[B! research] rawwellã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

rawwell id:rawwell

researchã«é–¢ã™ã‚‹rawwellã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (160)

${{author_name}}$

{{author_name}} {{created}}

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

{{#following_bookmarks}}

${{author_name}}$

{{author_name}} {{created}}

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

{{/following_bookmarks}}

{{/is_wiped}}

Minstrel: tool for structural prompt generation with Multi-Agents Coordination based on LangGPTâ€¦
rawwell 2024/10/09
research
ãƒªãƒ³ã‚¯
RAGã«é–¢ã™ã‚‹ä¸»è¦ãªè«–æ–‡ã‚’æ™‚ç³»åˆ—é †ã«ã¾ã¨ã‚ã¦ã„ã(2024å¹´åº¦ç‰ˆ)ï½œR
RAGã«é–¢ã™ã‚‹ä¸»è¦ãªè«–æ–‡ã¾ã¨ã‚ã¦ã„ãã¾ã™ã€‚(éŽåŽ»ã®åˆ†å«ã‚ã¦éšæ™‚æ›´æ–°äºˆå®š) è¦‹ã¤ã‘ãŸã‚‚ã®ã‹ã‚‰ã¾ã¨ã‚ã¦ã„ã‚‹ã®ã§ã€æœ€æ–°ã®2024å¹´ä»¥é™ã®è«–æ–‡å¤šã‚ã§ã™ã€‚ Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks(22/05/2022) ä¸€è¨€ç´¹ä»‹â“ LLMã£ã¦ã€äº‹å‰å¦ç¿’ã•ã‚ŒãŸçŸ¥è˜ã«é–¢ã—ã¦ã¯ç”ãˆã¦ãã‚Œã‚‹ã‘ã©ã€æœ€æ–°ã®ãƒ‹ãƒ¥ãƒ¼ã‚¹ã ã£ãŸã‚Šã€å°‚é–€çš„ãªæƒ…å ±ã‚„çµ„ç¹”å›ºæœ‰ã®æƒ…å ±ã«ã¯å¯¾å¿œã§ããªã„ã‚ˆãªã ðŸ’¡ å¤–éƒ¨çŸ¥è˜ã‚’LLMã«æ¤œç´¢ã•ã›ã‚ˆã†!â†’RAGã®èª•ç”Ÿ Abstractæ—¥æœ¬èªžè¨³å¤§è¦æ¨¡ãªäº‹å‰å¦ç¿’æ¸ˆã¿è¨€èªžãƒ¢ãƒ‡ãƒ«ã¯ã€ãã®ãƒ‘ãƒ©ãƒ¡ãƒ¼ã‚¿ã«äº‹å®ŸçŸ¥è˜ã‚’è“„ç©ã—ã€ä¸‹æµã®è‡ªç„¶è¨€èªžå‡¦ç†ï¼ˆNLPï¼‰ã‚¿ã‚¹ã‚¯ã«å¾®èª¿æ•´ã•ã‚ŒãŸã¨ãã«æœ€å…ˆç«¯ã®æˆæžœã‚’é”æˆã™ã‚‹ã“ã¨ãŒç¤ºã•ã‚Œã¦ã„ã¾ã™ã€‚ã—ã‹ã—ã€çŸ¥è˜ã‚’ã‚¢ã‚¯ã‚»ã‚¹ã—ã¦æ£ç¢ºã«æ“ä½œã™ã‚‹èƒ½åŠ›ã¯ä¾ç„¶ã¨ã—ã¦é™ã‚‰ã‚Œã¦ãŠã‚Šã€çŸ¥è˜é›†ç´„åž‹ã‚¿ã‚¹ã‚¯ã§ã¯ã€ã‚¿ã‚¹ã‚¯å›ºæœ‰ã®ã‚¢ãƒ¼ã‚
rawwell 2024/09/29
RAG

ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Entity Linking and Relationship Extraction With Relik in LlamaIndex
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Introducing Qwen2-Math
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Build a Knowledge Graph-based Agent With Llama 3.1, NVIDIA NIM, and LangChain
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Mastering RAG: How To Evaluate LLMs For RAG - Galileo AI
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
METAs Segment Anything Model (SAM) Complete Breakdown
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Building a Multi-Modal RAG System for Visual Question Answering
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Speeding Up the Vision Transformer with BatchNorm
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Causal Inference with Python: A Guide to Propensity Score Matching
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
ChatGPT vs. Claude vs. Gemini for Data Analysis (Part 1)
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
Paper Walkthrough: Vision Transformer (ViT)
rawwell 2024/08/16
ã‚ã¨ã§èªã‚€

research
ãƒªãƒ³ã‚¯
What if we drop the causal mask in auto-regressive Transformer?
rawwell 2024/08/10
â€œThis property ensures that the model can only attend to previous positions in the sequence, not future positions, in order to generate predictions sequentially.â€

LLM

research
ãƒªãƒ³ã‚¯
Pretraining LLMs - DeepLearning.AI
rawwell 2024/07/18
LLM

research
ãƒªãƒ³ã‚¯
Introducing DoRA, a High-Performing Alternative to LoRA for Fine-Tuning | NVIDIA Technical Blog
rawwell 2024/07/11
â€œDoRA demonstrates the ability to make only substantial directional adjustments with relatively minimal changes in magnitude or the reverse, while showing learning patterns closer to FT. This signifies its superior learning capacity over LoRAâ€

research
ãƒªãƒ³ã‚¯
LLMã®ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚° ã§ ä½•ãŒã§ãã¦ ä½•ãŒã§ããªã„ã®ã‹ï½œnpaka
LLMã®ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã§ä½•ãŒã§ãã¦ã€ä½•ãŒã§ããªã„ã®ã‹ã¾ã¨ã‚ã¾ã—ãŸã€‚ 1. LLMã®ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°LLMã®ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã®ç›®çš„ã¯ã€ã€Œç‰¹å®šã®ã‚¢ãƒ—ãƒªã‚±ãƒ¼ã‚·ãƒ§ãƒ³ã®ãƒ‹ãƒ¼ã‚ºã¨ãƒ‡ãƒ¼ã‚¿ã«åŸºã¥ã„ã¦ã€ãƒ¢ãƒ‡ãƒ«ã®å‡ºåŠ›ã®å“è³ªã‚’å‘ä¸Šã•ã›ã‚‹ã“ã¨ã€ã«ã‚ã‚Šã¾ã™ã€‚ OpenAIã®ãƒ‰ã‚ãƒ¥ãƒ¡ãƒ³ãƒˆã«ã¯ã€æ¬¡ã®ã‚ˆã†ã«è¨˜è¿°ã•ã‚Œã¦ã„ã¾ã™ã€‚ ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã¯ã€ãƒ—ãƒãƒ³ãƒ—ãƒˆã«åŽã¾ã‚‹ã‚ˆã‚Šã‚‚å¤šãã®ä¾‹ã§å¦ç¿’ã™ã‚‹ã“ã¨ã§ã€Few-Shotå¦ç¿’ã‚’æ”¹å–„ã—ã¾ã™ã€‚ä¸€åº¦ãƒ¢ãƒ‡ãƒ«ã‚’ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã™ã‚Œã°ã€ãƒ—ãƒãƒ³ãƒ—ãƒˆã«ãã‚Œã»ã©å¤šãã®ä¾‹ã‚’æä¾›ã™ã‚‹å¿…è¦ãŒãªããªã‚Šã¾ã™ã€‚ã“ã‚Œã«ã‚ˆã‚Šã€ã‚³ã‚¹ãƒˆã‚’å‰Šæ¸›ã—ã€ä½Žãƒ¬ã‚¤ãƒ†ãƒ³ã‚·ã®ãƒªã‚¯ã‚¨ã‚¹ãƒˆã‚’å¯èƒ½ã«ã—ã¾ã™ã€‚ ã—ã‹ã—å®Ÿéš›ã«ã¯ã€ãã‚Œã‚ˆã‚Šã‚‚ã‹ãªã‚Šè¤‡é›‘ã§ã™ã€‚ LLMã«ã¯ã€Œå¤§é‡ã®ãƒ‡ãƒ¼ã‚¿ã‚’æŠ•ã’ã‚Œã°è‡ªå‹•çš„ã«è§£æ±ºã™ã‚‹ã€ã‚ˆã†ãªå‰µç™ºçš„ãªç‰¹æ€§ãŒã‚ã‚‹ãŸã‚ã€ãƒ•ã‚¡ã‚¤ãƒ³ãƒãƒ¥ãƒ¼ãƒ‹ãƒ³ã‚°ã‚‚ãã®ã‚ˆã†ã«æ©Ÿèƒ½ã™ã‚‹ã¨äººã€…ã¯è€ƒãˆã¦ã„ã¾ã™ãŒã€å¿…ãšã—ã‚‚ãã†ã§ã¯ã‚ã‚Šã¾ã›
rawwell 2024/07/04
research
ãƒªãƒ³ã‚¯
Microsoft: 'Skeleton Key' attack unlocks the worst of AI
Microsoft on Thursday published details about Skeleton Key â€“ a technique that bypasses the guardrails used by makers of AI models to prevent their generative chatbots from creating harmful content. As of May, Skeleton Key could be used to coax an AI model - like Meta Llama3-70b-instruct, Google Gemini Pro, or Anthropic Claude 3 Opus - into explaining how to make a Molotov cocktail. The combination
rawwell 2024/07/03
The only exception was GPT-4, which resisted the attack as direct text prompt, but was still affected if the behavior modification request was part of a user-defined system message â€“ something developers working with OpenAI's API can specify.

research
ãƒªãƒ³ã‚¯
Solving the 'Lost-in-the-Middle' Problem in Large Language Models: A Breakthrough in Attention Calibration
rawwell 2024/07/03
â€œThe ability of LLMs to accurately find and use information within that context weakens as the relevant information gets further away from the beginning or end. In other words, they tend to focus on the information at the beginning and end, neglecting whatâ€™s sandwiched in between.â€

research
ãƒªãƒ³ã‚¯
Why is observability so expensive?
Itâ€™s no secret that observability costs are top of mind for many organizations in the post-zero interest rate phenomenon (ZIRP) era (see here, here, and here for example discussions, though similar sentiments can be found far and wide). Organizations are frustrated with the percentage of infrastructure spend (sometimes > 25%!) allocated towards logging, metrics, and traces, and are struggling to u
rawwell 2024/06/29
research

observability
ãƒªãƒ³ã‚¯
å¤§è¦æ¨¡è¨€èªžãƒ¢ãƒ‡ãƒ« (LLM) ã®æŠ€è¡“ã¨æœ€æ–°å‹•å‘
ãƒžãƒ«ãƒãƒ¡ãƒ‡ã‚£ã‚¢ï¼Œåˆ†æ•£ï¼Œå”èª¿ã¨ãƒ¢ãƒã‚¤ãƒ«ï¼ˆDICOMO2024ï¼‰ã‚·ãƒ³ãƒã‚¸ã‚¦ãƒ ã§ã®æ‹›å¾…è¬›æ¼”ã®è³‡æ–™ã§ã™ã€‚ https://dicomo.org/
rawwell 2024/06/29
research
ãƒªãƒ³ã‚¯
1 2 3 4 5 6 7 8 æ¬¡ã®ãƒšãƒ¼ã‚¸

ãŠçŸ¥ã‚‰ã›

ã‚‚ã£ã¨èªã‚€

å…¬å¼Twitter

@HatenaBookmark
ãƒªãƒªãƒ¼ã‚¹ã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“ã‚¹ã®ãŠçŸ¥ã‚‰ã›
@hatebu
æœ€æ–°ã®äººæ°—ã‚¨ãƒ³ãƒˆãƒªãƒ¼ã®é…ä¿¡

ã‚ãƒ¼ãƒœãƒ¼ãƒ‰ã‚·ãƒ§ãƒ¼ãƒˆã‚«ãƒƒãƒˆä¸€è¦§

jæ¬¡ã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

kå‰ã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

lã‚ã¨ã§èªã‚€

eã‚³ãƒ¡ãƒ³ãƒˆä¸€è¦§ã‚’é–‹ã

oãƒšãƒ¼ã‚¸ã‚’é–‹ã

è¨å®šã‚’å¤‰æ›´ã—ã¾ã—ãŸx