All pull requests

Pull requests

• fix(deployment): wait for all nodes to join Ray cluster in mcjax ray templates

  #2611·andytwigg opened Oct 2, 2026Collaborator·

  ·
  4

  

• Wuhao/fix orch nodepool and ckpt opt

  #2609·knightwh opened Oct 2, 2026Collaborator·6 tasks·

  ·
  2

  

• Bump maxtext to 80424f6 and add CHECKPOINT_ABANDON_FAILED_SAVES

  #2607·A9isha opened Oct 2, 2026Collaborator·

  ·
  1

  

• fix(distributed): verify worker_id on RPCs, hash JOB_PREFIX ports, and configure sandbox RBAC/NetworkPolicy

  #2606·andytwigg opened Oct 2, 2026Collaborator·

  ·
  4

  

• Update pinned Raiden wheel to dev991963055 in install_raiden.sh

  #2605·justinlu13 opened Oct 2, 2026·

  ·
  2

  

• Feat(Deployment): Support colocated_python checkpointing

  #2603·YixuanWang-99 opened Oct 2, 2026Collaborator·

  ·
  

• DeepSWE RL: non-blocking log fields for per-step performance attribution

  #2602·ultrons opened Oct 2, 2026·

  ·
  6

  

• Add the tunix/experimental/zero_tim_kernel package and Qwen3 canonical model benchmark suite (tunix/experimental/zero_tim_kernel/bench/qwen3):

  #2600·copybara-service[bot] opened Oct 1, 2026Bot·

  ·
  

• [MLPerf] Support GCS persistence and restore for JAX compilation cache in rollout and trainer

  #2599·wenxindongwork opened Oct 1, 2026Collaborator·

  ·
  

• Decouple trajectory store interfaces and implementations from TunixTrajectoryMetadata so stores can statically type, preserve, and rehydrate arbitrary TrajectoryMetadata subclasses.

  #2598·copybara-service[bot] opened Oct 1, 2026Bot·

  ·
  

• Update mlperf_base.sh for the dataset

  #2595·knightwh opened Oct 1, 2026Collaborator·6 tasks·

  ·
  1

  

• Prioritize the current step's rollouts in vLLM under max_staleness

  #2593·s-noghabi opened Oct 1, 2026Collaborator·

  ·
  4

  

• feat(weight_sync): overlap Raiden H2H transfer with rollout sampler p…

  #2581·knightwh opened Oct 1, 2026Collaborator·6 tasks·

  ·
  7

  

• feat(mlperf): automatic offline eval loop for MLPerf RL checkpoints

  #2574·uwlei opened Oct 1, 2026Collaborator·

  ·
  16

  

• Router replay: pack routed_experts contiguously and defer its padding past the trainer RPC

  #2572·lc5211 opened Sep 30, 2026Collaborator·

  ·
  3

  

• fix(utils): prevent trajectory drops and parallelize per-step GCS writes in AsyncTrajectoryLogger

  #2571·niting opened Sep 30, 2026Collaborator·

  ·
  5

  

• [experimental] orchestrator: fail the run when a worker rejoins (preemption)

  #2569·s-noghabi opened Sep 30, 2026Collaborator·

  ·
  8

  

• feat(orchestrator): support elastic rollout worker leave, re-registration, and recovery

  #2568·andytwigg opened Sep 30, 2026Collaborator·

  ·
  6

  

• feat(recipes): add FP8_MOE environment variable directly to mlperf recipes

  #2548·wenxindongwork opened Sep 30, 2026Collaborator·

  ·
  4

  

• [tunix] Add prepared diffusion policy scoring

  #2543·ethannnnnn opened Sep 29, 2026Contributor·

  ·
  4

  

• [tunix] Add prepared diffusion distillation batches

  #2542·ethannnnnn opened Sep 29, 2026Contributor·

  ·
  5

  

• Consolidates static rollout configuration onto RolloutConfig across the rollout stack:

  #2540·copybara-service[bot] opened Sep 29, 2026Bot·

  ·
  

• enable RPC retry

  #2539·copybara-service[bot] opened Sep 29, 2026Bot·

  ·
  

• Extends the progressive load recovery benchmark harness (data_generator.py, benchmark_lib.py, and run_benchmark.py) to simulate multi-worker concurrent trajectory appends and measure write throughput across both shared-writer and dedicated per-worker writer topologies.

  #2536·copybara-service[bot] opened Sep 29, 2026Bot·

  ·
  1

  

• Integrates SqlTrajectoryStore into the progressive load recovery benchmark harness (run_benchmark.py and benchmark_lib_test.py).

  #2535·copybara-service[bot] opened Sep 29, 2026Bot·

  ·
  1