๋›ฐ์–ด๋‚œ ์„ฑ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Œ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค. ํฅ๋ฏธ๋กญ๊ฒŒ๋„, ๋” ๋งŽ์€ ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ์ƒ˜ํ”Œ์„ ํฌํ•จํ•œ๋‹ค๊ณ  ํ•ด์„œ ๋ฐ˜๋“œ์‹œ ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ํ‰๊ฐ€์—์„œ ๋” ๋†’์€ ์ •ํ™•๋„ ์ ์ˆ˜๋ฅผ ์–ป๋Š” ๊ฒƒ์€ ์•„๋‹™๋‹ˆ๋‹ค.
HumanAnnotatedConvQA์—์„œ ๋‹ต๋ณ€ํ•  ์ˆ˜ ์—†๋Š” ์ƒ˜ํ”Œ์˜ ์ˆ˜ ์ธก๋ฉด์—์„œ ์ ˆ์ œ ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ์†Œ๋Ÿ‰์˜ ๋‹ต๋ณ€ ๋ถˆ๊ฐ€๋Šฅํ•œ ์ƒ˜ํ”Œ(์˜ˆ: 1.5k)์„ ์‚ฌ์šฉํ•˜๋ฉด ๋‹ต๋ณ€ ๋ถˆ๊ฐ€๋Šฅํ•œ ํ‰๊ฐ€์™€ ์ „์ฒด ๋Œ€ํ™”ํ˜• QA ์ ์ˆ˜(์ฆ‰, Avg-CQA)์—์„œ ๋†€๋ผ์šด ๊ฒฐ๊ณผ๋ฅผ ์–ป์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ํฅ๋ฏธ๋กญ๊ฒŒ๋„ ๋” ๋งŽ์€ ๋‹ต๋ณ€ ๋ถˆ๊ฐ€๋Šฅํ•œ ์ƒ˜ํ”Œ์„ ํ†ตํ•ฉํ•œ๋‹ค๊ณ  ํ•ด์„œ ๋‹ต๋ณ€ ๋ถˆ๊ฐ€๋Šฅํ•œ ํ‰๊ฐ€์—์„œ ๋ฐ˜๋“œ์‹œ ๋” ๋†’์€ ์ •ํ™•๋„ ์ ์ˆ˜๋กœ ์ด์–ด์ง€์ง€๋Š” ์•Š์Šต๋‹ˆ๋‹ค.
Language identification.We use the fastText language classifier of CCNet (Wenzek et al., 2020) at the document-level: it uses characters n-gram and was trained on Wikipedia, supporting 176 languages. We remove documents for which the top language scores below 0.65: this usually corresponds to pages without any natural text. For this paper, we focus on English; RefinedWeb can also be derived for other languages, see Appendix D for details.
์–ธ์–ด ์‹๋ณ„. ์šฐ๋ฆฌ๋Š” ๋ฌธ์„œ ์ˆ˜์ค€์—์„œ CCNet (Wenzek et al., 2020)์˜ fastText ์–ธ์–ด ๋ถ„๋ฅ˜๊ธฐ๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ถ„๋ฅ˜๊ธฐ๋Š” ๋ฌธ์ž n-gram์„ ์‚ฌ์šฉํ•˜๋ฉฐ, 176๊ฐœ ์–ธ์–ด๋ฅผ ์ง€์›ํ•˜๋Š” Wikipedia์—์„œ ํ•™์Šต๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ฃผ์š” ์–ธ์–ด ์ ์ˆ˜๊ฐ€ 0.65 ์ดํ•˜์ธ ๋ฌธ์„œ๋Š” ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์ž์—ฐ์–ด ํ…์ŠคํŠธ๊ฐ€ ์—†๋Š” ํŽ˜์ด์ง€์— ํ•ด๋‹นํ•ฉ๋‹ˆ๋‹ค. ์ด ๋…ผ๋ฌธ์—์„œ๋Š” ์˜์–ด์— ์ค‘์ ์„ ๋‘๊ณ  ์žˆ์œผ๋ฉฐ, RefinedWeb์€ ๋‹ค๋ฅธ ์–ธ์–ด๋กœ๋„ ํŒŒ์ƒ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก D๋ฅผ ์ฐธ์กฐํ•˜์‹ญ์‹œ์˜ค.
์–ธ์–ด ์‹๋ณ„.๋ฌธ์„œ ์ˆ˜์ค€์—์„œ CCNet(Wenzek et al., 2020)์˜ fastText ์–ธ์–ด ๋ถ„๋ฅ˜๊ธฐ๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. ๋ฌธ์ž n-๊ทธ๋žจ์„ ์‚ฌ์šฉํ•˜๋ฉฐ 176๊ฐœ ์–ธ์–ด๋ฅผ ์ง€์›ํ•˜๋Š” ์œ„ํ‚คํ”ผ๋””์•„์—์„œ ํ•™์Šต๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ƒ์œ„ ์–ธ์–ด ์ ์ˆ˜๊ฐ€ 0.65 ๋ฏธ๋งŒ์ธ ๋ฌธ์„œ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์ž์—ฐ ํ…์ŠคํŠธ๊ฐ€ ์—†๋Š” ํŽ˜์ด์ง€์— ํ•ด๋‹นํ•˜๋ฏ€๋กœ ์ œ๊ฑฐํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ฐฑ์„œ์—์„œ๋Š” ์˜์–ด์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์žˆ์œผ๋ฉฐ, ๋‹ค๋ฅธ ์–ธ์–ด์— ๋Œ€ํ•ด์„œ๋„ RefinedWeb์„ ๋„์ถœํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๋‚ด์šฉ์€ ๋ถ€๋ก D๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.
We collected responses from all models with a temperature of 0.01 and set the maximum number of tokens to be generated to 500. All remaining parameters were kept at the default (e.g. top_p = 0.7). Sampling with a temperature of 0.01 is not fully deterministic, but it reduces the variability of models' responses. This also reduces their creativity, which can minimize the risk of models giving more unusual and potentially unsafe responses.
๋ชจ๋“  ๋ชจ๋ธ์—์„œ ์˜จ๋„๋ฅผ 0.01๋กœ ์„ค์ •ํ•˜๊ณ  ์ƒ์„ฑํ•  ์ตœ๋Œ€ ํ† ํฐ ์ˆ˜๋ฅผ 500์œผ๋กœ ์„ค์ •ํ•˜์—ฌ ์‘๋‹ต์„ ์ˆ˜์ง‘ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋‚˜๋จธ์ง€ ๋ชจ๋“  ๋งค๊ฐœ๋ณ€์ˆ˜๋Š” ๊ธฐ๋ณธ๊ฐ’์œผ๋กœ ์œ ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค(์˜ˆ: top_p = 0.7). ์˜จ๋„ 0.01๋กœ ์ƒ˜ํ”Œ๋งํ•˜๋Š” ๊ฒƒ์€ ์™„์ „ํžˆ ๊ฒฐ์ •๋ก ์ ์ด์ง€๋Š” ์•Š์ง€๋งŒ, ๋ชจ๋ธ ์‘๋‹ต์˜ ๋ณ€๋™์„ฑ์„ ์ค„์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋Š” ๋˜ํ•œ ๋ชจ๋ธ์˜ ์ฐฝ์˜์„ฑ์„ ๊ฐ์†Œ์‹œ์ผœ, ๋” ์ด์ƒํ•˜๊ณ  ์ž ์žฌ์ ์œผ๋กœ ์•ˆ์ „ํ•˜์ง€ ์•Š์€ ์‘๋‹ต์„ ์ค„์ด๋Š” ๋ฐ ๋„์›€์ด ๋ฉ๋‹ˆ๋‹ค.
์˜จ๋„๊ฐ€ 0.01์ธ ๋ชจ๋“  ๋ชจ๋ธ์˜ ์‘๋‹ต์„ ์ˆ˜์ง‘ํ•˜๊ณ  ์ƒ์„ฑํ•  ์ตœ๋Œ€ ํ† ํฐ ์ˆ˜๋ฅผ 500์œผ๋กœ ์„ค์ •ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋‚˜๋จธ์ง€ ๋ชจ๋“  ๋งค๊ฐœ๋ณ€์ˆ˜๋Š” ๊ธฐ๋ณธ๊ฐ’(์˜ˆ: top_p = 0.7)์œผ๋กœ ์œ ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜จ๋„๊ฐ€ 0.01์ธ ์ƒ˜ํ”Œ๋ง์€ ์™„์ „ํžˆ ๊ฒฐ์ •๋ก ์ ์ด์ง€๋Š” ์•Š์ง€๋งŒ ๋ชจ๋ธ ์‘๋‹ต์˜ ๋ณ€๋™์„ฑ์„ ์ค„์—ฌ์ค๋‹ˆ๋‹ค. ๋˜ํ•œ ๋ชจ๋ธ์˜ ์ฐฝ์˜์„ฑ๋„ ๊ฐ์†Œํ•˜์—ฌ ๋ชจ๋ธ์ด ๋” ํŠน์ดํ•˜๊ณ  ์ž ์žฌ์ ์œผ๋กœ ์•ˆ์ „ํ•˜์ง€ ์•Š์€ ์‘๋‹ต์„ ์ œ๊ณตํ•  ์œ„ํ—˜์„ ์ตœ์†Œํ™”ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
This paper introduces v0.5 of the **AI Safety Benchmark** which has been created by the MLCommons AI Safety Working Group (WG). The MLCommons AI Safety WG is a consortium of industry and academic researchers, engineers, and practitioners. The primary goal of the WG is to advance the state of the art for evaluating AI safety. We hope to facilitate better AI safety processes and stimulate AI safety innovation across industry and research.
์ด ๋…ผ๋ฌธ์€ MLCommons AI Safety Working Group (WG)์— ์˜ํ•ด ๋งŒ๋“ค์–ด์ง„ **AI Safety Benchmark** v0.5๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. MLCommons AI Safety WG๋Š” ์‚ฐ์—… ๋ฐ ํ•™๊ณ„ ์—ฐ๊ตฌ์ž, ์—”์ง€๋‹ˆ์–ด, ์‹ค๋ฌด์ž๋“ค๋กœ ๊ตฌ์„ฑ๋œ ์ปจ์†Œ์‹œ์—„์ž…๋‹ˆ๋‹ค. WG์˜ ์ฃผ์š” ๋ชฉํ‘œ๋Š” AI ์•ˆ์ „์„ฑ ํ‰๊ฐ€์˜ ์ตœ์ฒจ๋‹จ ๊ธฐ์ˆ ์„ ๋ฐœ์ „์‹œํ‚ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋” ๋‚˜์€ AI ์•ˆ์ „์„ฑ ํ”„๋กœ์„ธ์Šค๋ฅผ ์ด‰์ง„ํ•˜๊ณ  ์‚ฐ์—… ๋ฐ ์—ฐ๊ตฌ ์ „๋ฐ˜์— ๊ฑธ์ณ AI ์•ˆ์ „์„ฑ ํ˜์‹ ์„ ์ž๊ทนํ•˜๊ธฐ๋ฅผ ํฌ๋งํ•ฉ๋‹ˆ๋‹ค.
์ด ๋ฐฑ์„œ์—์„œ๋Š” MLCommons AI Safety Working Group(WG)์—์„œ ๋งŒ๋“  **AI Safety Benchmark** v0.5๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. MLCommons AI Safety WG๋Š” ์—…๊ณ„ ๋ฐ ํ•™๊ณ„ ์—ฐ๊ตฌ์ž, ์—”์ง€๋‹ˆ์–ด, ์‹ค๋ฌด์ž๋กœ ๊ตฌ์„ฑ๋œ ์ปจ์†Œ์‹œ์—„์ž…๋‹ˆ๋‹ค. WG์˜ ์ฃผ์š” ๋ชฉํ‘œ๋Š” AI ์•ˆ์ „์„ฑ ํ‰๊ฐ€์˜ ์ตœ์ฒจ๋‹จ ๊ธฐ์ˆ ์„ ๋ฐœ์ „์‹œํ‚ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ๋” ๋‚˜์€ AI ์•ˆ์ „ ํ”„๋กœ์„ธ์Šค๋ฅผ ์ด‰์ง„ํ•˜๊ณ  ์—…๊ณ„์™€ ์—ฐ๊ตฌ ๋ถ„์•ผ ์ „๋ฐ˜์— ๊ฑธ์ณ AI ์•ˆ์ „ ํ˜์‹ ์„ ์ด‰์ง„ํ•˜๊ธฐ๋ฅผ ํฌ๋งํ•ฉ๋‹ˆ๋‹ค.
In Figure 13, we display the dynamic changes in token selection tendencies throughout the SLM training process. We chose four checkpoints during the training process (0%, 33%, 66%, and 100%) to analyze the current tendencies in token selection. The preferences for token selection are indicated by different colors, ranging from high to low preference, typically represented as deep blue, blue, black, orange, and dark orange, respectively.
Figure 13์—์„œ๋Š” SLM ํ›ˆ๋ จ ๊ณผ์ • ์ „๋ฐ˜์— ๊ฑธ์นœ ํ† ํฐ ์„ ํƒ ๊ฒฝํ–ฅ์˜ ๋™์  ๋ณ€ํ™”๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ›ˆ๋ จ ๊ณผ์ • ์ค‘ 4๊ฐœ์˜ ์ฒดํฌํฌ์ธํŠธ(0%, 33%, 66%, 100%)๋ฅผ ์„ ํƒํ•˜์—ฌ ํ˜„์žฌ ํ† ํฐ ์„ ํƒ ๊ฒฝํ–ฅ์„ ๋ถ„์„ํ–ˆ์Šต๋‹ˆ๋‹ค. ํ† ํฐ ์„ ํƒ์— ๋Œ€ํ•œ ์„ ํ˜ธ๋„๋Š” ๋‹ค์–‘ํ•œ ์ƒ‰์ƒ์œผ๋กœ ํ‘œ์‹œ๋˜๋ฉฐ, ์ผ๋ฐ˜์ ์œผ๋กœ ๋†’์€ ์„ ํ˜ธ๋„์—์„œ ๋‚ฎ์€ ์„ ํ˜ธ๋„๋กœ ๊นŠ์€ ํŒŒ๋ž€์ƒ‰, ํŒŒ๋ž€์ƒ‰, ๊ฒ€์€์ƒ‰, ์ฃผํ™ฉ์ƒ‰, ์ง™์€ ์ฃผํ™ฉ์ƒ‰์œผ๋กœ ๋‚˜ํƒ€๋‚ฉ๋‹ˆ๋‹ค.
๊ทธ๋ฆผ 13์—์„œ๋Š” SLM ํ•™์Šต ๊ณผ์ •์—์„œ ํ† ํฐ ์„ ํƒ ๊ฒฝํ–ฅ์˜ ๋™์  ๋ณ€ํ™”๋ฅผ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ํ•™์Šต ๊ณผ์ •์—์„œ ๋„ค ๊ฐœ์˜ ์ฒดํฌํฌ์ธํŠธ(0%, 33%, 66%, 100%)๋ฅผ ์„ ํƒํ•˜์—ฌ ํ† ํฐ ์„ ํƒ์˜ ํ˜„์žฌ ๊ฒฝํ–ฅ์„ ๋ถ„์„ํ–ˆ์Šต๋‹ˆ๋‹ค. ํ† ํฐ ์„ ํƒ์— ๋Œ€ํ•œ ์„ ํ˜ธ๋„๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์ง„ํ•œ ํŒŒ๋ž€์ƒ‰, ํŒŒ๋ž€์ƒ‰, ๊ฒ€์€์ƒ‰, ์ฃผํ™ฉ์ƒ‰, ์ง„ํ•œ ์ฃผํ™ฉ์ƒ‰์œผ๋กœ ํ‘œ์‹œ๋˜๋Š” ๋†’์€ ์„ ํ˜ธ๋„์—์„œ ๋‚ฎ์€ ์„ ํ˜ธ๋„๊นŒ์ง€ ๋‹ค์–‘ํ•œ ์ƒ‰์ƒ์œผ๋กœ ํ‘œ์‹œ๋ฉ๋‹ˆ๋‹ค.
\(\bullet\)**HH-RLHF**[170] consists of around 169K instances, and can be divided into two parts that focus on the helpfulness and harmlessness of LLMs, respectively. Each instance is an open-ended conversation between a crowdworker and a chat model, about seeking assistance, advice, or task completion. The chat model provides two responses to each user query, and the more helpful or harmful responses will be chosen as the annotations.
\(\bullet\)**HH-RLHF**[170]๋Š” ์•ฝ 169K๊ฐœ์˜ ์ธ์Šคํ„ด์Šค๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๊ฐ๊ฐ LLM์˜ ์œ ์šฉ์„ฑ๊ณผ ๋ฌดํ•ด์„ฑ์— ์ค‘์ ์„ ๋‘” ๋‘ ๋ถ€๋ถ„์œผ๋กœ ๋‚˜๋ˆŒ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐ ์ธ์Šคํ„ด์Šค๋Š” ํฌ๋ผ์šฐ๋“œ์›Œ์ปค์™€ ์ฑ„ํŒ… ๋ชจ๋ธ ๊ฐ„์˜ ๊ฐœ๋ฐฉํ˜• ๋Œ€ํ™”๋กœ, ๋„์›€, ์กฐ์–ธ ๋˜๋Š” ์ž‘์—… ์™„๋ฃŒ๋ฅผ ์š”์ฒญํ•˜๋Š” ๋‚ด์šฉ์ž…๋‹ˆ๋‹ค. ์ฑ„ํŒ… ๋ชจ๋ธ์€ ๊ฐ ์‚ฌ์šฉ์ž ์งˆ๋ฌธ์— ๋‘ ๊ฐ€์ง€ ์‘๋‹ต์„ ์ œ๊ณตํ•˜๋ฉฐ, ๋” ์œ ์šฉํ•˜๊ฑฐ๋‚˜ ํ•ด๋กœ์šด ์‘๋‹ต์ด ์ฃผ์„์œผ๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค.
\(\bullet\)**HH-RLHF**[170]๋Š” ์•ฝ 169K ์ธ์Šคํ„ด์Šค๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์œผ๋ฉฐ, ๊ฐ๊ฐ LLM์˜ ์œ ์šฉ์„ฑ๊ณผ ๋ฌดํ•ด์„ฑ์— ์ดˆ์ ์„ ๋งž์ถ˜ ๋‘ ๋ถ€๋ถ„์œผ๋กœ ๋‚˜๋ˆŒ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐ ์ธ์Šคํ„ด์Šค๋Š” ํฌ๋ผ์šฐ๋“œ์›Œ์ปค์™€ ์ฑ„ํŒ… ๋ชจ๋ธ ๊ฐ„์˜ ๋Œ€ํ™”๋กœ, ๋„์›€์ด๋‚˜ ์กฐ์–ธ, ์ž‘์—… ์™„๋ฃŒ๋ฅผ ๊ตฌํ•˜๋Š” ๋‚ด์šฉ์ž…๋‹ˆ๋‹ค. ์ฑ„ํŒ… ๋ชจ๋ธ์€ ๊ฐ ์‚ฌ์šฉ์ž ์ฟผ๋ฆฌ์— ๋Œ€ํ•ด ๋‘ ๊ฐœ์˜ ์‘๋‹ต์„ ์ œ๊ณตํ•˜๋ฉฐ, ๋” ์œ ์šฉํ•˜๊ฑฐ๋‚˜ ์œ ํ•ดํ•œ ์‘๋‹ต์ด ์ฃผ์„์œผ๋กœ ์„ ํƒ๋ฉ๋‹ˆ๋‹ค.
1. Legality varies across jurisdictions, even for countries with broadly similar legal systems. 2. Laws are often incomplete, difficult to understand, and may have surprising gaps, asymmetries and limitations. 3. The legality of specific activities is difficult to assess and requires trained legal expertise. 4. AI companies have restrictions on the use of their models, which are defined not solely on the basis of legal prohibitions 34.
1. ๋ฒ•์  ์ƒํ™ฉ์€ ๋น„์Šทํ•œ ๋ฒ• ์ฒด๊ณ„๋ฅผ ๊ฐ€์ง„ ๊ตญ๊ฐ€๋“ค ์‚ฌ์ด์—์„œ๋„ ๊ด€ํ•  ๊ตฌ์—ญ์— ๋”ฐ๋ผ ๋‹ค๋ฆ…๋‹ˆ๋‹ค. 2. ๋ฒ•๋ฅ ์€ ์ข…์ข… ๋ถˆ์™„์ „ํ•˜๊ณ  ์ดํ•ดํ•˜๊ธฐ ์–ด๋ ค์šฐ๋ฉฐ, ์˜ˆ์ƒ์น˜ ๋ชปํ•œ ๊ณต๋ฐฑ, ๋น„๋Œ€์นญ์„ฑ ๋ฐ ์ œํ•œ์„ ๊ฐ€์งˆ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. 3. ํŠน์ • ํ™œ๋™์˜ ํ•ฉ๋ฒ•์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” ๊ฒƒ์€ ์–ด๋ ต๊ณ , ํ›ˆ๋ จ๋œ ๋ฒ•๋ฅ  ์ „๋ฌธ๊ฐ€์˜ ์ „๋ฌธ ์ง€์‹์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. 4. AI ํšŒ์‚ฌ๋“ค์€ ๋ฒ•์  ๊ธˆ์ง€ ์‚ฌํ•ญ์—๋งŒ ๊ทผ๊ฑฐํ•˜์ง€ ์•Š๊ณ  ๋ชจ๋ธ ์‚ฌ์šฉ์— ๋Œ€ํ•œ ์ œํ•œ์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
1. ๋ฒ•๋ฅ ์€ ๋ฒ•๋ฅ  ์ฒด๊ณ„๊ฐ€ ๋น„์Šทํ•œ ๊ตญ๊ฐ€์—์„œ๋„ ๊ด€ํ• ๊ถŒ์— ๋”ฐ๋ผ ๋‹ค๋ฆ…๋‹ˆ๋‹ค. 2. ๋ฒ•๋ฅ ์€ ๋ถˆ์™„์ „ํ•˜๊ณ  ์ดํ•ดํ•˜๊ธฐ ์–ด๋ ค์šฐ๋ฉฐ ๋†€๋ผ์šด ๊ฒฉ์ฐจ, ๋น„๋Œ€์นญ์„ฑ ๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. 3. ํŠน์ • ํ™œ๋™์˜ ํ•ฉ๋ฒ•์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ๋Š” ์–ด๋ ต๊ณ  ์ˆ™๋ จ๋œ ๋ฒ•๋ฅ  ์ „๋ฌธ ์ง€์‹์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. 4. AI ํšŒ์‚ฌ๋Š” ๋ฒ•์  ๊ธˆ์ง€ ์‚ฌํ•ญ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ ๋ชจ๋ธ ์‚ฌ์šฉ์— ๋Œ€ํ•œ ์ œํ•œ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
LLaMA is a collection of foundation language models, released by Meta. Unlike GPT models, LLaMA models are open-source, i.e., model weights are released to the research community under a noncommercial license. Thus, the LLaMA family grows rapidly as these models are widely used by many research groups to develop better open-source LLMs to compete the closed-source ones or to develop task-specific LLMs for mission-critical applications.
LLaMA๋Š” Meta์—์„œ ์ถœ์‹œํ•œ ๊ธฐ์ดˆ ์–ธ์–ด ๋ชจ๋ธ ๋ชจ์Œ์ž…๋‹ˆ๋‹ค. GPT ๋ชจ๋ธ๊ณผ ๋‹ฌ๋ฆฌ, LLaMA ๋ชจ๋ธ์€ ์˜คํ”ˆ ์†Œ์Šค์ด๋ฉฐ, ์ฆ‰ ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๊ฐ€ ๋น„์ƒ์—…์  ๋ผ์ด์„ ์Šค ํ•˜์— ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ๊ณต๊ฐœ๋ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ LLaMA ๊ณ„์—ด์€ ๋งŽ์€ ์—ฐ๊ตฌ ๊ทธ๋ฃน์ด ๋” ๋‚˜์€ ์˜คํ”ˆ ์†Œ์Šค LLM์„ ๊ฐœ๋ฐœํ•˜๊ฑฐ๋‚˜ ํ์‡„ํ˜• ๋ชจ๋ธ๊ณผ ๊ฒฝ์Ÿํ•˜๊ฑฐ๋‚˜ ์ค‘์š”ํ•œ ์ž„๋ฌด์— ํŠนํ™”๋œ LLM์„ ๊ฐœ๋ฐœํ•˜๊ธฐ ์œ„ํ•ด ๋„๋ฆฌ ์‚ฌ์šฉํ•จ์— ๋”ฐ๋ผ ๋น ๋ฅด๊ฒŒ ์„ฑ์žฅํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
LLaMA๋Š” Meta์—์„œ ๊ณต๊ฐœํ•œ ๊ธฐ๋ฐ˜ ์–ธ์–ด ๋ชจ๋ธ ๋ชจ์Œ์ž…๋‹ˆ๋‹ค. GPT ๋ชจ๋ธ๊ณผ ๋‹ฌ๋ฆฌ LLaMA ๋ชจ๋ธ์€ ์˜คํ”ˆ ์†Œ์Šค, ์ฆ‰ ๋ชจ๋ธ ๊ฐ€์ค‘์น˜๊ฐ€ ๋น„์ƒ์šฉ ๋ผ์ด์„ ์Šค์— ๋”ฐ๋ผ ์—ฐ๊ตฌ ์ปค๋ฎค๋‹ˆํ‹ฐ์— ๊ณต๊ฐœ๋ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด๋Ÿฌํ•œ ๋ชจ๋ธ์€ ๋งŽ์€ ์—ฐ๊ตฌ ๊ทธ๋ฃน์—์„œ ๋„๋ฆฌ ์‚ฌ์šฉ๋˜์–ด ๋น„๊ณต๊ฐœ ๋ชจ๋ธ๊ณผ ๊ฒฝ์Ÿํ•  ์ˆ˜ ์žˆ๋Š” ๋” ๋‚˜์€ ์˜คํ”ˆ ์†Œ์Šค LLM์„ ๊ฐœ๋ฐœํ•˜๊ฑฐ๋‚˜ ๋ฏธ์…˜ ํฌ๋ฆฌํ‹ฐ์ปฌ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ์œ„ํ•œ ์ž‘์—…๋ณ„ LLM์„ ๊ฐœ๋ฐœํ•˜๊ธฐ ๋•Œ๋ฌธ์— LLaMA ์ œํ’ˆ๊ตฐ์ด ๋น ๋ฅด๊ฒŒ ์„ฑ์žฅํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
Footnote 4: A further possibility is that rather than using the final losses as per their description of Approach 3, they also used intermediate losses during training. If so, confidence intervals would be substantially narrower, but likely not by an order of magnitude if standard errors were appropriately clustered. Overall, we think this is unlikely to explain the tight intervals we see reported by Hoffman et al. (see Appendix A.2).
Footnote 4: ๋˜ ๋‹ค๋ฅธ ๊ฐ€๋Šฅ์„ฑ์€ Approach 3์˜ ์„ค๋ช…์— ๋”ฐ๋ฅธ ์ตœ์ข… ์†์‹ค์„ ์‚ฌ์šฉํ•˜๋Š” ๋Œ€์‹ , ํ›ˆ๋ จ ์ค‘๊ฐ„์— ์ค‘๊ฐ„ ์†์‹ค๋„ ์‚ฌ์šฉํ–ˆ์„ ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋งŒ์•ฝ ๊ทธ๋ ‡๋‹ค๋ฉด, ์‹ ๋ขฐ ๊ตฌ๊ฐ„์€ ์ƒ๋‹นํžˆ ์ข์•„์ง€๊ฒ ์ง€๋งŒ, ํ‘œ์ค€ ์˜ค์ฐจ๊ฐ€ ์ ์ ˆํžˆ ํด๋Ÿฌ์Šคํ„ฐ๋ง๋˜์—ˆ๋‹ค๋ฉด ๊ทธ ์ •๋„๋Š” ์•„๋‹ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ „๋ฐ˜์ ์œผ๋กœ, ์šฐ๋ฆฌ๋Š” ์ด๊ฒƒ์ด Hoffman ๋“ฑ(๋ถ€๋ก A.2 ์ฐธ์กฐ)์ด ๋ณด๊ณ ํ•œ ์ข์€ ๊ตฌ๊ฐ„์„ ์„ค๋ช…ํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
๊ฐ์ฃผ 4: ๋˜ ๋‹ค๋ฅธ ๊ฐ€๋Šฅ์„ฑ์€ ์ ‘๊ทผ ๋ฐฉ์‹ 3์— ๋Œ€ํ•œ ์„ค๋ช…๋Œ€๋กœ ์ตœ์ข… ์†์‹ค์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด ์•„๋‹ˆ๋ผ ํ›ˆ๋ จ ์ค‘์— ์ค‘๊ฐ„ ์†์‹ค์„ ์‚ฌ์šฉํ–ˆ์„ ์ˆ˜๋„ ์žˆ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ทธ๋ ‡๋‹ค๋ฉด ์‹ ๋ขฐ ๊ตฌ๊ฐ„์ด ํ›จ์”ฌ ์ข์•„์ง€๊ฒ ์ง€๋งŒ ํ‘œ์ค€ ์˜ค์ฐจ๋ฅผ ์ ์ ˆํžˆ ํด๋Ÿฌ์Šคํ„ฐ๋งํ•œ ๊ฒฝ์šฐ 10๋ฐฐ ์ •๋„๋Š” ์•„๋‹ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ „๋ฐ˜์ ์œผ๋กœ Hoffman ๋“ฑ์ด ๋ณด๊ณ ํ•œ ๊ธด์ถ•๋œ ๊ฐ„๊ฒฉ์„ ์„ค๋ช…ํ•˜๊ธฐ์—๋Š” ์ด ๋ฐฉ๋ฒ•์œผ๋กœ๋Š” ์„ค๋ช…ํ•˜๊ธฐ ์–ด๋ ต๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค(๋ถ€๋ก A.2 ์ฐธ์กฐ).
As an example, the recruitment step allows agents to be removed or added based on the progress towards the goal. This helps ensure that the right agents are participating at any given stage of problem solving. The researchers found that horizontal teams are generally best suited for collaborative tasks like consulting, while vertical teams are better suited for tasks that require clearer isolation of responsibilities for tool calling.
์˜ˆ๋ฅผ ๋“ค์–ด, ๋ชจ์ง‘ ๋‹จ๊ณ„์—์„œ๋Š” ๋ชฉํ‘œ ๋‹ฌ์„ฑ ์ง„ํ–‰ ์ƒํ™ฉ์— ๋”ฐ๋ผ ์—์ด์ „ํŠธ๋ฅผ ์ œ๊ฑฐํ•˜๊ฑฐ๋‚˜ ์ถ”๊ฐ€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋ฌธ์ œ ํ•ด๊ฒฐ์˜ ๊ฐ ๋‹จ๊ณ„์—์„œ ์ ์ ˆํ•œ ์—์ด์ „ํŠธ๊ฐ€ ์ฐธ์—ฌํ•˜๋„๋ก ๋ณด์žฅํ•˜๋Š” ๋ฐ ๋„์›€์ด ๋ฉ๋‹ˆ๋‹ค. ์—ฐ๊ตฌ์ž๋“ค์€ ์ˆ˜ํ‰ ํŒ€์ด ์ผ๋ฐ˜์ ์œผ๋กœ ์ปจ์„คํŒ…๊ณผ ๊ฐ™์€ ํ˜‘์—… ์ž‘์—…์— ๊ฐ€์žฅ ์ ํ•ฉํ•˜๋ฉฐ, ์ˆ˜์ง ํŒ€์€ ๋„๊ตฌ ํ˜ธ์ถœ์— ๋Œ€ํ•œ ์ฑ…์ž„์˜ ๋ช…ํ™•ํ•œ ๋ถ„๋ฆฌ๊ฐ€ ํ•„์š”ํ•œ ์ž‘์—…์— ๋” ์ ํ•ฉํ•˜๋‹ค๋Š” ๊ฒƒ์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค.
ํ•œ ์˜ˆ๋กœ, ๋ชจ์ง‘ ๋‹จ๊ณ„์—์„œ๋Š” ๋ชฉํ‘œ์— ๋Œ€ํ•œ ์ง„ํ–‰ ์ƒํ™ฉ์— ๋”ฐ๋ผ ์—์ด์ „ํŠธ๋ฅผ ์ œ๊ฑฐํ•˜๊ฑฐ๋‚˜ ์ถ”๊ฐ€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๋ฌธ์ œ ํ•ด๊ฒฐ์˜ ํŠน์ • ๋‹จ๊ณ„์—์„œ ์ ์ ˆํ•œ ์—์ด์ „ํŠธ๊ฐ€ ์ฐธ์—ฌํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ฉ๋‹ˆ๋‹ค. ์—ฐ๊ตฌ์ง„์€ ์ˆ˜ํ‰ํ˜• ํŒ€์€ ์ผ๋ฐ˜์ ์œผ๋กœ ์ปจ์„คํŒ…๊ณผ ๊ฐ™์€ ํ˜‘์—… ์ž‘์—…์— ๊ฐ€์žฅ ์ ํ•ฉํ•œ ๋ฐ˜๋ฉด, ์ˆ˜์งํ˜• ํŒ€์€ ๋„๊ตฌ ํ˜ธ์ถœ์— ๋Œ€ํ•œ ์ฑ…์ž„์„ ๋” ๋ช…ํ™•ํ•˜๊ฒŒ ๋ถ„๋ฆฌํ•ด์•ผ ํ•˜๋Š” ์ž‘์—…์— ๋” ์ ํ•ฉํ•˜๋‹ค๋Š” ์‚ฌ์‹ค์„ ๋ฐœ๊ฒฌํ–ˆ์Šต๋‹ˆ๋‹ค.
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.