Mistral Large 4: 1T-parameter 'Le Chonk' now trainable in TRL with open SFT example
QGallouedec · x · 2026-10-30
QGallouedec of the TRL team announced that Mistral's new flagship Mistral Large 4 (dubbed "Le Chonk", 1T parameters, claimed strongest European model) is now trainable in TRL. The repo adds a full SFT example with a training script (sftmistrallarge4.py) and a Slurm launch script, enabling researchers to fine-tune the trillion-parameter model on their own clusters. The author quipped this is how you concretely build AI sovereignty.
More from Models
- Why LLMs can crack Millennium Prize Problems but still ship broken UIs: verifiable rewards — iamsahaj_xyz · 2026-10-30
- Token math suggests quota-hungry users may be behind Anthropic's aggressive bans — teortaxesTex · 2026-10-30
- Le Chonk: Mistral's rumored open-weights model teased on Hugging Face, reportedly its most cyber-capable yet — Bam4d · 2026-10-30
- ChatGPT Slaps a Suicide Hotline Ad Under a Statistics Study Guide — skdh · 2026-10-30
- Groot-19B-A3B Local Test: Spiral Galaxy in 90 Seconds on a MacBook, Qwen 27B Takes 20 Minutes — mhdfaran · 2026-10-30
- Mistral researcher teases dropping 'the best western OSS model' — Bam4d · 2026-10-30