Pull requests: google/tunix
Pull requests list
Support DeepSWE evaluation pipeline on v5p-32 cluster with MaxText and Pathways
#2071
opened Sep 1, 2026 by
susanbao
Collaborator
Loading…
5 tasks done
[Tunix] Training Step Lineage Correlation & Sink Flush for Distributed RL.
#2070
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
Wire the TrajectoryStore through rollout collection workflows:
#2067
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
Integrate TrajectoryStore into TrajectoryCollectEngine to log turn steps and trajectory metadata during episode collection:
#2066
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
Support max_response_length to dynamically cap generation tokens during rollouts.
#2062
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
decouple force_on_policy_ratio from step0 reforward.
#2061
opened Sep 1, 2026 by
entrpn
Loading…
6 tasks done
refactor: move worker registration logic into orchestrator.
#2060
opened Sep 1, 2026 by
copybara-service
Bot
Loading…
Enable Raiden FFI for Pathways weight sync
ready-to-submit
#2059
opened Aug 31, 2026 by
wang2yn84
Collaborator
Loading…
6 tasks
Enable router replay in the GSM8K demo launchers
#2057
opened Aug 31, 2026 by
khatwanimohit
Collaborator
Loading…
6 tasks
Add optional trajectory_ids filtering to TrajectoryStore.get_trajectories_metadata across store implementations:
#2056
opened Aug 31, 2026 by
copybara-service
Bot
Loading…
Optimize Gemma 4 KV cache initialization to allocate on-device in HBM.
#2055
opened Aug 31, 2026 by
copybara-service
Bot
Loading…
Bootstrap rollout policy from target state
#2054
opened Aug 31, 2026 by
wang2yn84
Collaborator
Loading…
6 tasks
Support DeepSWE on MaxText models in Tunix
#2051
opened Aug 31, 2026 by
niting
Contributor
Loading…
6 tasks done
fix: handle JSON-string finish arguments in ToolEnvironment._extract_llm_answer
#2049
opened Aug 31, 2026 by
howl-anderson
Loading…
Carry rollout MoE routing from the vLLM sampler to the maxtext trainer
#2046
opened Aug 31, 2026 by
khatwanimohit
Collaborator
Loading…
6 tasks done
[Tunix] Refactor SequencePackedBatchAssembler to use centralized sequence packing utility.
#2045
opened Aug 31, 2026 by
copybara-service
Bot
Loading…
fix(peft_trainer): make scalar-metric host-fetch multi-process safe
#2041
opened Aug 30, 2026 by
lokic233
Loading…
ProTip!
What’s not been updated in a month: updated:<2026-08-01.