AbdulElahGwaith's picture
Upload folder using huggingface_hub
a9bd396 verified

Gemma2 [[gemma2]]

๊ฐœ์š” [[overview]]

Gemma2 ๋ชจ๋ธ์€ Google์˜ Gemma2 ํŒ€์ด ์ž‘์„ฑํ•œ Gemma2: Open Models Based on Gemini Technology and Research์—์„œ ์ œ์•ˆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ํŒŒ๋ผ๋ฏธํ„ฐ ํฌ๊ธฐ๊ฐ€ ๊ฐ๊ฐ 90์–ต(9B)๊ณผ 270์–ต(27B)์ธ ๋‘ ๊ฐ€์ง€ Gemma2 ๋ชจ๋ธ์ด ์ถœ์‹œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

๋ธ”๋กœ๊ทธ ๊ฒŒ์‹œ๋ฌผ์˜ ์ดˆ๋ก์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค:

์ด์ œ ์šฐ๋ฆฌ๋Š” ์ „ ์„ธ๊ณ„์˜ ์—ฐ๊ตฌ์ž์™€ ๊ฐœ๋ฐœ์ž๋“ค์—๊ฒŒ Gemma 2๋ฅผ ๊ณต์‹์ ์œผ๋กœ ์ถœ์‹œํ•ฉ๋‹ˆ๋‹ค. 90์–ต(9B)๊ณผ 270์–ต(27B) ํŒŒ๋ผ๋ฏธํ„ฐ ํฌ๊ธฐ๋กœ ์ œ๊ณต๋˜๋Š” Gemma 2๋Š” 1์„ธ๋Œ€๋ณด๋‹ค ๋” ๋†’์€ ์„ฑ๋Šฅ๊ณผ ์ถ”๋ก  ํšจ์œจ์„ฑ์„ ์ œ๊ณตํ•˜๋ฉฐ, ์ƒ๋‹นํ•œ ์•ˆ์ „์„ฑ ํ–ฅ์ƒ์„ ํฌํ•จํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ์‹ค 270์–ต ๊ทœ๋ชจ์˜ ๋ชจ๋ธ์€ ํฌ๊ธฐ๊ฐ€ ๋‘ ๋ฐฐ ์ด์ƒ์ธ ๋ชจ๋ธ๊ณผ ๋น„๊ตํ•ด๋„ ๊ฒฝ์Ÿ๋ ฅ ์žˆ๋Š” ๋Œ€์•ˆ์„ ์ œ๊ณตํ•˜๋ฉฐ, ์ด๋Š” ์ž‘๋…„ 12์›”๊นŒ์ง€๋งŒ ํ•ด๋„ ๋…์  ๋ชจ๋ธ์—์„œ๋งŒ ๊ฐ€๋Šฅํ–ˆ๋˜ ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

ํŒ:

  • ์›๋ณธ ์ฒดํฌํฌ์ธํŠธ๋Š” ๋ณ€ํ™˜ ์Šคํฌ๋ฆฝํŠธ src/transformers/models/Gemma2/convert_Gemma2_weights_to_hf.py๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ณ€ํ™˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

์ด ๋ชจ๋ธ์€ Arthur Zucker, Pedro Cuenca, Tom Arsen์ด ๊ธฐ์—ฌํ–ˆ์Šต๋‹ˆ๋‹ค.

Gemma2Config [[transformers.Gemma2Config]]

[[autodoc]] Gemma2Config

Gemma2Model [[transformers.Gemma2Model]]

[[autodoc]] Gemma2Model - forward

Gemma2ForCausalLM [[transformers.Gemma2ForCausalLM]]

[[autodoc]] Gemma2ForCausalLM - forward

Gemma2ForSequenceClassification [[transformers.Gemma2ForSequenceClassification]]

[[autodoc]] Gemma2ForSequenceClassification - forward

Gemma2ForTokenClassification [[transformers.Gemma2ForTokenClassification]]

[[autodoc]] Gemma2ForTokenClassification - forward