Hugging Face's logo
Activity Feed

AI & ML interests

A one-year long research workshop on large language models: the Summer of Language Models 21 ๐ŸŒธ

Recent Activity
christopherย  new activity about 9 hours ago
bigscience/bloom:Test PR
christopherย  new activity about 9 hours ago
bigscience/bloom:Test discussion
christopherย  new activity about 9 hours ago
bigscience/bloom:Test discussion
View all activity
christopherย 
in bigscience/bloom about 9 hours ago

Test PR
#286 opened about 10 hours ago by
FIRSTACCOUNT69

Test discussion
#287 opened about 9 hours ago by
FIRSTACCOUNT69

Test discussion
#288 opened about 9 hours ago by
FIRSTACCOUNT69
albertvillanovaย 
posted an update 3 days ago
view post
Post
1576
๐Ÿš€ TRL v0.29.0 introduces trl-training: an agent-native training skill.

This makes the TRL CLI a structured, agent-readable capability, allowing AI agents to reliably execute training workflows such as:
- Supervised Fine-Tuning (SFT)
- Direct Preference Optimization (DPO)
- Group Relative Policy Optimization (GRPO)

Weโ€™re excited to see what the community builds on top of this.

If youโ€™re working on AI agents, alignment research, or scalable RL training infrastructure: give TRL v0.29.0 a try! ๐Ÿค—

The future of ML tooling is agent-native.
๐Ÿ”— https://github.com/huggingface/trl/releases/tag/v0.29.0
armancย 
authored a paper 9 days ago
lewtunย 
submitted a paper to Daily Papers 15 days ago
lewtunย 
submitted a paper to Daily Papers 17 days ago
albertvillanovaย 
posted an update 18 days ago
view post
Post
1682
5 years already working in democratizing AI ๐Ÿค—
Grateful to be part of such an awesome team making it happen every day.
julien-cย 
submitted a paper to Daily Papers 30 days ago
yjerniteย 
authored a paper about 1 month ago
stasย 
authored a paper about 1 month ago
armancย 
authored a paper about 1 month ago
ybelkadaย 
authored a paper about 2 months ago
christopherย 
in bigscience/bloomz-560m 3 months ago

Fails to load with transformers v4.57+

4
#14 opened 3 months ago by
qgallouedec
monsoon-nlpย 
posted an update 3 months ago
christopherย 
in bigscience/petals-api 4 months ago

Bloom
#2 opened 4 months ago by
Raz-Test
nouamanetaziย 
posted an update 4 months ago
view post
Post
4492
After training ๐’๐ฆ๐จ๐ฅ๐‹๐Œ๐Ÿ‘ on ๐Ÿ‘๐Ÿ–๐Ÿ’ ๐‡๐Ÿ๐ŸŽ๐ŸŽ๐ฌ for nearly a month, I've come to realize something most people overlook: ๐ข๐ง๐Ÿ๐ซ๐š๐ฌ๐ญ๐ซ๐ฎ๐œ๐ญ๐ฎ๐ซ๐ž ๐ข๐ฌ ๐ญ๐ก๐ž ๐ฆ๐š๐ค๐ž-๐จ๐ซ-๐›๐ซ๐ž๐š๐ค ๐Ÿ๐š๐œ๐ญ๐จ๐ซ ๐ข๐ง ๐‹๐‹๐Œ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐ . ๐Ÿ”ฅ

Everyone talks about model architecture and data quality. And yes, those matter immensely. But here's what nobody tells you: when your training run fails at 2 AM because of mysterious ๐๐‚๐‚๐‹ ๐ž๐ซ๐ซ๐จ๐ซ๐ฌ, or when your expensive GPU cluster is running at ๐Ÿ”๐ŸŽ% ๐ž๐Ÿ๐Ÿ๐ข๐œ๐ข๐ž๐ง๐œ๐ฒ, the problem isn't your model. It's most probably a ๐ฆ๐ข๐ฌ๐ฎ๐ฌ๐ž ๐จ๐Ÿ ๐ญ๐ก๐ž ๐ก๐š๐ซ๐๐ฐ๐š๐ซ๐ž. ๐Ÿ› ๏ธ

Questions that seemed simple but had no clear answers: Why is ๐Œ๐จ๐„ ๐ญ๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐ฌ๐ฅ๐จ๐ฐ๐ž๐ซ ๐ญ๐ก๐š๐ง ๐๐ž๐ง๐ฌ๐ž ๐ฆ๐จ๐๐ž๐ฅ๐ฌ? Which ๐๐‚๐‚๐‹ ๐Ÿ๐ฅ๐š๐ ๐ฌ should we actually set? How often should we checkpoint without killing throughput?

That's why we built ๐“๐ก๐ž ๐’๐ฆ๐จ๐ฅ ๐“๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐๐ฅ๐š๐ฒ๐›๐จ๐จ๐ค ๐Ÿ“–: a complete guide covering everything from model architecture and data curation to the SmolLM3 training marathon, post-training techniques, and crucially, the ๐ข๐ง๐Ÿ๐ซ๐š๐ฌ๐ญ๐ซ๐ฎ๐œ๐ญ๐ฎ๐ซ๐ž ๐ฅ๐š๐ฒ๐ž๐ซ that most teams get wrong.

We validated real vs theoretical bandwidth across the entire stack: ๐‡๐๐Œ๐Ÿ‘ ๐ก๐ข๐ญ๐ญ๐ข๐ง๐  ๐Ÿ‘ ๐“๐/๐ฌ, ๐๐•๐‹๐ข๐ง๐ค ๐Ÿ’.๐ŸŽ ๐ซ๐ž๐š๐œ๐ก๐ข๐ง๐  ๐Ÿ•๐Ÿ–๐Ÿ” ๐†๐/๐ฌ, ๐๐‚๐ˆ๐ž ๐†๐ž๐ง๐Ÿ’ ๐š๐ญ ๐Ÿ๐Ÿ’.๐Ÿ ๐†๐/๐ฌ. Then we ran collective operations across ๐Ÿ๐Ÿ๐Ÿ– ๐†๐๐”๐ฌ (16 nodes, 8xH100s each) and measured how performance degrades at scale: all-reduce drops from ๐Ÿ’๐Ÿ–๐ŸŽ ๐†๐/๐ฌ on a single node to ๐Ÿ‘๐Ÿ๐ŸŽ-๐Ÿ‘๐Ÿ“๐ŸŽ ๐†๐/๐ฌ across 16 nodes.

If you've ever wondered why your training runs are slower than they should be, or you're planning to scale up and want to avoid expensive mistakes, this guide might save you weeks of debugging.

๐“๐ก๐ž ๐’๐ฆ๐จ๐ฅ ๐“๐ซ๐š๐ข๐ง๐ข๐ง๐  ๐๐ฅ๐š๐ฒ๐›๐จ๐จ๐ค: https://lnkd.in/e5MKXUHS

Shared with โค๏ธ by the HuggingFace team
Zaidย 
authored a paper 4 months ago
thomwolfย 
authored a paper 5 months ago
vumichienย 
authored a paper 5 months ago