All pull requests

Pull requests

• feat(pdd): add Qwen-Image 2.1 PDD training from SGLang rollouts

  #365·niehen6174 opened Nov 7, 2026Collaborator·

  ·

• ci: install sglang's radix_transfer_plugins beside sglang on both CI paths

  #364·zhihengy opened Nov 6, 2026Collaborator·

  ·

• build: install sglang's radix_transfer_plugins in the image

  #363·zhihengy opened Nov 6, 2026Collaborator·

  ·

• feat(fsdp): read a shard's store slices back from the output store before training on it

  run-ci-e2e
  #361·zhihengy opened Nov 6, 2026Collaborator·3/3·

  ·

• feat(rollout): carry a rollout's stored tensors as store slices and release them after its train

  run-ci-e2e
  #360·zhihengy opened Nov 6, 2026Collaborator·2/3·

  ·

• feat(rollout): add --use-output-store with its Mooncake master and the engines' store config

  run-ci-e2e
  #359·zhihengy opened Nov 6, 2026Collaborator·1/3·

  ·

• feat(fsdp): support torch 2.14 in the param-dtype patch

  run-ci-fsdprun-ci-torch
  #357·Rockdu opened Nov 5, 2026Collaborator·5 of 6 tasks·

  ·

• feat(sft): invert H3's distilled CFG to supervise plain velocity

  #353·Rockdu opened Nov 4, 2026Collaborator·4 of 6 tasks·

  ·

• feat(pi05): add native SFT support and optional LIBERO workflow

  #352·zhangwt20011015 opened Nov 3, 2026·

  ·
  

• [DMD2 10/10] feat(dmd2): add --dmd2-critic-scoring original, the reference's separate scoring before the critic's last step

  #351·zhihengy opened Nov 2, 2026Collaborator·10/10·5 of 6 tasks·

  ·

• fix(sft): retry transient encode-server failures and keep partial progress

  #349·zhihengy opened Oct 30, 2026Collaborator·4/4·

  ·

• feat(flux3-image): add FLUX.3 Image SFT through an encode-only server

  #345·zhihengy opened Oct 28, 2026Collaborator·3/4·

  ·

• feat(sft): encode SFT cache misses through an sglang-diffusion encode-only server

  #344·zhihengy opened Oct 28, 2026Collaborator·2/4·

  ·

• [DMD2 9/10] feat(dmd2): add the MiniMax H3 4-step DMD2 recipe with eval media and DMD2 metrics

  #343·zhihengy opened Oct 26, 2026Collaborator·9/10·5 of 6 tasks·

  ·

• [DMD2 8/10] perf(ray): sleep each trainer group right after its train

  #342·zhihengy opened Oct 26, 2026Collaborator·8/10·5 of 6 tasks·

  ·

• [DMD2 4/10] feat(fsdp): add a train-input hook that preprocesses train pairs before micro-batching

  #338·zhihengy opened Oct 24, 2026Collaborator·4/10·6 tasks done·

  ·
  1

• [DMD2 7/10] feat(dmd2): train the student with the distribution matching gradient from the critic group's scores

  #334·zhihengy opened Oct 19, 2026Collaborator·7/10·5 of 6 tasks·

  ·

• [DMD2 6/10] feat(dmd2): train a fake-score critic group from the teacher and score the student's samples

  #333·zhihengy opened Oct 19, 2026Collaborator·6/10·5 of 6 tasks·

  ·
  1

• [DMD2 5/10] feat(rollout): stop the rollout at a random step for DMD2 backward simulation

  run-ci-basic
  #331·zhihengy opened Oct 19, 2026Collaborator·5/10·6 tasks done·

  ·
  1

• [DMD2 3/10] feat(ray): hand earlier trainer groups' outputs to later groups as external_data

  run-ci-basic
  #330·zhihengy opened Oct 19, 2026Collaborator·3/10·6 tasks done·

  ·
  2

• [DMD2 2/10] feat(ray): train several colocated trainer groups in the sync driver

  run-ci-basic
  #329·zhihengy opened Oct 19, 2026Collaborator·2/10·6 tasks done·

  ·
  1

• fix(rollout): stop dynamic sampling from leaking generation past the rollout

  #324·OnePunchMonk opened Oct 17, 2026Contributor·6 tasks done·

  ·
  3

• feat(flux3-image): add FLUX 3 Image Klein RL support

  #316·Rockdu opened Oct 16, 2026Collaborator·1/4·

  ·
  1

• perf: startup timing probes and env-gated startup optimizations

  #310·OnePunchMonk opened Oct 11, 2026Contributor·

  ·
  7

• fix(rollout): wrap data source across epochs and cancel leftover rollout tasks

  #301·OnePunchMonk opened Oct 7, 2026Contributor·

  ·
  1

  1