You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session. You switched accounts on another tab or window. Reload to refresh your session. Dismiss alert
â»æ¬ãã¼ã¸ã¯ãã¢ãã£ãªã¨ã¤ãåºåãå©ç¨ãã¦ãã¾ãã ãã¯ããã¦ã§ãããã¾ã§ã§ãããã®æ¬æ¸ã¿ã¤ãã«éããèªãã§3æéã»ã©ã§Stable Diffusionã®ç»åçæãæãéãã«ã§ããããã«ãªãã¾ããï¼ Stable Diffusionã§ã©ããªãã¨ãã§ããã®ãã¯ç¥ã£ã¦ãã¾ããããå®éã«èªåã®PCã«ã»ããã¢ããããããã³ããã«ããç»åçæãã¯ãããè²å¡ãæå®ãç»é¢¨æå®ããã£ã©ãAIã«å¦ç¿ããã¦ã®ç»åçæã¾ã§ã§ãã¾ãããStable Diffusionåå¿è ã®äººã«ãå§ãã®è§£èª¬æ¸ãç´¹ä»ãã¾ãã ã¾ããå°ããã£ã¦ã¿ããã©Textual InversionãControlNetãLoRAãªã©ããã¾ã使ããªãããããªäººã«ãããªããå§ãã§ãã æ¬æ¸ã¯ãã¯ããã¦ã§ãããã¾ã§ã§ãããã¨ããã¿ã¤ãã«éããåå¿è åãã®è§£èª¬æ¸ã§ããããã¦ãæ¬æ ¼æ´»ç¨ãã¨ããããã«ããã 使ããããã«ãªãã ãã§ãªããå¡ãã®æ¸ãè¾¼ã¿æãè²å¡
å種Prompt Engineeringã®æ¥æ¬èªå®ä¾éï¼Zero-CoTãmockãReActãToTãMetacogãStep BackãIEPãªã©ï¼Pythonæ©æ¢°å¦ç¿å ¥éChatGPTLLM CoTãZero-CoTãToTãmockãReActãStep BackãMetacogãIEPãªã©ãå種Prompt Engineeringææ³ã®æ¦èª¬ã¨ãæ¥æ¬èªã§ã®å®éã®ããã³ããä¾ãã¾ã¨ããè¨äºã§ãã å種Prompt Engineeringææ³ãæ¥æ¬èªã§å®è£ ãããæ¹åãã®è¨äºã¨ãªãã¾ãã æ¬è¨äºã§åãæ±ãææ³ã¯ä»¥ä¸ã®éãã§ãã é ç®æ°ãå¤ããããè¨äºå³ä¸ã®ç®æ¬¡ãªã³ã¯ããæ´»ç¨ãã ããã æ¬è¨äºã®å 容 01: é常ã®Prompt 02: Few-shot Learning 03: CoTï¼Chain of Thoughtï¼ 04: åºåå½¢å¼ã®æå®æ¹æ³ 05: Zero-shot CoTï¼â
ãæé©åæ³ã第15å(ã²ã¹ããã¼ã¯), 2023å¹´1æ20æ¥, åå¿ç¤¾å¤§å¦.
Let's enjoy math and physics! YouTubeãã£ã³ãã«ãäºåæ ¡ã®ããªã§å¦ã¶ã大å¦ã®æ°å¦ã»ç©çããï¼ç¥ç§°ï¼ã¨ãããªï¼ã®ãã¼ã ãã¼ã¸ã¸ããããï¼ ãã®ãµã¤ãã§ã¯ãåç»ã®ç´¹ä»ãããããã®åèæ¸ããä¼ããã¦ããã¾ãã大å¦ã®æ°å¦ã»ç©çãç§ãã¡ã¨æ¥½ãã¿ã¾ãããï¼ Channel Funding
LLMããã®åºåå½¢å¼ã¯ãããã³ããã§ç´æ¥æå®ããæ¹æ³ãã·ã³ãã«ã§ãããLLMã®åºåãå®å®ããªãå ´åãã¾ã¾ããã¨æãã®ã§ãLangChainã«ã¯ãæ§é åããåºåå½¢å¼ãæå®ã§ãããã¼ãµã¼æ©è½ãããã¾ãã LangChainã«ã¯ãããã¤ãåºåãã¼ãµã¼ãããã¾ãããæ±ç¨çã«ä½¿ãéããããããªãPydantic/JSON åºåãã¼ãµã¼ã試ãã¦ã¿ã¾ããã ï¼ï¼ã©ã¤ãã©ãªã»ãæºåfrom langchain.llms import OpenAI from langchain.output_parsers import PydanticOutputParser from pydantic import BaseModel, Field, validator from typing import Listä»åã®ä¸»å½¹ã¯ããPydanticOutputParserãã§ãã model = OpenAI(mode
å®ã¯ãã®LoRAã®ä½ãæ¹ã¯ä½è æ§ãæ å ±ãå ±æãã¦ãã ãã£ã¦ãããç 究ç±å¿ãªæ¹ã®éã§ã¯èå³æ·±ãç 究ãã¼ãã¨ãã¦ãã°ããåã«è©±é¡ã«ãªã£ã¦ãã¾ããããã®LoRAã®ä½ææ¹æ³ã¯ãã³ãã¼æ©LoRAå¦ç¿æ³ãã¨å¼ã°ãã¦ãã¦ãå¿ç¨ããã¨ç»é¢¨ã調æ´å¯è½ãªLoRAãä½ããã¨ã®ãã¨ã ã£ãã®ã§ä»æ´ãªããç§ã試ãã¦ã¿ããã¨ã«ãã¾ããã ããã§ã¯ãã®ã³ãã¼æ©LoRAå¦ç¿æ³ã使ã£ãLoRAã®ä½ãæ¹ãã¾ã¨ãã¦ã¿ã¾ããã
StarCoderã¨StarCoderBaseã¯ã80以ä¸ã®ããã°ã©ãã³ã°è¨èªãGitã³ããããGitHub IssuesãJupyter Notebookãªã©ãGitHubããã©ã¤ã»ã³ã¹ã¨ãã¦è¨±å¯ãããã³ã¼ãã®ãã¼ã¿ã§ãã¬ã¼ãã³ã°ï¼æ©æ¢°å¦ç¿ï¼ããã¦ãããMeta製LLMã®ãLLaMAãã¨åæ§ã«ã1å ã®ãã¼ã¯ã³ã«å¯¾ãã¦ç´150åã®ãã©ã¡ã¼ã¿ã¼ã¢ãã«ããã¬ã¼ãã³ã°ãããHugging Faceã¯350åã®Pythonãã¼ã¯ã³ç¨ã«StarCoderBaseã¢ãã«ã微調æ´ããStarCoderãä½æããã StarCoderBaseã¯ãä¸è¬çãªããã°ã©ãã³ã°ãã³ããã¼ã¯ã§ãæ¢åã®ãªã¼ãã³ã³ã¼ãLLMãããåªããããã©ã¼ãã³ã¹ã示ããOpenAIã®ãcode-cushman-001ãï¼ãGitHub Copilotãã®åæãã¼ã¸ã§ã³ãå¼·åãããªãªã¸ãã«ã®ãCodexãã¢ãã«ï¼ã®ãããªã¯ãã¼ãºãã¢ã
ãã«ã¬ã¯ã¯AIãå«ããæå 端æè¡ãæçè·é¢ã§å¦ã¶ãã¨ãã§ããå¦ç¿ãã©ãããã©ã¼ã ã§ããä»ãªãæ°è¦ã¦ã¼ã¶ã¼ç»é²ï¼ç¡æï¼ã§ãUdemy ã§ãã¹ãã»ã©ã¼ã®ãè±ãã©ãã¯ããã¯ã¹ã³ã¼ã¹ãå®å ¨çããã¬ã¼ã³ãï¼ãå®åã§ä½¿ãããã¹ãã«ã身ã«çãã次ã®æ代ã«æ´»èºãã人æãç®æãã¾ãããã
StableLMã®ãã¡ã¤ã³ãã¥ã¼ãã³ã°ã£ã¦ã§ããã®ããªï¼ã¨èª¿ã¹ãã¨ãããGitHubã®Issueã§ãã¢ãã«èªä½ã®æ§è½ãã¾ã è¯ããªãããããã¡ã¤ã³ãã¥ã¼ãã³ã°ã®æ®µéã§ã¯ãªãããã¨ããã³ã¡ã³ããããã¾ãã¦ã ã·ã¼ãã®ä¸èº«ãè¦ã¦ã¿ãlm-evalã·ã¼ãstablelm-base-alpha-7bã¯54è¡ç®ã«ããã¾ãã¦ã確ãã«ä»ã®è¨èªã¢ãã«ã¨æ¯è¼ããã¨ã¾ã ã¾ã ãªæ§è½ã®ããã§ããå¿æ´ãããã§ããã ã·ã¼ãã®åã®æå³ããããã®åã®æå³ã«ã¤ãã¦ã¯æ¨å®ã§ãã以ä¸ã®éãã§ãã RAM è¨èªã¢ãã«ã®GPUã¡ã¢ãªæ¶è²»éã lambada(ppl) LAMBADAãã¼ã¿ã»ããã«ãã測å®å¤ããã³ã°ã¬ã³ã¸ã®è¨èªç解è½åããã¹ãããï¼æç« å ¨ä½ãèªã¾ãªãã¨çããããªãã¿ã¹ã¯ã§ã®è©ä¾¡ï¼ãPPLã¯Perplexityã¨ããææ¨ã§ãã¢ãã«ã®äºæ¸¬ã®ä¸ç¢ºããã示ããPPLãä½ãã»ã©ãã¢ãã«ã®äºæ¸¬ç²¾åº¦ãé«ãã lambada(acc
ãã£ã±GPTãä»çµã¿ããåå¼·ããããã¨ããæ¬ãããã¤ãè¦ã¤ããã®ã§ã¾ã¨ãã¦ããã¾ãã ã¾ãçè«çãªæ¦è¦ã æ©æ¢°å¦ç¿ãããã¥ã¼ã©ã«ãããã¯ã¼ã¯ãCNNã§ã®ç»åå¦çããã©ã³ã¹ãã©ã¼ãã¼ã§ã®èªç¶è¨èªå¦çãé³å£°èªèã»åæãããã¦ããããçµã¿åããããã«ãã¢ã¼ãã«ã¨ç« ãé²ãã®ã§ãçè«çãªæ¦è¦³ãå¾ãã®ã«ããã¨æãã¾ãã æåã¯æ°å¼ãå¤ãã®ã ãã©ãΣãforæã ã¨æãã¤ã¤ãå®ç¾©ã説ææä¸ã«åã¾ã£ã¦ãPerlãããã¡ãæªãè¨å·ä¸»ä½è¨èªã ã¨æãã°èªãããã©ããã©ãããã®ã§é£ã°ãã¦ãåé¡ãªãã¨æãã¾ãã 深層å¦ç¿ãããã«ãã¢ã¼ãã«æ å ±å¦ç㸠(AI/ãã¼ã¿ãµã¤ã¨ã³ã¹ã©ã¤ãã©ãªâåºç¤ããå¿ç¨ã¸â 3) ä½è :ä¸å±± è±æ¨¹,äºåç° ç¯¤å²,ç°æ æè£,äºä¸ ä¸é ,çä¹ ç¥¥åãµã¤ã¨ã³ã¹ç¤¾Amazon ã§ããããããªãä½ãã ãã¼ã¯ãã¤ã¶ã¼ããå ¨é¨ã¤ãã£ã¦ããã¾ããTensorFlowã§BERTãã¤ãã£ã¦GPT2ãã¤ãã
ãã©ã¡ã¼ã¿ãåºå®ããäºåå¦ç¿æ¸ã¿ã¢ãã«ã«å¯¾ãã¦ãããå°æ°ã®ãã©ã¡ã¼ã¿ãããªãä½ã©ã³ã¯è¡åãå°å ¥ã»å¦ç¿ãããã¨ã§ãã¢ãã«å ¨ä½ã®fine-tuningã¨åçã®æ§è½ãçºæ®ã§ããææ³ã§ããLoRAã¨ããã®è«æã«ã¤ãã¦è§£èª¬ããè³æã§ãã 深層å¦ç¿ãç¨ããèªç¶è¨èªå¦çã®æ´å²çãªå¤é·ã¨å¨è¾ºæè¡ãããLoRAãå¿ è¦ã¨â¦
ããã¡ã«ã¯ã ãã¼ã¿ã¢ããªãã£ã¯ã¹äºæ¥æ¬é¨ ã¤ã³ãã°ã¬ã¼ã·ã§ã³é¨ æ©æ¢°å¦ç¿ãã¼ã ã®ä¸æã§ãã LangChainã®Agentã®ã«ã¹ã¿ã ãã¼ã«ã¨ãã¦ãLlamaIndexã®ã¤ã³ããã¯ã¹ã使ç¨ããæ¹æ³ãç´¹ä»ãã¾ãã LangChainã¨ã¯ LlamdaIndexã¨åæ§ã大è¦æ¨¡è¨èªã¢ãã«(LLM)ãã«ã¹ã¿ãã¤ãºãã¦ä½¿ç¨ããããã®ã©ã¤ãã©ãªã§ãã ã«ã¹ã¿ã ããããã£ããããããã¨ã¼ã¸ã§ã³ããä½æãããã¨ãå¯è½ã§ã以ä¸ã®ãããªæ§ã ãªå¤é¨ãªã½ã¼ã¹ã»ãµã¼ãã¹ã¨é£æºå¯è½ã§ãã OpenAIã®ChatCompletionï¼è¦ããã«ChatGPTï¼ Googleæ¤ç´¢ãOpenSearchãªã©ã®æ¤ç´¢ãµã¼ãã¹ Hugging Faceã®ã¨ã³ã·ã¹ãã Pythonã®ã³ã¼ãã£ã³ã°ããããã¼ã« ä¾ï¼PyTorchã§hogehogeãªæ§æã®ãµã³ãã«ã³ã¼ããæãã¦ããªã© SQLãã¼ã¿ãã¼ã¹ãªã©ã®ãªã½ã¼ã¹ ä¾ï¼hogehog
ã¤ã³ããæè¿ãChatGPTãGPT-4ãªã©ã®å¤§è¦æ¨¡è¨èªã¢ãã«(LLM)ãæ¥éã«æ³¨ç®ãéãã¦ãã¾ããè¦ç´ã¿ã¹ã¯ã質çå¿çã¿ã¹ã¯ãªã©æ§ã ãªã¿ã¹ã¯ã§é«ãæ§è½ãçºæ®ãã¦ãã¾ãããããã®ã¢ãã«ã¯ãã¸ãã¹åéã§ã®å¿ç¨ãé常ã«æå¾ ããã¦ãããGoogleãMicrosoftãèªç¤¾ãµã¼ãã¹ã¨ã®é£æºãé²ãã¦ããã¨ããå ±éãç¸æ¬¡ãã§ãã¾ãã ä»åã¯ãæå ã§åä½ãã軽éãªå¤§è¦æ¨¡è¨èªã¢ãã«ãAlpaca-LoRAããæ¥æ¬èªã«ãã¡ã¤ã³ãã¥ã¼ãã³ã°ãã¦ã¿ã¾ããããã®è¨äºã§ã¯ããã®ãã¡ã¤ã³ãã¥ã¼ãã³ã°ã®ããã»ã¹ãå¿ç¨ä¾ã«ã¤ãã¦ç°¡åã«è§£èª¬ãã¦ããã¾ããChatGPTãGPT-4ã¯ãã¢ãã«ããã©ãã¯ããã¯ã¹ã§ããAPIçµç±ã§ããå ¥åãä¸ãã¦åºåãå¾ããã¨ãã§ãã¾ãããããã«ãç¾æç¹ï¼2023å¹´3æ20æ¥ï¼ã§ã¯ãç¬èªã®ãã¼ã¿ãç¨ãã¦ãã¡ã¤ã³ãã¥ã¼ãã³ã°ãè¡ããã¨ãã§ãã¾ããã ããããAlpaca-LoRAã¨ããã¢ãã«ãç¨
Guanaco - Generative Universal Assistant for Natural-language Adaptive Context-aware Omnilingual outputs Image by NobbiP (CC BY 3.0) Our new ðSOTA work: CausalLM 72B(preview), 14B & 7B, #1 SOTA model of its size, outperforming ALL open models. Recent update: Added support for multimodal VQA. Guanaco is an advanced instruction-following language model built on Meta's LLaMA 7B model. Expanding upon
LINEæ ªå¼ä¼ç¤¾ã¯ã2023å¹´10æ1æ¥ã«LINEã¤ãã¼æ ªå¼ä¼ç¤¾ã«ãªãã¾ãããLINEã¤ãã¼æ ªå¼ä¼ç¤¾ã®æ°ããããã°ã¯ãã¡ãã§ãã LINEã¤ãã¼ Tech Blog ããã«ã¡ã¯ãLINEã®å°ææ»æ²³(@kajyuuen)ã§ããNLPãã¼ã ã§åºæ表ç¾æ½åºãæ害表ç¾ã®æ¤ç¥ãLINEã¹ã¿ã³ãæ¨è¦ã®æ¹åãªã©èªç¶è¨èªå¦çã«é¢ããä»äºããã¦ãã¾ãã ãã®è¨äºã§ã¯LINEãå ¬éããè¨èªã¢ãã«ãLINE DistilBERTãã«ã¤ãã¦ç´¹ä»ãã¾ãã https://huggingface.co/line-corporation/line-distilbert-base-japanese https://github.com/line/LINE-DistilBERT-Japanese LINE DistilBERTã¯æ¬¡ã®ãããªç¹å¾´ãæã¤æ¥æ¬èªè¨èªã¢ãã«ã§ãã é«æ§è½ã»é«éã»è»½é Hugging Faceã®tra
æè¿ ChatGPT API ã«ã³ã¼ããæ¸ããã¦ããã eval ããã¨ããä»çµã¿ãä½ãã¾ããã ãã®éã«èããã¨ã©ã¼ãã³ããªã³ã°ã®ãã©ã¯ãã£ã¹ããChatGPT API ãªãã§ã¯ã§å人çã«é¢ç½ãã£ãã®ã§è¨äºã«æ®ãã¦ããã¾ãã ð tl;dr ChatGPT ã¯ã³ã¼ããæ¸ããã¨ãæå¦ããå ´åããããèæ ®ãããªãã©ã¤ã®ä»çµã¿ãä½ã ChatGPT ã®æ¸ããã³ã¼ãã§çºçããã¨ã©ã¼ã¯ ChatGPT ã«è§£æ±ºãã¦ããã ð¬ ä½ããããã¨ããï¼ ä»¥åã®è¨äºã«ã¦ Minecraft ä¸ã§ ChatGPT ã«ä½ããããã®ãä¼ããã¨éæ³ã¿ããã«å®ç¾ãã¦ãããã³ãã³ãã¨ãããã®ãç´¹ä»ãã¾ããã 豪è¯(?)ãªå®¶ãä½ã£ã¦ãããä¾ ãã®ã³ãã³ãã®å®è£ ã¯ã¨ã¦ãã·ã³ãã«ã§ãã¦ã¼ã¶ã¼ã®æ示ï¼promptï¼ãã ChatGPT API ã«ã³ã¼ããæ¸ããã¦ãããã eval ãã¦ããã ãã§ãã def think_co
2023/03/09 è¿½è¨ ãã®ã³ãã³ãã§ä½¿ã£ã¦ããã¨ã©ã¼ãã³ããªã³ã°ã®ä»çµã¿ã«ã¤ãã¦å¥éè¨äºã«ãã¾ããã ChatGPTãªãã§ã¯ã§é¢ç½ãã®ã§ãèªãã§ããããã¨å¬ããã§ãï¼ ð ChatGPT API ã«ã³ã¼ããæ¸ãã㦠eval ããéã®ã¨ã©ã¼ãã³ããªã³ã°ã»ãã©ã¯ãã£ã¹ æ¯åã¨ä¸ç·ã«éã³ãããªã¨ãMinecraft 㨠ChatGPT ãç¹ãã¦ããããããã¨ãæ¥æ¬èªã§ä¼ããã¨ããæãã«å®ç¾ãã¦ãããã³ãã³ããä½ãã¾ãã ðª ä½ã£ããã® /py magic ã«ç¶ãã¦ãããããã¨ãä¼ããã¨ãããæãã«å®ç¾ãã¦ããã¾ãã 家ãã¤ããä¾ /py magic 10ãã¹å ã«è±ªè¯ãªå®¶ãä½ã£ã¦ãçªã¯åºãã§ã æ°´æµã¨ã¬ãã¼ã¿ã¼ãã¤ããä¾ /py magic æ°´æµã¨ã¬ãã¼ã¿ã¼ãä½ã£ã¦ãå¨ãã¯ç³ãããã¯ã§ã ð ï¸ å®è£ äºåã« Minecraft 㧠Python ãå®è¡ã§ããç°å¢ãæ´ãã¾ãã mcpi
ãªãªã¼ã¹ãé害æ å ±ãªã©ã®ãµã¼ãã¹ã®ãç¥ãã
ææ°ã®äººæ°ã¨ã³ããªã¼ã®é ä¿¡
å¦çãå®è¡ä¸ã§ã
j次ã®ããã¯ãã¼ã¯
kåã®ããã¯ãã¼ã¯
lãã¨ã§èªã
eã³ã¡ã³ãä¸è¦§ãéã
oãã¼ã¸ãéã
{{#tags}}- {{label}}
{{/tags}}