Sakana AI
This is the second in a series of blog posts presenting the results of Sakana AI’s research projects that were supported by the Japanese Ministry of Economy, Trade and Industry’s GENIAC supercomputing grant. Summary At Sakana AI, we draw inspiration from nature to extend the capabilities of foundation models. This was evidenced in our recent work on improving performance with evolutionary model me
LLMã‚’æ‚©ã¾ã›ã‚‹"Excel文書"ã‚’ã†ã¾ã扱ã†æ–¹æ³•
ã¯ã˜ã‚ã« æ ªå¼ä¼šç¤¾ãƒ•ã‚¡ãƒ¼ã‚¹ãƒˆãƒ»ã‚ªãƒ¼ãƒˆãƒ¡ãƒ¼ã‚·ãƒ§ãƒ³CTOã®ç”°ä¸(ã—ã‚ãã¾)ã§ã™ï¼ 弊社ã§ã¯è£½é€ æ¥å‘ã‘ã®RAGを使ã£ãŸãƒãƒ£ãƒƒãƒˆãƒœãƒƒãƒˆã®é–‹ç™ºã‚’è¡Œã£ã¦ã„ã¾ã™ãŒã€ RAGã§èªã¿å–ã‚Šã¥ã‚‰ã„ãªã¨æ„Ÿã˜ã¦ã„るドã‚ュメントãŒ"Excel文書"ã§ã™ã€‚ LLMã‚’æ‚©ã¾ã›ã‚‹"Excel文書"ã¨ã¯ ã“ã“ã§"Excel文書"ã¨å‘¼ã‚“ã§ã„るドã‚ュメントã¯ã€ ã€Œæ§‹é€ åŒ–ã•ã‚ŒãŸãƒ†ãƒ¼ãƒ–ルをä¿å˜ã—ã¦ã„ã‚‹Excelファイル〠ã§ã¯ãªã〠「 セルã«æ–‡æ›¸ã‚’書ã„ãŸã‚Šã€ã‚ªãƒ–ジェクトや画åƒã‚’挿入ã™ã‚‹ã“ã¨ã§ã€ã„ã‚ゆる一般的ãªæ–‡æ›¸ã‚’作æˆã—ã¦ã„ã‚‹Excelファイル 〠ã®ã“ã¨ã‚’呼んã§ã„ã¾ã™ã€‚ ãã‚‚ãも一般的ãªæ–‡æ›¸ä½œæˆã«ãŠã„ã¦ã¯Excelã§ã¯ãªãã€Wordを使ãˆã°ã‚ˆã„ã®ã§ã¯ï¼Ÿã¨æ€ã‚れるã‹ã‚‚ã§ã™ãŒã€ä»¥ä¸‹ã®ç‚¹ã§Excelã§æ–‡æ›¸ã®è³‡æ–™ã‚’作æˆã™ã‚‹ã“ã¨ã¯ãã‚Œãªã‚Šã«ä¾¿åˆ©ãªéƒ¨åˆ†ãŒã‚ã‚‹ã¨æ€ã£ã¦ã„ã¾ã™ã€‚ æž ã‚’ä½¿ã£ã¦ã€ã‚°ãƒ«ãƒ¼ãƒ”ングã™ã‚‹ã“ã¨ã§ãƒ‰ã‚ュメントã®æ§‹æˆãŒè¦‹ã‚„ã™ããªã‚‹
Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters
Enabling LLMs to improve their outputs by using more test-time computation is a critical step towards building generally self-improving agents that can operate on open-ended natural language. In this paper, we study the scaling of inference-time computation in LLMs, with a focus on answering the question: if an LLM is allowed to use a fixed but non-trivial amount of inference-time compute, how muc
勉強会「ã€LangChainゆる勉強会#15】LangChainã«ã‚ˆã‚‹ãƒ—ãƒãƒ³ãƒ—ト最é©åŒ– 「Promptimã€ã€ã‚’開催ã—ã¾ã—㟠#StudyCo - Generative Agents Tech Blog
ジェãƒãƒ©ãƒ†ã‚£ãƒ–エージェンツã®å¤§å¶‹ã§ã™ã€‚ é‹å–¶ã—ã¦ã„る勉強会コミュニティStudyCoã§ã€Œã€LangChainゆる勉強会#15】LangChainã«ã‚ˆã‚‹ãƒ—ãƒãƒ³ãƒ—ト最é©åŒ– 「Promptimã€ã€ã¨ã„ã†ã‚¤ãƒ™ãƒ³ãƒˆã‚’開催ã—ã¾ã—ãŸã€‚ studyco.connpass.com アーカイブ動画ã¯ã“ã¡ã‚‰ã§ã™ã€‚ youtube.com 今回ã¯LangChainã®å…¬å¼ãƒ–ãƒã‚°ã§ç´¹ä»‹ã•ã‚ŒãŸãƒ—ãƒãƒ³ãƒ—ト最é©åŒ–ツール「Promptimã€ã‚’å‹•ã‹ã—ã¦ã¿ã¾ã—ãŸã€‚ Promptimã«ã¤ã„ã¦ã®LangChainå…¬å¼ãƒ–ãƒã‚°ã¯ã“ã¡ã‚‰ã§ã™ã€‚ blog.langchain.dev Promptimã®GitHubリãƒã‚¸ãƒˆãƒªã¯ã“ã¡ã‚‰ã§ã™ã€‚ github.com 今回ã®ãƒã‚¤ãƒ³ãƒˆ Promptimã®æ¦‚è¦ Promptimã®æ¦‚è¦ã¯ã€ä¸Šè¨˜ã®å…¬å¼ãƒ–ãƒã‚°ã‚„GitHubã®READMEã«æ›¸ã‹ã‚Œã¦ã„ã¾ã™ã€‚ ç°¡å˜ã«è¨€ãˆã°LangSmithを使ã£ãŸãƒ—ãƒãƒ³ãƒ—ト最é©åŒ–
OpenAIã®è’¸ç•™æ©Ÿèƒ½(Model Distillation)を使ã£ã¦é‹ç”¨ä¸ã®LLMã®ã‚³ã‚¹ãƒˆã‚’削減ã™ã‚‹å–り組ã¿
ç§ãŸã¡ã¯ä»¥å‰ã«ã‚‚GPT-4o-miniをファインãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã—ã¦é‹ç”¨ã—ã¦ã„ã‚‹ã¨ã„ã†ãŠè©±ã‚’シェアã—ã¦å¤§ããªå響ã„ãŸã ãã¾ã—ãŸã€‚ 上ã®è¨˜äº‹ã§ã¯ã€GPT-4oã®å‡ºåŠ›ã‚’人手ã§ä¿®æ£ï¼ˆã‚¢ãƒŽãƒ†ãƒ¼ã‚·ãƒ§ãƒ³ï¼‰ã—ã¦ã€GPT-4-miniをファインãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã™ã‚‹é‹ç”¨ã‚’ã”紹介ã—ã¦ã„ã¾ã™ãŒã€è’¸ç•™ã¯GPT-4oã®å‡ºåŠ›ã‚’ãã®ã¾ã¾ä½¿ç”¨ã™ã‚‹ã®ãŒå¤§ããªé•ã„ã§ã™ã€‚ ã¾ãŸã€ä»Šå›žã®ç™ºè¡¨ã®è‚ã¯ã€ã“ã®è’¸ç•™ã‚’OpenAIã®Dashboad上ã§ç°¡å˜ã«è¡Œãˆã‚‹ã‚ˆã†ã«ãªã£ãŸã“ã¨ã§ã™ã€‚ ã§ã™ãŒã€ä»–社ã•ã‚“ã§ã‚‚実際ã«è’¸ç•™ã‚’使ã£ã¦ã„ã‚‹ã¨ã„ã†ãŠè©±ã¯ã‚ã¾ã‚Šèžãªã„ã®ã§ã€è’¸ç•™ã®åŸºæœ¬ã¨Model Distillationã®ä½¿ã„方をã”紹介ã—ã¤ã¤ã€ä½¿ã„所ã¨æ³¨æ„点をシェアã—ãŸã„ã¨æ€ã„ã¾ã™ã€‚ o1-previewã¨o1-miniãŒåŒæ™‚ã«ç™ºè¡¨ã•ã‚ŒãŸã“ã¨ã‚’見ã¦ã‚‚ã€ä»Šå¾Œã‚‚"高性能ã§é«˜ä¾¡ãªãƒ¢ãƒ‡ãƒ«"ã¨"å°‘ã—性能ã¯åŠ£ã‚‹ãŒå®‰ä¾¡ãªãƒ¢ãƒ‡ãƒ«"ã¯ï¼ˆå¤šå°‘ã®æ™‚期ã®é•ã„ãŒã‚ã£ã¦ã‚‚）セットã§ãƒªãƒªãƒ¼ã‚¹
The Rise of LLMOps
第11回 Data-Centric AI勉強会 ~MLOps勉強会コラボ回~ / 第46回 MLOps 勉強会 ã®ç™»å£‡è³‡æ–™ã§ã™
Introduction to GPT-4o and GPT-4o mini | OpenAI Cookbook
GPT-4o ("o" for "omni") and GPT-4o mini are natively multimodal models designed to handle a combination of text, audio, and video inputs, and can generate outputs in text, audio, and image formats. GPT-4o mini is the lightweight version of GPT-4o. Background Before GPT-4o, users could interact with ChatGPT using Voice Mode, which operated with three separate models. GPT-4o integrates these capabil
ランã‚ング
ランã‚ング
ランã‚ング
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
aie-book/resources.md at main · chiphuyen/aie-book
Large Concept Models: Language Modeling in a Sentence Representation Space | Research - AI at Meta
GitHub - gpt-omni/mini-omni: open-source multimodal large language model that can hear, talk while thinking. Featuring real-time end-to-end speech input and streaming audio output conversational capabilities.
ZenML - LLMOps Database
https://openai.com/index/openai-o1-system-card/
An update on our general capability evaluations
5-Day Gen AI Intensive Course with Google Learn Guide
https://openai.com/index/learning-to-reason-with-llms/
The Amazon Nova family of models: Technical report and model card
GitHub - llm-jp/llm-jp-eval
Next-Gen Models Show Limited Gains, Real-Time Video Generation, China AI Chips Blocked, Transformer Training Streamlined
Next-Gen Models Show Limited Gains, Real-Time Video Generation, China AI Chips Blocked, Transformer Training Streamlined
What is an AI Engineer? - Gradient Flow
{{#tags}}- {{label}}
{{/tags}}