All pull requests

Pull requests

• Fix sampler-trainer Pearson correlation via Chan's parallel covariance formula

  #2466·copybara-service[bot] opened Sep 26, 2026Bot·

  ·
  

• Plumb sampling temperature and vLLM processed_logprobs across RL pipeline

  #2465·copybara-service[bot] opened Sep 26, 2026Bot·

  ·
  

• Update mlperf_397b_256_v7x.sh

  #2462·knightwh opened Sep 25, 2026Collaborator·

  ·
  3

  

• Refactor step lifecycle management in TrajectoryCollectEngine and Trajectory:

  #2461·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  

• (v2 sampler) Add update_from_output to Scheduler.

  #2460·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• (v2 sampler) Add schedule_step to Scheduler.

  #2459·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• Add matched FrozenLake RL benchmark and trainer parity.

  #2456·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  

• Add distributed DeepSWE evaluation and episode summaries.

  #2455·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  

• [Tunix] Bind source FFI weights in RaidenSynchronizer.bind() under direct device buffer mode

  #2454·justinlu13 opened Sep 25, 2026·

  ·
  1

  

• Remove unused name_resolver and worker_rpc_client from RaidenHandler

  #2453·justinlu13 opened Sep 25, 2026·

  ·
  

• - Add _use_direct_device_buffer() in raiden_synchronizer.py to check RAIDEN_FFI_USE_DIRECT_DEVICE_BUFFER.

  #2452·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• cherry-pick(main): Auto-run TPU unit tests on Copybara-synced PRs and enable CI on atwigg/mlperf

  #2451·fineguy opened Sep 25, 2026Contributor·

  ·
  1

  

• add more test coverage for invalid masking in advantage

  #2449·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  

• (v2 sampler) Add request admission to Scheduler.

  #2447·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• fix(checkpointing): translate overwrite to force in MaxText adapter

  #2444·andytwigg opened Sep 25, 2026Collaborator·

  ·
  8

  

• (v2 sampler) Introduce Scheduler.

  ready-to-submit
  #2443·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• (v2 sampler) Add prefix caching to KVCacheManager.

  ready-to-submit
  #2442·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• (v2 sampler) Introduce KVCacheManager.

  ready-to-submit
  #2441·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• (v2 sampler) Add allocation to KVCacheManager.

  ready-to-submit
  #2440·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  1

  

• fuse TIS

  #2439·copybara-service[bot] opened Sep 25, 2026Bot·

  ·
  

• feat(rl): exclude invalid and masked trajectories from advantage calculation

  #2437·s-noghabi opened Sep 24, 2026Collaborator·

  ·
  4

  

• (v2 sampler) Add public helpers to SingleTypeKVCacheManager.

  ready-to-submit
  #2436·copybara-service[bot] opened Sep 24, 2026Bot·

  ·
  1

  

• (v2 sampler) Add allocation to SingleTypeKVCacheManager.

  ready-to-submit
  #2434·copybara-service[bot] opened Sep 24, 2026Bot·

  ·
  1

  

• (v2 sampler) Add prefix cache lookup to SingleTypeKVCacheManager.

  ready-to-submit
  #2433·copybara-service[bot] opened Sep 24, 2026Bot·

  ·
  1

  

• (v2 sampler) Add state sync to SingleTypeKVCacheManager for v2 sampler.

  ready-to-submit
  #2432·copybara-service[bot] opened Sep 24, 2026Bot·

  ·
  1