Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Jun 26, 2025

• [GRPO] Make sure special tokens aren't lost when truncating prompt. (#3651)

  pramodith

  authored
  79ec242View commit details

  Copy full SHA for 79ec242

  Browse repository at this point

Commits on Jun 25, 2025

• Add entropy based filtering inside the GRPOTrainer. (#3563)

  Show description for 7e8ef86
  

  3 peopleauthored
  7e8ef86View commit details

  Copy full SHA for 7e8ef86

  Browse repository at this point

• 🤝 validate gradient_accumulation_steps vs steps_per_generation for on-policy GRPO (#3493)

  Show description for 32df093
  

  HarryHsing

  and

  shirinyamani

  authored
  32df093View commit details

  Copy full SHA for 32df093

  Browse repository at this point

• ☕️ GRPO script reward_funcs error (#3639)

  Show description for 0336e4b
  

  tcapelle

  and

  shirinyamani

  authored
  0336e4bView commit details

  Copy full SHA for 0336e4b

  Browse repository at this point

Commits on Jun 24, 2025

• Update dpo_vlm.py (#3629)

  Show description for ab331bf
  

  Clement25

  and

  shirinyamani

  authored
  ab331bfView commit details

  Copy full SHA for ab331bf

  Browse repository at this point

• env var for vllm colocate exp added (#3638)

  shirinyamani

  authored
  84d7b5bView commit details

  Copy full SHA for 84d7b5b

  Browse repository at this point

• fixing num_processes (#3637)

  shirinyamani

  authored
  b40c959View commit details

  Copy full SHA for b40c959

  Browse repository at this point

• 🐛 fix grpo generation_kwargs (#3634)

  Show description for 34fa6b9
  

  ahatamiz

  and

  kashif

  authored
  34fa6b9View commit details

  Copy full SHA for 34fa6b9

  Browse repository at this point

Commits on Jun 22, 2025

• Revert "🔍 Add guidance on choosing `max_length` value and include visualization tool"

  Show description for eef7a43
  qgallouedec

  committed
  eef7a43View commit details

  Copy full SHA for eef7a43

  Browse repository at this point

• 🔍 Add guidance on choosing `max_length` value and include visualization tool

  qgallouedec

  committed
  89c699fView commit details

  Copy full SHA for 89c699f

  Browse repository at this point

Commits on Jun 20, 2025

• ⬆️ Bump dev version (#3626)

  qgallouedec

  authored
  559a99fView commit details

  Copy full SHA for 559a99f

  Browse repository at this point

• Release: v0.19 (#3625)

  qgallouedec

  authored
  5b3ea9dView commit details

  Copy full SHA for 5b3ea9d

  Browse repository at this point

• 🧰 [SFT] Tool support (#3597)

  Show description for c262674
  

  qgallouedec

  and

  kashif

  authored
  c262674View commit details

  Copy full SHA for c262674

  Browse repository at this point

• 🔍 Add test to verify chat template consistency (#3624)

  qgallouedec

  authored
  5c3dd3aView commit details

  Copy full SHA for 5c3dd3a

  Browse repository at this point

• ⚔️ Fix bf16 fp16 config conflict issue (#3598)

  Show description for 4c92de0
  

  yao-matrix

  and

  qgallouedec

  authored
  4c92de0View commit details

  Copy full SHA for 4c92de0

  Browse repository at this point

• 📜 Add `chat_template_path` parameter to `SFTConfig` (#3599)

  qgallouedec

  authored
  67f17f7View commit details

  Copy full SHA for 67f17f7

  Browse repository at this point

• 🧬 Add `generation_kwargs` as a property of `GRPOConfig` to support additional generation arguments. (#3617)

  Show description for 37a71e8
  

  pramodith

  and

  qgallouedec

  authored
  37a71e8View commit details

  Copy full SHA for 37a71e8

  Browse repository at this point

• [GRPO] Fix prompt truncation (`max_prompt_length`) with vLLM. (#3601)

  Show description for b0958c6
  

  LeonEricsson

  and

  qgallouedec

  authored
  b0958c6View commit details

  Copy full SHA for b0958c6

  Browse repository at this point

Commits on Jun 19, 2025

• ⭐ Add `vllm_gpu_memory_utilization` recommendation script (#3554)

  Show description for 8bad863
  

  3 peopleauthored
  8bad863View commit details

  Copy full SHA for 8bad863

  Browse repository at this point

• 🎁 Put the reward computation in a separate function (#3620)

  Show description for d004415
  

  ajtejankar

  and

  qgallouedec

  authored
  d004415View commit details

  Copy full SHA for d004415

  Browse repository at this point

• 🤵‍♂️ SFT on assistant messages only (#3586)

  Show description for 9554c2f
  

  qgallouedec

  and

  lewtun

  authored
  9554c2fView commit details

  Copy full SHA for 9554c2f

  Browse repository at this point

• 🦘 Skip no-op ChatML conversion for datasets already in ChatML format (#3594)

  qgallouedec

  authored
  712afd5View commit details

  Copy full SHA for 712afd5

  Browse repository at this point

• 📚 SFTTrainer support chat template kwargs (#3609)

  qgallouedec

  authored
  086e9d5View commit details

  Copy full SHA for 086e9d5

  Browse repository at this point

• 🔖 Fix: ensure user-provided `labels` are retained in self._signature_columns (#3589)

  Show description for 5206c92
  

  sxndqc

  and

  qgallouedec

  authored
  5206c92View commit details

  Copy full SHA for 5206c92

  Browse repository at this point

• 👔 Apply doc-builder style (#3615)

  Show description for e4b586a
  

  qgallouedec

  and

  kashif

  authored
  e4b586aView commit details

  Copy full SHA for e4b586a

  Browse repository at this point

• 🏛️ Fix CI and Iterative SFT (#3614)

  qgallouedec

  authored
  0576346View commit details

  Copy full SHA for 0576346

  Browse repository at this point

Commits on Jun 18, 2025

• 🏁 Refactor reference model initialization in GRPOTrainer (#3575)

  Show description for e63588a
  

  3 peopleauthored
  e63588aView commit details

  Copy full SHA for e63588a

  Browse repository at this point

• [SFT] Clarify default collator docs (#3606)

  LeonEricsson

  authored
  d9d25a7View commit details

  Copy full SHA for d9d25a7

  Browse repository at this point

• Change `enforce_eager` default value in vLLM server. (#3607)

  LeonEricsson

  authored
  58ea227View commit details

  Copy full SHA for 58ea227

  Browse repository at this point

• Fix Typos in Comments and Improve Clarity in Trainer Modules (#3596)

  maximevtush

  authored
  a768484View commit details

  Copy full SHA for a768484

  Browse repository at this point

• Fix: list-typed tags handling in `Trainer::create_model_card` (#3613)

  LeonEricsson

  authored
  d17ec7aView commit details

  Copy full SHA for d17ec7a

  Browse repository at this point

• 🗳️ Remove `logging_steps` parameter from for simpler setup (#3612)

  qgallouedec

  authored
  ed9b78aView commit details

  Copy full SHA for ed9b78a

  Browse repository at this point

• ♻️ Avoids redundant calculation of ref logps in the new policy update loop (#3600)

  Show description for d6a969f
  

  zkpranav

  and

  qgallouedec

  authored
  d6a969fView commit details

  Copy full SHA for d6a969f

  Browse repository at this point

Commits on Jun 15, 2025

• Fix Typo in Documentation and Notebook; Improve Library Installation Comment (#3593)

  zeevick10

  authored
  8a235a9View commit details

  Copy full SHA for 8a235a9

  Browse repository at this point

• Fix typos and improve metric descriptions in documentation (#3585)

  vtjl10

  authored
  afa06c3View commit details

  Copy full SHA for afa06c3

  Browse repository at this point