Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on May 6, 2025

• 🎲 [GRPO] Shuffle mini batches (#3391)

  Show description for adfa7fd
  

  3 peopleauthored
  adfa7fdView commit details

  Copy full SHA for adfa7fd

  Browse repository at this point

• 💔 [GRPO] Decouple gradient accumulation from the number of minibatches generated (#3388)

  Show description for cf5183d
  

  edbeeching

  and

  qgallouedec

  authored
  cf5183dView commit details

  Copy full SHA for cf5183d

  Browse repository at this point

• 🤝 Compatibility of the TRL CLI with accelerate arguments (#3409)

  Show description for 1954c02
  

  qgallouedec

  and

  lewtun

  authored
  1954c02View commit details

  Copy full SHA for 1954c02

  Browse repository at this point

• ✌️ Add support for FSDP2 (#3317)

  Show description for 45f4c58
  

  lewtun

  and

  qgallouedec

  authored
  45f4c58View commit details

  Copy full SHA for 45f4c58

  Browse repository at this point

Commits on May 2, 2025

• 🕊️ Un-restrict diffusers (#3407)

  qgallouedec

  authored
  cc044e3View commit details

  Copy full SHA for cc044e3

  Browse repository at this point

• 🕺 Migrate setup configuration from `setup.py` to `setup.cfg` and make `rich` an optional dep (#3403)

  Show description for 999acd5
  

  qgallouedec

  and

  lewtun

  authored
  999acd5View commit details

  Copy full SHA for 999acd5

  Browse repository at this point

• 🪪 Remove license classifier (#3402)

  qgallouedec

  authored
  8606b1aView commit details

  Copy full SHA for 8606b1a

  Browse repository at this point

• 👉 [DPO] Model forward pass padding side fix (#3307)

  Show description for a673da5
  

  LeonEricsson

  and

  qgallouedec

  authored
  a673da5View commit details

  Copy full SHA for a673da5

  Browse repository at this point

• 🦁 Fix liger initialization (#3401)

  Show description for 00b8e31
  

  shivam15s

  and

  qgallouedec

  authored
  00b8e31View commit details

  Copy full SHA for 00b8e31

  Browse repository at this point

Commits on May 1, 2025

• 💔 [SFT] Raise error when `formatting_func` is used with `completion_only_loss` (#3385)

  Show description for c163cf5
  

  LeonEricsson

  and

  qgallouedec

  authored
  c163cf5View commit details

  Copy full SHA for c163cf5

  Browse repository at this point

• ✨ [IterativeSFT] Small refresher (#3378)

  LeonEricsson

  authored
  bc9c019View commit details

  Copy full SHA for bc9c019

  Browse repository at this point

• 🧑‍🤝‍🧑 Co-Locating vLLM w/ training to for higher throughput and GPU utilization (#3394)

  Show description for 18596cf
  

  toslali-ibm

  and

  qgallouedec

  authored
  18596cfView commit details

  Copy full SHA for 18596cf

  Browse repository at this point

• 🌊 Add MLflow metrics in profiling context (#3400)

  Show description for 280d353
  

  dhruvmullick

  and

  qgallouedec

  authored
  280d353View commit details

  Copy full SHA for 280d353

  Browse repository at this point

• [GRPO] Reference model initialization bug fix (#3397)

  LeonEricsson

  authored
  13fa840View commit details

  Copy full SHA for 13fa840

  Browse repository at this point

Commits on Apr 30, 2025

• [🐯+GRPO] Support FSDP + Fix bug when using LigerGRPO with DDP (#3260)

  Show description for 09b669f
  

  5 peopleauthored
  09b669fView commit details

  Copy full SHA for 09b669f

  Browse repository at this point

Commits on Apr 29, 2025

• Deprecate `TextEnvironment` and tools (#3389)

  lewtun

  authored
  01d0be1View commit details

  Copy full SHA for 01d0be1

  Browse repository at this point

• DPO fixes for evaluations (#3377)

  winglian

  authored
  3a42af1View commit details

  Copy full SHA for 3a42af1

  Browse repository at this point

• PEFT support for Liger GRPO (#3355)

  Show description for aaf3960
  

  salmanmohammadi

  and

  kashif

  authored
  aaf3960View commit details

  Copy full SHA for aaf3960

  Browse repository at this point

Commits on Apr 28, 2025

• 📋 Allow calling trl cli in sft mode with config file (#3380)

  Show description for 2bf4847
  

  CloseChoice

  and

  qgallouedec

  authored
  2bf4847View commit details

  Copy full SHA for 2bf4847

  Browse repository at this point

Commits on Apr 26, 2025

• ⚰️ Remove deprecated (#3364)

  qgallouedec

  authored
  a8cfca6View commit details

  Copy full SHA for a8cfca6

  Browse repository at this point

• Better guards for DeepSpeed imports (#3351)

  lewtun

  authored
  1bca495View commit details

  Copy full SHA for 1bca495

  Browse repository at this point

• 🎭 Fix train and eval mode checking in `GRPOTrainer` and `SFTTrainer` (#3337)

  Show description for 39e9639
  

  3 peopleauthored
  39e9639View commit details

  Copy full SHA for 39e9639

  Browse repository at this point

• 🥸🔢 Adding pad_multiple to SFT trainer (#3365)

  shirinyamani

  authored
  8e6ed93View commit details

  Copy full SHA for 8e6ed93

  Browse repository at this point

Commits on Apr 25, 2025

• 🔢 Pad to multiple of (#3362)

  qgallouedec

  authored
  29c5e05View commit details

  Copy full SHA for 29c5e05

  Browse repository at this point

Commits on Apr 24, 2025

• ⬆️ Bump dev version (#3357)

  qgallouedec

  authored
  a9b27f8View commit details

  Copy full SHA for a9b27f8

  Browse repository at this point

• Release: v0.17 (#3356)

  qgallouedec

  authored
  cd6b3deView commit details

  Copy full SHA for cd6b3de

  Browse repository at this point

• ⚡ Up to 4x faster: Data Parallel for vLLM server (#3310)

  Show description for 36685c8
  

  3 peopleauthored
  36685c8View commit details

  Copy full SHA for 36685c8

  Browse repository at this point

Commits on Apr 23, 2025

• 🍡 Fix using reward model and DeepSpeed ZeRO 3 (#3326)

  qgallouedec

  authored
  89556c8View commit details

  Copy full SHA for 89556c8

  Browse repository at this point

• Define default chat template for SFT (#3309)

  Show description for f3e8c23
  

  lewtun

  and

  qgallouedec

  authored
  f3e8c23View commit details

  Copy full SHA for f3e8c23

  Browse repository at this point

• 🏁 Fix adding special tokens in SFT (#3328)

  qgallouedec

  authored
  9ee6c3aView commit details

  Copy full SHA for 9ee6c3a

  Browse repository at this point

Commits on Apr 22, 2025

• [CPO] Check that `max_prompt_length < max_length` (#3341)

  LeonEricsson

  authored
  ef05331View commit details

  Copy full SHA for ef05331

  Browse repository at this point

• 🦄 Add optional uvicorn log level for vLLM serve (#3338)

  Show description for 05e2ba6
  

  3 peopleauthored
  05e2ba6View commit details

  Copy full SHA for 05e2ba6

  Browse repository at this point

• 💡 Fix type hint in `_generate_and_score_completions` (#3336)

  syt-nju

  authored
  1b4f189View commit details

  Copy full SHA for 1b4f189

  Browse repository at this point

• 🧸 Fix unset tokenizer pad_token (#3290)

  Show description for 1faa7f9
  

  LeonEricsson

  and

  qgallouedec

  authored
  1faa7f9View commit details

  Copy full SHA for 1faa7f9

  Browse repository at this point

• ✅ [doc] Update sft_trainer.md in table x->✓ (#3313)

  Show description for 66e6eab
  

  HERIUN

  and

  qgallouedec

  authored
  66e6eabView commit details

  Copy full SHA for 66e6eab

  Browse repository at this point