All pull requests

Pull requests

• fix: LLaMA bridge on unscaled RoPE under transformers 5; add Mistral bridge

  #163·jbelofamd opened Nov 3, 2026·

  ·
  2

• fix: handle null Hugging Face RoPE scaling

  #162·kaixih opened Oct 10, 2026·

  ·

• fix: support fused RoPE for Qwen3.5 mRoPE

  #160·ewan0x79 opened Sep 23, 2026·

  ·

• fix(core): clamp the first CP chunk slice in preprocess_packed_seqs

  #157·gaohongkui opened Aug 17, 2026·

  ·

• [fix]: bugfix for RAY_EXPERIMENTAL_NOSET_ASCEND/CUDA_RT_VISIBLE_DEVICES in RL

  #151·xiazhahe opened Jun 11, 2026·

  ·

• [feat] Add DeepSeek-V4 support

  #147·Meirtz opened May 27, 2026·

  ·

• Add Megatron-FSDP load and export support

  #146·conver334 opened May 25, 2026·

  ·

• bugfix:Fix mbridge weight save logic

  #119·zyang6 opened Apr 20, 2026·

  ·

• Add layer number extraction for layernorm weights

  #118·rrrsj opened Apr 19, 2026·

  ·

• feat: add load_weights_from_async_generator to Bridge

  #117·sl-1314 opened Apr 14, 2026·

  ·

• [fix] handle FSDP DTensor in broadcast_from_megatron_pp

  #113·yxs opened Apr 2, 2026·

  ·
  1

• [fix] Fix qwen3.5 MoE vp_stage propagation under VPP

  #112·none0663 opened Apr 1, 2026·

  ·

• try fit in new transformers moe

  #110·ETOgaosion opened Mar 30, 2026Collaborator·

  ·
  1

• feat: Support quantization before weight resharding

  #92·hy2826 opened Mar 8, 2026·

  ·

• feat: support LongCat-Flash-Chat Model

  #82·VegetaPn opened Feb 14, 2026·

  ·

• feat: online load/save huggingface model weights for Megatron-FSDP

  #80·conver334 opened Feb 12, 2026·

  ·

• fix: get_tensor_model_parallel_group_if_none is missing for mcore v0.12.1

  #53·quancs opened Nov 13, 2025Contributor·

  ·