GitHub - huggingface/setfit: Efficient few-shot learning with Sentence Transformers
SetFit is an efficient and prompt-free framework for few-shot fine-tuning of Sentence Transformers. It achieves high accuracy with little labeled data - for instance, with only 8 labeled examples per class on the Customer Reviews sentiment dataset, SetFit is competitive with fine-tuning RoBERTa Large on the full training set of 3k examples 🤯! Compared to other few-shot learning methods, SetFit ha
ã€Model Context Protocol】ClaudeãŒå…¨ã¦ã®ã‚¢ãƒ—リã¨ç¹‹ãŒã‚‹ï¼ï¼ŸSlackã‚„GitHubã¨é€£æºã—ã¦ã¿ãŸ | WEEL
Introducing the Model Context Protocol (MCP) An open standard we've been working on at Anthropic that solves a core challenge with LLM apps – connecting them to your data. No more building custom integrations for every data source. MCP provides one protocol to connect them all: pic.twitter.com/kYsivQyPDq — Alex Albert (@alexalbert__) November 25, 2024 Model Context Protocolã®æ¦‚è¦ Model Context Protoc
GoogleãŒãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã—ã‚„ã™ã„ビジュアル言語モデル「PaliGemma 2ã€ã‚’リリース
GoogleãŒ2024å¹´12月5æ—¥ã«ã€ã‚ªãƒ¼ãƒ—ンã‹ã¤è»½é‡ãªè¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒGemma 2ã€ã‚’ベースã«è¦–è¦šæ©Ÿèƒ½ã‚’è¿½åŠ ã—ãŸãƒ“ジュアル言語モデルã®ã€ŒPaliGemma 2ã€ã‚’発表ã—ã¾ã—ãŸã€‚ Introducing PaliGemma 2: Powerful Vision-Language Models, Simple Fine-Tuning - Google Developers Blog https://developers.googleblog.com/en/introducing-paligemma-2-powerful-vision-language-models-simple-fine-tuning/ Welcome PaliGemma 2 – New vision language models by Google https://huggingface.co/blog/paligemma
Llama 3.1 Swallow – Swallow LLM
Llama 3.1 Swallow Llama 3.1 Swallowã¯Llama 3.1ã®è‹±èªžã®èƒ½åŠ›ã‚’ç¶æŒã—ãªãŒã‚‰ã€æ—¥æœ¬èªžã®èƒ½åŠ›ã‚’強化ã—ãŸå¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ« (8B, 70B) ã§ã™ã€‚モデルã®ãƒ‘ラメータ（é‡ã¿ï¼‰ãŒHuggingFace上ã§å…¬é–‹ã•ã‚Œã¦ã„ã¾ã™ã®ã§ã€Llama 3.1ライセンスã«å¾“ã†é™ã‚Šã€ç ”究や商æ¥ç›®çš„ãªã©ã§åˆ©ç”¨ã§ãã¾ã™ï¼ˆãŸã ã—ã€æŒ‡ç¤ºãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°æ¸ˆã¿ãƒ¢ãƒ‡ãƒ«ã®åˆ©ç”¨ã«ãŠã„ã¦ã¯ã€Llama 3.1ライセンスã«å¾“ã£ãŸã†ãˆã§ã€Gemma利用è¦ç´„ã®åˆ©ç”¨åˆ¶é™ã«æŠµè§¦ã—ã¦ã¯ã„ã‘ã¾ã›ã‚“）。Llama 3.1 Swallowã¯ç±³Meta社ã®Llama 3.1をベースã«ã€æ±äº¬ç§‘å¦å¤§å¦æƒ…å ±ç†å·¥å¦é™¢ã®å²¡å´Žç ”究室ã¨æ¨ªç”°ç ”究室ã€å›½ç«‹ç ”究開発法人産æ¥æŠ€è¡“ç·åˆç ”究所ã®ç ”究ãƒãƒ¼ãƒ ã§é–‹ç™ºã•ã‚Œã¾ã—ãŸã€‚Built with Llama. View on HuggingFace Hugging Face Llama 3.1
PLaMo-13Bを公開ã—ã¾ã—㟠- Preferred Networks Research & Development
Preferred Networksã§ã¯ã€9月28æ—¥ã«PLaMo-13Bã¨ã„ã†å¤§è¦æ¨¡ãªè¨€èªžãƒ¢ãƒ‡ãƒ« (LLM) を公開ã—ã¾ã—ãŸã€‚公開ã•ã‚Œã¦ã„ã‚‹ä»–ã®ãƒ¢ãƒ‡ãƒ«ã¨æ¯”較ã—ã¦ã€æ—¥è‹±ï¼’言語をåˆã‚ã›ãŸèƒ½åŠ›ã§ä¸–界トップレベルã®é«˜ã„性能を示ã—ã¦ã„ã¾ã™ã€‚実際ã«å¦ç¿’を回ã™ã¾ã§ã®æŠ€è¡“開発ã«ã¯è‡ªç¤¾ã‚¹ãƒ¼ãƒ‘ーコンピューターã§ã‚ã‚‹MN-2を利用ã—ã€å¦ç¿’ã¯AI橋渡ã—クラウド（AI Bridging Cloud Infrastructureã€ABCI)　ã®â€ç¬¬ä¸€å›ž 大è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«æ§‹ç¯‰æ”¯æ´ãƒ—ãƒã‚°ãƒ©ãƒ â€ã‚’利用ã—ã¦è¡Œã„ã¾ã—ãŸã€‚ 今後ã€PLaMo-13Bを基ã«ã—ãŸäº‹å‰å¦ç¿’モデルやã€æŒ‡ç¤ºå¦ç¿’ã‚’è¡Œã£ãŸãƒ¢ãƒ‡ãƒ«ã«ã¤ã„ã¦ã‚‚公開を予定ã—ã¦ã„ã¾ã™ã€‚ PLaMo-13Bã®æ¦‚è¦ PLaMo-13Bã¯ç´„130億個ã®ãƒ‘ラメータã‹ã‚‰ãªã‚‹è¨€èªžãƒ¢ãƒ‡ãƒ«ã§ã™ã€‚ PLaMo-13Bã¯æ—¥æœ¬èªžãƒ»è‹±èªžã®2ã¤ã®è¨€èªžã®ãƒ™ãƒ³ãƒãƒžãƒ¼ã‚¯ã‚¿ã‚¹ã‚¯ã§é«˜ã„性能を示ã—ã¦ã„ã¾ã™ã€‚日本ã§ä½¿ã‚れるL
ä¸å‹•ç”£è²¡å‹™ãƒ¢ãƒ‡ãƒ«â‘ ：ã¯ã˜ã‚ã«ï½œinvestdebater
財務モデルã¨èžã‘ã°åŸºæœ¬çš„ã«ã¯ä¼æ¥ã¸ã®æŠ•è³‡ã™ã‚‹ãŸã‚ã®è²¡å‹™ãƒ¢ãƒ‡ãƒ«ã‚’イメージã™ã‚‹æ–¹ãŒå¤šãã€LBOモデルã«ã¤ã„ã¦ã¯æ—¥æœ¬èªžã®æ•™æも英語ã®ã‚‚ã®ã«æ¯”ã¹ã¦å°‘ãªã„ã¨ã¯ã„ãˆã€è‰²ã€…ãªæ–¹ãŒè§£èª¬è³‡æ–™ã‚’出ã—ã¦ãŠã‚Šã€ã‹ãªã‚Šä¸€èˆ¬çš„ã«ãªã£ã¦ããŸã‚ˆã†ã«æ„Ÿã˜ã¦ã„る。 モデリングテストを行ã£ã¦ã‚‚ã€LBOモデルã¯ä½•ã¨ãªãå½¢ã«ã§ãる人ãŒãã‚Œãªã‚Šã«å¢—ãˆã¦ãã¦ãŠã‚Šã€å°‘ã—å‰ã®BSã‚’ãƒãƒ©ãƒ³ã‚¹ã•ã›ã‚‹ã“ã¨ãŒã§ãã¾ã›ã‚“ã€ã¨ã„ã†äººãŒå¤§åŠã ã£ãŸæ™‚代ã«æ¯”ã¹ã‚‹ã¨LBOモデル/財務モデルãŒæ¯”較的一般化ã—ã¦ããŸã¨æ€ã‚れる。 ã—ã‹ã—ä¸å‹•ç”£ã®è²¡å‹™ãƒ¢ãƒ‡ãƒ«ã¯ã€ä¸å‹•ç”£æŠ•è³‡ãŒæ—¥æœ¬ã§ã¯ã¨ã¦ã‚‚活発ã§ã‚ã‚‹ã®ã«ã‚‚é–¢ã‚らãšã€æ—¥æœ¬èªžã®æ•™æãŒã»ã¨ã‚“ã©è¦‹ã¤ã‹ã‚‰ãªã„。æらãä¸å‹•ç”£è²¡å‹™ãƒ¢ãƒ‡ãƒ«ã®æ–¹ãŒã€LBOモデルよりも必è¦ã¨ãªã‚‹äººã¯ãŸãã•ã‚“ã„ã‚‹ã¯ãšã§ã‚る。 外資系ä¸å‹•ç”£ãƒ•ã‚¡ãƒ³ãƒ‰ã«ã¦å‹¤å‹™ã™ã‚‹æ–¹ã¯å¿…ãšå¿…è¦ã¨ãªã‚Šã€ãれ以外ã«ã‚‚事æ¥ä¼šç¤¾ã§ä¸å‹•ç”£æŠ•è³‡ã‚’è¡Œã†æ–¹ã€ä¸å‹•ç”£ä»²ä»‹ã®æ–¹ã€ä¸å‹•ç”£ãƒ•ã‚¡ãƒ³ãƒ‰ã®çµ„æˆã‚’生æ¥
ã€ã“ã‚ŒãŒã‚²ãƒ¼ãƒ ãƒã‚§ãƒ³ã‚¸ãƒ£ãƒ¼ã‹!】æ¾å°¾ç ”ã®Tanuki-8Bã¨Tanuki-8x8Bを試ã™ï½œshi3z
ãªãœã‚¿ãƒŒã‚ãªã®ã‹? ãã®è¬Žã¯è¬Žã®ã¾ã¾ã ãŒã€ã¨ã«ã‹ã日本語性能ãŒGemini1.5Proã«æ¬¡ãŽã€å°‘ã—å‰ã®GPT-4よりも高ã„上ã«å•†ç”¨åˆ©ç”¨å¯èƒ½ã¨ã„ã†å¤ªã£è…¹ä»•æ§˜ãªã®ã§ä½¿ã‚ãªã„手ã¯ãªã„。むã—ã‚ã“ã‚Œã•ãˆã‚ã‚Œã°ã‚‚ã†èª°ã‚‚GPTã«èª²é‡‘ã—ãªãã¦æ¸ˆã‚€ã€ãã‚“ãªæœªæ¥ãŒæ¥ã‚‹ã®ã‹ã‚‚ã—ã‚Œãªã„。 ã—ã‹ã—ã€Tanukiã¯ç‰¹æ®Šãªä½•ã‹ã‚’ã—ã¦ã‚‹ã‚‰ã—ãMLXã«ç°¡å˜ã«ã‚³ãƒ³ãƒãƒ¼ãƒˆã§ããšvllmã§å‹•ã‹ã™ã¨ãã‚‚ã¡ã‚‡ã£ã¨é”æ”¹é€ ã—ãŸvllmãŒå¿…è¦ã«ãªã‚‹ã¨ã„ã†ã€‚ 最近ãƒãƒ¼ã‚«ãƒ«ã¥ã„ã¦ã‚‹å¾è¼©ã¨ã—ã¦ã¯ã§ãã‚Œã°MLXã§å‹•ã‹ã—ãŸã„ã®ã ãŒMLXãŒã¾ã Tanukiã«å¯¾å¿œã—ã¦ãªã„(ã—ã€ãã‚‚ãも何をã™ã‚Œã°Tanuki対応ã«ã§ãã‚‹ã®ã‹ã‚ˆãã‚ã‹ã‚‰ãªã„)ã®ã§ã€ã¨ã‚Šã‚ãˆãšã€ŒéžæŽ¨å¥¨ã€ã¨ã•ã‚Œã¦ã¯ã„ã‚‹ã‚‚ã®ã®ã€Macã§ã‚‚ギリギリ動ããã†ãªGGUF版を動ã‹ã—ã¦ã¿ãŸã€‚ from llama_cpp import Llama filename = "Tanuki-8B-dpo-v1.
メタãŒæ¡é•ã„ã®AIインフラ構築　「完全ãªæ±Žç”¨çŸ¥èƒ½ã€ã¸ - 日本経済新èž
米メタãŒäººå·¥çŸ¥èƒ½ï¼ˆAI）å‘ã‘ã®ITï¼ˆæƒ…å ±æŠ€è¡“ï¼‰ã‚¤ãƒ³ãƒ•ãƒ©æŠ•è³‡ã‚’å†æ‹¡å¤§ã—始ã‚ãŸã€‚リストラã®ãŸã‚2023å¹´ã¯è¨å‚™æŠ•è³‡ãªã©ã®è³‡æœ¬çš„支出（CAPEX）を減らã—ã¦ã„ãŸãŒã€24å¹´ã¯å†ã³å¢—ã‚„ã—ã¦300億〜370億ドル（約4.4兆〜5.5兆円）を投ã˜ã‚‹ã€‚マーク・ザッカーãƒãƒ¼ã‚°æœ€é«˜çµŒå–¶è²¬ä»»è€…（CEO）ã¯ã€Œå®Œå…¨ãªæ±Žç”¨çŸ¥èƒ½ã®å®Ÿç¾ã‚’目指ã™ã€ã¨æ˜Žè¨€ã—ã¦ã„る。å‰å¹´ã‚ˆã‚Šã‚‚減ã£ãŸã‚‚ã®ã®ã€è¨å‚™æŠ•è³‡è‡ªä½“ã¯23å¹´ã‚‚å·¨é¡ã ã£ãŸã€‚åŒç¤¾ãŒ
日本語LLMã«ãŠã‘るトークナイザーã®é‡è¦æ€§ | データアナリティクスラボ
ELYZA-Japanese-Llama-2-7b ELYZA-Japanese-Llama-2-7bã¯ã€ELYZAãŒå…¬é–‹ã—ãŸæ—¥æœ¬èªžã«ç‰¹åŒ–ã—ãŸLLMã§ã™ã€‚ 公開ã•ã‚ŒãŸã®ã¯ä»¥ä¸‹ã®ãƒ¢ãƒ‡ãƒ«ã§ã™ã€‚ ELYZA-japanese-Llama-2-7bELYZA-japanese-Llama-2-7b-fastELYZA-japanese-Llama-2-7b-instructELYZA-japanese-Llama-2-7b-fast-instruct instruct：instruction tuningã«ã‚ˆã£ã¦è¿½åŠ å¦ç¿’ã•ã‚ŒãŸãƒ¢ãƒ‡ãƒ«ã€‚ fast：日本語ã®èªžå½™ã®è¿½åŠ ã«ã‚ˆã£ã¦å‡¦ç†ã‚’高速化ã—ãŸãƒ¢ãƒ‡ãƒ«ã€‚ モデルã®æ¦‚è¦ ELYZA-japanese-Llama-2-7bã¯Llama2をベースã¨ã—ã¦æ—¥æœ¬èªžå‡¦ç†ã®æ€§èƒ½ã‚’å‘上ã•ã›ã‚‹ãŸã‚ã«è¿½åŠ å¦ç¿’ã—ãŸãƒ¢ãƒ‡ãƒ«ã§ã™ã€‚英語ã§å¦ç¿’済ã¿ã®LLMã®è¨€èªžèƒ½åŠ›ã‚’引ã継ãã“ã¨ã§ã€å°‘ãª
google/siglip-base-patch16-256-multilingual · Hugging Face
SigLIP (base-sized model, multilingual) SigLIP model pre-trained on WebLi at resolution 256x256. It was introduced in the paper Sigmoid Loss for Language Image Pre-Training by Zhai et al. and first released in this repository. Disclaimer: The team releasing SigLIP did not write a model card for this model so this model card has been written by the Hugging Face team. Model description SigLIP is CLI
Grok-1 Model Card
Grok-1 is an autoregressive Transformer-based model pre-trained to perform next-token prediction. The model was then fine-tuned using extensive feedback from both humans and the early Grok-0 models. The initial Grok-1 has a context length of 8,192 tokens and is released in Nov 2023. Grok-1 is intended to be used as the engine behind Grok for natural language processing tasks including question ans
生æˆAIã®åŸºç›¤æŠ€è¡“を無償公開　æ±å¤§ãƒ»æ¾å°¾æ•™æŽˆã®ç ”究室 - 日本経済新èž
人工知能（AIï¼‰ç ”ç©¶ã®ç¬¬ä¸€äººè€…ã¨ã—ã¦çŸ¥ã‚‰ã‚Œã‚‹æ±äº¬å¤§å¦ã®æ¾å°¾è±Šæ•™æŽˆã®ç ”究室ã¯18æ—¥ã€ç”ŸæˆAIã®åŸºç›¤æŠ€è¡“ã§ã‚る「大è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ï¼ˆLLM）ã€ã‚’開発ã—ãŸã¨ç™ºè¡¨ã—ãŸã€‚日本語ã¨è‹±èªžã«å¯¾å¿œã—ã€å¤–部ã®ç ”究者らãŒåˆ©ç”¨ã§ãるよã†ç„¡å„Ÿã§å…¬é–‹ã—ãŸã€‚日本ã®ç”ŸæˆAI活用を後押ã—ã™ã‚‹ã€‚　米オープンAIã®ã€ŒChat（ãƒãƒ£ãƒƒãƒˆï¼‰GPTã€ãªã©ã®LLMã¯ä¸»ã«è‹±èªžã®æ–‡ç« ã‚’å¦ç¿’ã—ã¦ã„ã‚‹ãŸã‚ã€æ—¥æœ¬èªžã‚ˆã‚Šã‚‚英語ã§ç”Ÿæˆã•ã‚Œã‚‹æ–‡ç« ã®æ–¹ãŒç²¾åº¦
日本語画åƒè¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒJapanese InstructBLIP Alphaã€ã‚’リリースã—ã¾ã—㟠— Stability AI Japan
Stability AIã¯æ—¥æœ¬èªžå‘ã‘ç”»åƒè¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒJapanese InstructBLIP Alphaã€ã‚’一般公開ã—ã¾ã—ãŸã€‚入力ã—ãŸç”»åƒã«å¯¾ã—ã¦æ–‡å—ã§èª¬æ˜Žã‚’生æˆã§ãã‚‹ç”»åƒã‚ャプション機能ã«åŠ ãˆã€ç”»åƒã«ã¤ã„ã¦ã®è³ªå•ã‚’æ–‡å—ã§å…¥åŠ›ã™ã‚‹ã“ã¨ã§å›žç”ã™ã‚‹ã“ã¨ã‚‚ã§ãã¾ã™ã€‚ Japanese InstructBLIP Alpha「Japanese InstructBLIP Alphaã€ã¯ã€å…ˆæ—¥å…¬é–‹ã•ã‚ŒãŸæ—¥æœ¬èªžå‘ã‘指示応ç”言語モデル「Japanese StableLM Instruct Alpha 7Bã€ã‚’æ‹¡å¼µã—ãŸã€ç”»åƒã‚’å…ƒã«ã—ãŸãƒ†ã‚ストãŒç”Ÿæˆã•ã‚Œã‚‹ãƒ¢ãƒ‡ãƒ«ã§ã™ã€‚ 「Japanese InstructBLIP Alphaã€ã¯ã€é«˜ã„パフォーマンスãŒå ±å‘Šã•ã‚Œã¦ã„ã‚‹ç”»åƒè¨€èªžãƒ¢ãƒ‡ãƒ«InstructBLIPã®ãƒ¢ãƒ‡ãƒ«æ§‹é€ を用ã„ã¦ãŠã‚Šã¾ã™ã€‚å°‘ãªã„日本語データセットã§é«˜æ€§èƒ½ãªãƒ¢ãƒ‡ãƒ«ã‚’構築ã™ã‚‹ãŸã‚ã«ã€ãƒ¢ãƒ‡ãƒ«ã®ä¸€éƒ¨ã‚’大è¦æ¨¡ãª
ChatGPT（3.5）ã«åŒ¹æ•µã™ã‚‹ã€ŒLlama 2ã€ã‚’ãƒãƒ¼ã‚«ãƒ«PCã§å‹•ã‹ã—ã¦ã¿ãŸ
生æˆAIã®ãƒˆãƒƒãƒ—ランナーã¨ã„ãˆã°ã€ç±³OpenAIãŒæä¾›ã™ã‚‹GPT-4ãªã©ã‚’使ã£ãŸChatGPTã§ã™ãŒã€ãã®å¯¾æŠ—馬ã¨ã—ã¦æœŸå¾…ã•ã‚Œã¦ã„ã‚‹ã®ãŒç±³MetaãŒæä¾›ã™ã‚‹å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒLlama 2ã€ã§ã™ã€‚ ã“ã®Llama 2ã€GPT-3.5ã®3月1日時点ã®ãƒ¢ãƒ‡ãƒ«ã«åŒ¹æ•µã™ã‚‹æ€§èƒ½ã‚’æŒã£ã¦ã„ã‚‹ã¨ã„ã†ã®ãŒã‚¦ãƒªã§ã™ã€‚GPT-3.5ã¨ã„ãˆã°ã€ç„¡æ–™ç‰ˆã®ChatGPTã§ä½¿ã‚ã‚Œã¦ã„るモデルã§ã™ã€‚ãã‚ŒãŒã‚ªãƒ¼ãƒ—ンソースã¨ã—ã¦å…¬é–‹ã•ã‚ŒãŸã®ã§ã™ã‹ã‚‰ã€è¡æ’ƒçš„ã§ã™ã€‚ ã•ã‚‰ã«ã€é«˜æ€§èƒ½ãªã ã‘ã§ãªãモデルサイズãŒå°ã•ã„ã“ã¨ã‚‚特徴ã§ã™ã€‚GPT-3ã®ãƒ‘ラメータ数ã¯1750億（175B）ã€GPT-3.5ã¯æœªå…¬é–‹ã§ã¯ã‚ã‚‹ã‚‚ã®ã®3550億（355B）ã¨æŽ¨å®šã•ã‚Œã¦ã„ã¾ã™ã€‚一方ã§ã€Llama 2ã¯ã€700億（70B）パラメータã§ã€GPT-3.5並をã†ãŸã£ã¦ã„ã¾ã™ã€‚ パラメータãŒå°ã•ããªã‚Œã°å¿…è¦ãªGPUã®ãƒ¡ãƒ¢ãƒªã‚‚å°ã•ããªã‚Šã¾ã™ã€‚GPT-3.5ã¯ãƒ‡ãƒ¼
最近ã®è©±é¡Œã«ã‚‚詳ã—ã„14億パラメータã®æ—¥æœ¬èªžLLMã®å…¬é–‹
最近ã®è©±é¡Œã«ã‚‚詳ã—ã„14億パラメータã®æ—¥æœ¬èªžLLMã®å…¬é–‹
The Falcon has landed in the Hugging Face ecosystem
Falcon is a new family of state-of-the-art language models created by the Technology Innovation Institute in Abu Dhabi, and released under the Apache 2.0 license. Notably, Falcon-40B is the first “truly open†model with capabilities rivaling many current closed-source models. This is fantastic news for practitioners, enthusiasts, and industry, as it opens the door for many exciting use cases. Note
GitHub - karpathy/llama2.c: Inference Llama 2 in one file of pure C
Have you ever wanted to inference a baby Llama 2 model in pure C? No? Well, now you can! Train the Llama 2 LLM architecture in PyTorch then inference it with one simple 700-line C file (run.c). You might think that you need many billion parameter LLMs to do anything useful, but in fact very small LLMs can have surprisingly strong performance if you make the domain narrow enough (ref: TinyStories p
æ±å¤§ç™ºã‚¹ã‚¿ãƒ¼ãƒˆã‚¢ãƒƒãƒ—ã€67億パラメーターã®æ—¥æœ¬èªžLLMã‚’OSSã§å…¬é–‹
æ±äº¬å¤§å¦ç™ºã®ã‚¹ã‚¿ãƒ¼ãƒˆã‚¢ãƒƒãƒ—ä¼æ¥ã§ã‚ã‚‹Lightblue（ライトブルー）ã¯ã€å…¬é–‹ãƒ¢ãƒ‡ãƒ«ã¨ã—ã¦ã¯å›½å†…最大è¦æ¨¡ã®67億パラメーターã®æ—¥æœ¬èªžå¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã‚’開発ã—ã€ã‚ªãƒ¼ãƒ—ンソース・ソフトウェアã¨ã—ã¦å…¬é–‹ã—ãŸã€‚ライセンスã¯Apache 2.0。 ã“ã®è¨€èªžãƒ¢ãƒ‡ãƒ«ã¯ã€ç±³ãƒ¢ã‚¶ã‚¤ã‚¯ML （MosaicML）ãŒå…¬é–‹ã—ãŸå¤šè¨€èªžå¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ«ã€ŒMPT-7Bã€ã‚’基ã«ã—ãŸã‚‚ã®ã€‚グーグルãŒé–‹ç™ºã—ãŸå¤šè¨€èªžãƒ‡ãƒ¼ã‚¿ã‚»ãƒƒãƒˆã€ŒMC4ã€ã‚’ã‚¢ãƒ¬ãƒ³äººå·¥çŸ¥èƒ½ç ”ç©¶æ‰€ï¼ˆAllen Institute for AI）ãŒãã‚Œãžã‚Œã®è¨€èªžã”ã¨ã«åˆ©ç”¨å¯èƒ½ã«ã—ãŸã‚µãƒ–セットã®æ—¥æœ¬èªžéƒ¨åˆ†ã‚’使ã£ã¦è¿½åŠ å¦ç¿’ã—ãŸã€‚ Lightblueã¯ã€ä»Šå›žå…¬é–‹ã—ãŸãƒ¢ãƒ‡ãƒ«ã‚’法人å‘ã‘ã«æä¾›ã™ã‚‹ã€‚æ¥ç•Œç”¨èªžã‚„部署特有ã®å°‚門用語ã€æ…£ç¿’ãªã©ã«åˆã‚ã›ã¦è¨“練・調整ã™ã‚‹ã“ã¨ã§ã€ä¼æ¥ã‚„部署ã«ã‚ˆã£ã¦ç•°ãªã‚‹è¦æœ›ã«å¿œã˜ã‚‹ã¨ã„ã†ã€‚åŠ ãˆã¦ã€è‡ªç¤¾ã‚µãƒ¼ãƒ“スã®æ供も予定ã—ã¦ã„ã‚‹ã¨ã®ã“ã¨ã 。 （笹田） 6 人
メンテナンス
ãŠçŸ¥ã‚‰ã›
障害
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
Effortless tracking of your Vertex AI model training | Google Cloud Blog
A Comprehensive Survey of Compression Algorithms for Language Models
{{#tags}}- {{label}}
{{/tags}}