14 16 6

Rémi Ouazan Reboul

ror

AI & ML interests

None yet

Recent Activity

updated a dataset about 14 hours ago

huggingface/documentation-images

upvoted an article 1 day ago

Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law

new activity 14 days ago

huggingface/documentation-images:Upload images for the continuous async blog post

View all activity

Organizations

updated a dataset about 14 hours ago

huggingface/documentation-images

Viewer • Updated about 11 hours ago • 59 • 2.32M • 140

upvoted an article 1 day ago

Article

Two Years of Local AI on a Laptop: When Open Models Outpaced Moore's Law

mishig

•

1 day ago

• 15

New activity in huggingface/documentation-images 14 days ago

Upload images for the continuous async blog post

#611 opened 14 days ago by

ror

reacted to qgallouedec's post with 🔥 26 days ago

Post

1958

TRL v1.2 introduces the SSDTrainer 🚀

Simple Self-Distillation (SSD) from Apple's paper "Embarrassingly Simple Self-Distillation Improves Code Generation" is now available as an experimental trainer in TRL.

The recipe is as minimal as the name suggests: sample completions from the model itself at a training-time temperature, then fine-tune on those raw, unverified samples with plain cross-entropy. No reward model. No verifier. No teacher model. No reinforcement learning. Just prompts and the model.

from trl.experimental.ssd import SSDConfig, SSDTrainer

trainer = SSDTrainer(
    model="Qwen/Qwen3-4B-Instruct",
    args=SSDConfig(temperature=0.6, top_k=20, top_p=0.95),
    train_dataset=dataset,
)
trainer.train()

v1.2 also ships expanded tool-calling support (LLaMA 3.1 / 3.2, DeepSeek-V3), another round of KTO ↔ DPO alignment getting us closer to promoting KTO to stable, a big GRPO simplification for overlong tool results, deprecation of use_transformers_paged, and key fixes for VLM response parsing.

Full release notes: https://github.com/huggingface/trl/releases/tag/v1.2.0

liked a Space 3 months ago

Can LLMs Play the Game of Science?

📝

Explore LLM science benchmark scores

upvoted an article 4 months ago

Article

Architectural Choices in China's Open-Source AI Ecosystem: Building Beyond DeepSeek

huggingface

•

Jan 27

• 45

commented on Creating custom kernels for the AMD MI300 4 months ago

We use an int16 variable in the shared memory (a semaphore), which can be looked up by any warp on the SM. The thing you have to be carefull about is to have all instructions related to SMEM or GMEM finished before changing the state of the semaphore. For instance, you have to explicitly wait for SMEM stores to be over before switching the semaphore.