All pull requests

Pull requests

• Recognize the Muse Spark Open chat template revision with the corrected generation markers

  #8275·albertvillanova opened Nov 12, 2026Member·

  ·
  1

  1

• Final- Promote AsyncGRPOTrainer to stable

  #8274·AmineDiro opened Nov 12, 2026Member·

  ·
  1

• Recognize the SmolVLM2 chat template revision that renders string content

  #8272·albertvillanova opened Nov 12, 2026Member·

  ·
  1

  1

• Fix the CLI parser on Python 3.15

  #8271·qgallouedec opened Nov 12, 2026Member·

  ·
  2

• Merge LoRA adapters out of place for the vLLM weight sync

  #8269·qgallouedec opened Nov 12, 2026Member·

  ·
  1

  7

• Use non-reentrant gradient checkpointing in the Online DPO visual math example

  #8268·albertvillanova opened Nov 12, 2026Member·

  ·
  1

• Raise early when the MoE auxiliary loss meets reentrant gradient checkpointing

  #8267·albertvillanova opened Nov 12, 2026Member·1 task·

  ·

• Only call enable_input_require_grads() before transformers 5.0 in the experimental trainers

  #8266·albertvillanova opened Nov 12, 2026Member·

  ·
  1

• Let transformers set use_reentrant for QLoRA in experimental prepare_peft_model

  #8265·albertvillanova opened Nov 12, 2026Member·

  ·
  1

• Apply the tiny model revision overrides when the scripts pass revision="main"

  🔧 maintenance
  #8264·albertvillanova opened Nov 12, 2026Member·

  ·
  1

  1

• Pin every tiny test model to a Hub commit

  🔧 maintenance
  #8262·albertvillanova opened Nov 12, 2026Member·

  ·
  1

• Check the log-prob gap only on the first step in the GRPO structured outputs test

  🔧 maintenance
  #8261·albertvillanova opened Nov 12, 2026Member·

  ·
  1

  1

• Compare version checks against plain release versions, not .dev0

  #8256·qgallouedec opened Nov 11, 2026Member·

  ·
  1

• Test GRPO's environments through training and logged metrics

  🔧 maintenance
  #8251·qgallouedec opened Nov 10, 2026Member·

  ·
  4

• Count each AsyncGRPO reward function's mean once per scored rollout, as GRPO does

  #8250·qgallouedec opened Nov 10, 2026Member·

  ·
  2

• Keep the newlines the model writes before a Qwen3 tool call when parsing

  #8244·AmineDiro opened Nov 10, 2026Member·

  ·
  1

• Scale the FP8 blocks of the tiny DeepSeek-V3 models down to 32x32

  🔧 maintenance
  #8240·albertvillanova opened Nov 10, 2026Member·

  ·
  5

• Write the tiny Mistral Large 4 FP8 checkpoint in the reference layout

  🔧 maintenance
  #8239·albertvillanova opened Nov 10, 2026Member·

  ·
  3

• Compare tiny-model dtypes with references that only ship consolidated safetensors

  🔧 maintenance
  #8238·albertvillanova opened Nov 10, 2026Member·

  ·
  1

• Return the candidate set of each sampled token from the vLLM server in GRPO

  #8235·albertvillanova opened Nov 10, 2026Member·

  ·
  9

• Handle a missing content key in the DeepSeek training chat templates

  #8228·albertvillanova opened Nov 10, 2026Member·

  ·
  1

  2

• Quantize the tiny GLM-5.3-Flash model in FP8 like its reference

  🔧 maintenance
  #8227·albertvillanova opened Nov 10, 2026Member·1 task·

  ·

• Add URM loss for noisy preferences to RewardTrainer

  #8210·cswjl opened Nov 10, 2026Contributor·5 of 8 tasks·

  ·
  1

• Train on conversational data in trainer tests

  #8195·qgallouedec opened Nov 9, 2026Member·

  ·
  17

• Use the activation checkpointing methods on transformers 5.23

  #8194·albertvillanova opened Nov 9, 2026Member·1 task·

  ·
  1

  2