https://github.com/Tencent-Hunyuan/Hy3



GitHub - Tencent-Hunyuan/Hy3: Hy3 (295B A21B), a leading reasoning and agent model in its size, with great cost efficiency.Hy3 (295B A21B), a leading reasoning and agent model in its size, with great cost efficiency. - Tencent-Hunyuan/Hy3github.com


https://huggingface.co/tencent/Hy3




tencent/Hy3 ยท Hugging FaceWeโ€™re on a journey to advance and democratize artificial intelligence through open source and open science.huggingface.co



24b0d121e09c28a8699fe8b115ef046f5e4e989daf

Hy3 ๋Š” ํ…์„ผํŠธ Hy ํŒ€์—์„œ ๊ฐœ๋ฐœํ•œ 2์–ต 9,500๋งŒ ๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ๊ฐ€์ง„ ์ „๋ฌธ๊ฐ€ ํ˜ผํ•ฉ ๋ชจ๋ธ(MoE)๋กœ, ํ™œ์„ฑ ํŒŒ๋ผ๋ฏธํ„ฐ๋Š” 210์–ต ๊ฐœ, MTP ๋ ˆ์ด์–ด ํŒŒ๋ผ๋ฏธํ„ฐ๋Š” 38์–ต ๊ฐœ์ž…๋‹ˆ๋‹ค. 4์›” ๋ง Hy3 ํ”„๋ฆฌ๋ทฐ ์ถœ์‹œ ์ดํ›„ 50๊ฐœ ์ด์ƒ์˜ ์ œํ’ˆ์—์„œ ํ”ผ๋“œ๋ฐฑ์„ ์ˆ˜์ง‘ํ•˜๊ณ , ๋” ๋†’์€ ํ’ˆ์งˆ์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ํ•™์Šต ํ›„ ๊ทœ๋ชจ๋ฅผ ํ™•์žฅํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ค๋Š˜, Hy3๋ฅผ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. Hy3๋Š” ๋น„์Šทํ•œ ๊ทœ๋ชจ์˜ ๋ชจ๋ธ๋ณด๋‹ค ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ์„ ๋ณด์—ฌ์ฃผ๋ฉฐ, ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜๊ฐ€ 2~5๋ฐฐ ๋” ๋งŽ์€ ํ”Œ๋ž˜๊ทธ์‹ญ ์˜คํ”ˆ์†Œ์Šค ๋ชจ๋ธ๊ณผ๋„ ๊ฒฝ์Ÿํ•  ๋งŒํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ๋‹ค์–‘ํ•œ ์ œํ’ˆ ๋ฐ ์ƒ์‚ฐ์„ฑ ์ž‘์—…์—์„œ ์œ ์šฉ์„ฑ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋˜๋Š” ๊ฒƒ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.