All pull requests

Pull requests

• [tinker] Honour train_attn/train_mlp on the Megatron backend

  #2529·ArturNiederfahrenhorst opened Oct 22, 2026·

  ·
  1

  6

• [megatron] GLM-5.3-Flash long-context integration branch (all optimizations, not for merge)

  #2528·avigyabb opened Oct 22, 2026Collaborator·

  ·

• feat(train): time optimizer steps separately from forward and backward

  x-lila
  #2527·dyurk-lila opened Oct 22, 2026Collaborator·

  ·
  1

• [fix] Stop Gated DeltaNet state leaking between packed sequences

  #2526·anthonypetertc opened Oct 22, 2026·

  ·
  17

• [megatron] Fix OOM on native checkpoint save and resume

  x-lila
  #2525·sgabriel92 opened Oct 22, 2026·

  ·
  9

• [skyrl-train][bugfix] Run the sampler weight broadcast off the event loop

  x-lila
  #2524·dyurk-lila opened Oct 21, 2026Collaborator·

  ·
  8

• [sft] Count context parallelism in the batch-size check

  x-lila
  #2523·dyurk-lila opened Oct 21, 2026Collaborator·

  ·
  1

• [fix][megatron] Call mark_not_offload in the fused LM head path

  x-lila
  #2522·dyurk-lila opened Oct 21, 2026Collaborator·

  ·
  4

• perf(megatron): fuse sample-support candidate projection with Triton

  x-lila
  #2521·dyurk-lila opened Oct 21, 2026Collaborator·9/9·

  ·
  4

• [tinker] Reject backend_config values the server overrides on the SkyRL-Train backends

  #2516·ArturNiederfahrenhorst opened Oct 20, 2026·

  ·
  1

  13

  

• Add opt-in policies for unbridged Harbor skycap rollouts

  #2514·jadenfix opened Oct 19, 2026Contributor·

  ·
  1

• [megatron][vllm] Qwen3.8-Flash-Next (qwen4_exp) support: Megatron bridge, QSA sparse attention, GSM8K + DAPO recipes

  #2513·xinyuangui2 opened Oct 17, 2026Collaborator·8 of 12 tasks·

  ·
  1

• [megatron][vlm] freeze_language_model / freeze_vision_model / freeze_vision_projection

  #2509·gc-anyscale opened Oct 16, 2026Contributor·

  ·
  18

• [fsdp] Initialize a LoRA policy from a PEFT adapter (lora.init_adapter_path)

  #2507·avigyabb opened Oct 16, 2026Collaborator·

  ·

• [tinker] Free the colocate placement group after a failed model build

  #2504·spicysharkco opened Oct 16, 2026·

  ·
  1

  1

• [LoRA] Initialize LoRA policies from a PEFT adapter (lora.init_adapter_path): resume Tinker runs on Megatron/FSDP

  #2503·avigyabb opened Oct 16, 2026Collaborator·

  ·
  3

• [sft] Add lazy_vlm_dataset: tokenize VLM datasets per fetch instead of holding every row's pixel_values

  #2498·gc-anyscale opened Oct 15, 2026Contributor·

  ·
  7

  

• [docs][skycap] Add a skycap page to the SkyRL docs

  #2494·kouroshHakha opened Oct 15, 2026Collaborator·

  ·

• WIP: [Feature] Add ADAPO controller

  #2491·catherinelee274 opened Oct 15, 2026·

  ·
  8

• [feat][megatron] Support optimizer state offload to disk (optimizer_offload_to_disk)

  #2490·SumanthRH opened Oct 15, 2026Member·

  ·
  1

• [examples][skycap] Carry the sampler's support logprobs through harbor_skycap

  x-lila
  #2485·dyurk-lila opened Oct 14, 2026Collaborator·3/3·

  ·
  4

  

• [skyrl-train] Score centering over the recorded sample support

  run_gpu_cirun_megatron_gpu_cix-lila
  #2484·dyurk-lila opened Oct 14, 2026Collaborator·2/3·

  ·
  6

• [chore] Use random tokens in the full context trainer's dummy batch

  #2479·SumanthRH opened Oct 14, 2026Member·

  ·

• [harbor] Run Harbor's modal environment on Ray Sandbox

  #2478·xyuzh opened Oct 13, 2026·

  ·
  10

• [feat][megatron] Add an opt-in streaming write for async checkpoints

  #2477·bvolpato opened Oct 13, 2026Contributor·

  ·