All pull requests

Pull requests

• Add Electron + Flask desktop scaffold for writing-style generator (gpt-4)

  #125·Treat125 opened Sep 4, 2026·

  ·

• feat: add Tavily search backend to search-r1 example

  #122·manisrinivasan2k1 opened Jul 17, 2026·

  ·

• Fix Qwen3.5 mbridge wrapper config handling

  #121·allplepine opened Jul 9, 2026·

  ·
  1

• [codex] fix terminal remote setup uvicorn dependency

  #120·Squirtle-gpj opened Jul 6, 2026·

  ·

• [codex] fix remaining PYTHONUNBUFFERED typos

  #119·Squirtle-gpj opened Jul 6, 2026·

  ·

• fix: skip DDP and optimizer for prm_teacher (inference-only mode)

  #115·HeJW0303 opened May 14, 2026·3 tasks done·

  ·

• Add Ascend NPU support

  #103·quancs opened Apr 25, 2026·

  ·

• feat: support resuming existing W&B runs

  #102·Zhuang-Zhuang-Liu opened Apr 24, 2026·

  ·

• feat: implement single-GPU (24GB) INT4 QLoRA + offload-rollout + external PRM for OpenClaw-Combine and test

  #97·wuwanchun opened Apr 20, 2026·

  ·

• fix: repad rollout samples after sample filtering

  #79·xuyangc03 opened Apr 15, 2026Contributor·

  ·

• openclaw-tinker: support Qwen3.5 tool calling

  #57·ojus1 opened Mar 31, 2026·

  ·
  1

• Fix "strip_thinking function" and "default model name for OpenClaw requests"

  #53·zlh-source opened Mar 30, 2026·

  ·

• fix: support terminal-bench original-tasks layout

  #50·ningzimu opened Mar 28, 2026·

  ·

• Add initial Qwen3.5-4B support and SGLang text-runtime patch

  #39·ShenAC-SAC opened Mar 19, 2026·

  ·

• feat: add MiniMax M3 and M2.7 model metadata

  #37·octo-patch opened Mar 18, 2026·

  ·

• docs: add Chinese documentation for core RL methods

  #33·guozhihao-224 opened Mar 17, 2026·

  ·

• feat(openclaw-rl): add reward model feedback training

  #28·selfattn opened Mar 14, 2026·

  ·
  3

• feat: Support single gpu

  #27·wyqsss opened Mar 13, 2026·

  ·
  2

• fix: add cd to SLIME_ROOT and derive MEGATRON_ROOT dynamically

  #21·JasonOA888 opened Mar 12, 2026·2 tasks done·

  ·

• feat: add openclaw-training-free — zero-GPU experience loop for LLMs

  #6·zcyynl opened Mar 5, 2026·

  ·

• fix: add cd to SLIME_ROOT before train_async.py in all training scripts

  #5·kaysonyu opened Mar 4, 2026·

  ·

• fix: derive Megatron-LM PYTHONPATH from script location

  #3·koriyoshi2041 opened Mar 1, 2026·4 tasks done·

  ·

• docs: fix kl-loss-coef default value in instructions

  #2·koriyoshi2041 opened Mar 1, 2026·2 tasks done·

  ·

• fix: correct PYTHONBUFFERED typo to PYTHONUNBUFFERED

  #1·koriyoshi2041 opened Mar 1, 2026·3 tasks done·

  ·