Dokyoon PRO

leeloolee

AI & ML interests

ai

Recent Activity

upvoted a paper about 11 hours ago
GUI Agents: A Survey
reacted to m-ric's post with ๐Ÿ‘ 6 days ago
๐‡๐ฎ๐ ๐ ๐ข๐ง๐  ๐…๐š๐œ๐ž ๐ซ๐ž๐ฅ๐ž๐š๐ฌ๐ž๐ฌ ๐๐ข๐œ๐จ๐ญ๐ซ๐จ๐ง, ๐š ๐ฆ๐ข๐œ๐ซ๐จ๐ฌ๐œ๐จ๐ฉ๐ข๐œ ๐ฅ๐ข๐› ๐ญ๐ก๐š๐ญ ๐ฌ๐จ๐ฅ๐ฏ๐ž๐ฌ ๐‹๐‹๐Œ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐Ÿ’๐ƒ ๐ฉ๐š๐ซ๐š๐ฅ๐ฅ๐ž๐ฅ๐ข๐ณ๐š๐ญ๐ข๐จ๐ง ๐Ÿฅณ ๐Ÿ•ฐ๏ธ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years. ๐Ÿ‘ด๐Ÿป If they had needed all this time, we would have GPU stories from the time of Pharaoh ๐“‚€: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons " ๐Ÿ› ๏ธ But instead, they just parallelized the training on 24k H100s, which made it take just a few months. This required parallelizing across 4 dimensions: data, tensor, context, pipeline. And it is infamously hard to do, making for bloated code repos that hold together only by magic. ๐Ÿค ๐—•๐˜‚๐˜ ๐—ป๐—ผ๐˜„ ๐˜„๐—ฒ ๐—ฑ๐—ผ๐—ป'๐˜ ๐—ป๐—ฒ๐—ฒ๐—ฑ ๐—ต๐˜‚๐—ด๐—ฒ ๐—ฟ๐—ฒ๐—ฝ๐—ผ๐˜€ ๐—ฎ๐—ป๐˜†๐—บ๐—ผ๐—ฟ๐—ฒ! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry. And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening! โšก ๐—œ๐˜'๐˜€ ๐˜๐—ถ๐—ป๐˜†, ๐˜†๐—ฒ๐˜ ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ณ๐˜‚๐—น: Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this) Go take a look ๐Ÿ‘‰ https://github.com/huggingface/picotron/tree/main/picotron
View all activity

Organizations

sionic-ai's profile picture Multi๐Ÿค–Transformers's profile picture ์ธ์ŠคํŠธ๋ŸญํŠธ.ํ•œ๊ตญ's profile picture AI Safeguard's profile picture

leeloolee's activity

reacted to m-ric's post with ๐Ÿ‘ 6 days ago
view post
Post
2026
๐‡๐ฎ๐ ๐ ๐ข๐ง๐  ๐…๐š๐œ๐ž ๐ซ๐ž๐ฅ๐ž๐š๐ฌ๐ž๐ฌ ๐๐ข๐œ๐จ๐ญ๐ซ๐จ๐ง, ๐š ๐ฆ๐ข๐œ๐ซ๐จ๐ฌ๐œ๐จ๐ฉ๐ข๐œ ๐ฅ๐ข๐› ๐ญ๐ก๐š๐ญ ๐ฌ๐จ๐ฅ๐ฏ๐ž๐ฌ ๐‹๐‹๐Œ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐Ÿ’๐ƒ ๐ฉ๐š๐ซ๐š๐ฅ๐ฅ๐ž๐ฅ๐ข๐ณ๐š๐ญ๐ข๐จ๐ง ๐Ÿฅณ

๐Ÿ•ฐ๏ธ Llama-3.1-405B took 39 million GPU-hours to train, i.e. about 4.5 thousand years.

๐Ÿ‘ด๐Ÿป If they had needed all this time, we would have GPU stories from the time of Pharaoh ๐“‚€: "Alas, Lord of Two Lands, the shipment of counting-stones arriving from Cathay was lost to pirates, this shall delay the building of your computing temple by many moons "

๐Ÿ› ๏ธ But instead, they just parallelized the training on 24k H100s, which made it take just a few months.
This required parallelizing across 4 dimensions: data, tensor, context, pipeline.
And it is infamously hard to do, making for bloated code repos that hold together only by magic.

๐Ÿค ๐—•๐˜‚๐˜ ๐—ป๐—ผ๐˜„ ๐˜„๐—ฒ ๐—ฑ๐—ผ๐—ป'๐˜ ๐—ป๐—ฒ๐—ฒ๐—ฑ ๐—ต๐˜‚๐—ด๐—ฒ ๐—ฟ๐—ฒ๐—ฝ๐—ผ๐˜€ ๐—ฎ๐—ป๐˜†๐—บ๐—ผ๐—ฟ๐—ฒ! Instead of building mega-training codes, Hugging Face colleagues cooked in the other direction, towards tiny 4D parallelism libs. A team has built Nanotron, already widely used in industry.
And now a team releases Picotron, a radical approach to code 4D Parallelism in just a few hundred lines of code, a real engineering prowess, making it much easier to understand what's actually happening!

โšก ๐—œ๐˜'๐˜€ ๐˜๐—ถ๐—ป๐˜†, ๐˜†๐—ฒ๐˜ ๐—ฝ๐—ผ๐˜„๐—ฒ๐—ฟ๐—ณ๐˜‚๐—น:
Counting in MFU (Model FLOPs Utilization, how much the model actually uses all the compute potential), this lib reaches ~50% on SmolLM-1.7B model with 8 H100 GPUs, which is really close to what huge libs would reach. (Caution: the team is leading further benchmarks to verify this)

Go take a look ๐Ÿ‘‰ https://github.com/huggingface/picotron/tree/main/picotron
  • 1 reply
ยท
reacted to alimotahharynia's post with ๐Ÿ”ฅ 7 days ago
view post
Post
1543
Here's the space for our new article that leverages LLMs with reinforcement learning to design high-quality small molecules. Check it out at alimotahharynia/GPT-2-Drug-Generator. You can also access the article here: https://arxiv.org/abs/2411.14157.
I would be happy to receive your feedback.
reacted to cutechicken's post with โค๏ธ 7 days ago
view post
Post
2822
๐Ÿš€ RAGOndevice: High-Performance Local AI Document Analysis Assistant
๐Ÿ’ซ Core Value
RAGOndevice is a high-performance AI system running locally without cloud dependency. Using CohereForAI's optimized 7B model, it enables professional-grade document analysis on standard PCs. โœจ
๐ŸŒŸ Ondevice AI Advantages
1. ๐Ÿ”‹ Efficient Resource Utilization

๐ŸŽฏ Optimized 7B Model: Runs on standard PCs
โšก Local Processing: Instant response without cloud
๐Ÿ’ป Low-Spec Compatible: Performs well on regular GPUs
๐Ÿ”„ Optimized Memory: Ensures stable operation

2. ๐Ÿ›ก๏ธ Data Security & Cost Efficiency

๐Ÿ”’ Complete Privacy: No external data transmission
๐ŸŒ Offline Operation: No internet required
๐Ÿ’ฐ No Subscription: One-time installation
โš™๏ธ Resource Optimization: Uses existing hardware

๐ŸŽฎ Key Features
1. ๐Ÿ“Š Powerful Document Analysis

๐Ÿ“ Multi-Format Support: TXT, CSV, PDF, Parquet
๐Ÿง  Intelligent Analysis: Automatic structure recognition
๐Ÿ‘๏ธ OCR Support: Advanced PDF text extraction
๐Ÿ’ฌ Real-time Chat: Natural language interaction

2. ๐Ÿ” Local RAG System

๐ŸŽฏ Efficient Search: TF-IDF based local search
๐Ÿงฉ Context Understanding: Accurate information retrieval
๐Ÿ“š Wikipedia Integration: Rich background knowledge

๐ŸŽฏ Use Cases

๐Ÿข Enterprise: Secure confidential document processing
๐Ÿ”ฌ Personal Research: Private data analysis
๐Ÿ“š Education: Personal learning material analysis
๐Ÿ’ป Development: Local codebase analysis

โญ Differentiators

๐Ÿƒโ€โ™‚๏ธ Independent Operation: Zero cloud dependency
โšก Instant Response: No network latency
๐Ÿ” Complete Security: Full data control
๐Ÿ’Ž Cost Efficiency: No ongoing costs

๐Ÿ”ฎ Future Plans

๐Ÿš€ Enhanced model optimization
๐Ÿ“š Local knowledge base expansion
โšก Hardware optimization
๐Ÿ“ Extended file support


๐ŸŒŸ RAGOndevice democratizes high-performance AI, providing the optimal local AI solution for security-sensitive environments. ๐Ÿš€

๐Ÿ”ฅ Power of Local AI: Experience enterprise-grade AI capabilities right on your device!

VIDraft/RAGOndevice
reacted to julien-c's post with ๐Ÿ”ฅ 14 days ago
view post
Post
7605
After some heated discussion ๐Ÿ”ฅ, we clarify our intent re. storage limits on the Hub

TL;DR:
- public storage is free, and (unless blatant abuse) unlimited. We do ask that you consider upgrading to PRO and/or Enterprise Hub if possible
- private storage is paid above a significant free tier (1TB if you have a paid account, 100GB otherwise)

docs: https://huggingface.co/docs/hub/storage-limits

We optimize our infrastructure continuously to scale our storage for the coming years of growth in Machine learning, to the benefit of the community ๐Ÿ”ฅ

cc: @reach-vb @pierric @victor and the HF team
ยท
upvoted an article 27 days ago
view article
Article

LLaVA-o1: Let Vision Language Models Reason Step-by-Step

By mikelabs โ€ข
โ€ข 11