Quentin Gallouédec

@qgallouedec

PhD - Post-training @huggingface 🤗 TRL lead maintainer 🇫🇷 in 🇨🇦

following
followers

Mistral announces improved Le Chonk weights; main checkpoint listed as upcoming

Mistral announced the weights after a public preview, but its linked main Hugging Face listing showed “Upcoming release.”

117 Sources, 6d ago, updated 6d ago

Score centering added to TRL to tackle instability in LLM reinforcement learning

The announcement calls the addition stable; a September post says it targets instability when training and sampling policies differ.

6 Sources, 14d ago, updated 14d ago

Technology

Gallouédec posted “Next stop: 1M/week,” and Hugging Face reposted it

Quentin Gallouédec wrote that 1 million trainings a week is the next goal and framed post-training as a step toward a less centralized AI ecosystem. Hugging Face later reposted the post.

2 Sources, 40d ago, first seen 40d ago

Telemetry reportedly shows 1 million trainings a month with TRL

The poster says their telemetry showed the monthly figure on October 1 and sets 1 million a week as the next goal.

2 Sources, 40d ago, first seen 40d ago

Featured in

Mistral announces improved Le Chonk weights; main checkpoint listed as upcoming

Score centering added to TRL to tackle instability in LLM reinforcement learning

Gallouédec posted “Next stop: 1M/week,” and Hugging Face reposted it

Telemetry reportedly shows 1 million trainings a month with TRL

Featured in

Mistral announces improved Le Chonk weights; main checkpoint listed as upcoming

Score centering added to TRL to tackle instability in LLM reinforcement learning

Gallouédec posted “Next stop: 1M/week,” and Hugging Face reposted it

Telemetry reportedly shows 1 million trainings a month with TRL