Pull requests: google/tunix
Pull requests list
[Do not Submit] Enable GSM8K Qwen3-0.6B training in ci_smoke_gsm8k_qwen3_0p6b_maxtext.sh and patch cl/983986955
#2378
opened Sep 22, 2026 by
copybara-service
Bot
Loading…
feat: off-policy with batch-ordered trajectory consumption.
#2377
opened Sep 22, 2026 by
jiangyangmu
Collaborator
Loading…
feat(recipes): add frozenlake_35b_256 recipe and k8s launcher for frozenlake_dist
#2372
opened Sep 22, 2026 by
andytwigg
Collaborator
Loading…
feat(recipes): add gsm_8k_35b_256 recipe and enhance math_gsm8k_dist launcher
#2371
opened Sep 22, 2026 by
andytwigg
Collaborator
Loading…
Parse trajectory_store_config in orchestrator and propagate to RolloutWorkers
#2367
opened Sep 22, 2026 by
fineguy
Contributor
Loading…
fix(checkpoint,raiden): prevent 397B every-step checkpointing + Raiden weight-sync OOM
#2365
opened Sep 22, 2026 by
YixuanWang-99
Collaborator
Loading…
Wire TrajectoryStore CLI flags across distributed RL recipes and verify rollout parity.
#2363
opened Sep 22, 2026 by
copybara-service
Bot
Loading…
move weighted-metric reducer to common utils.
#2362
opened Sep 22, 2026 by
copybara-service
Bot
Loading…
Enable Qwen3.5-397B on the GSM8K and DeepSWE paths
#2356
opened Sep 21, 2026 by
khatwanimohit
Collaborator
Loading…
6 tasks
refactor: introduce a new get_group_batch to cover group granular batch use case.
#2355
opened Sep 21, 2026 by
copybara-service
Bot
Loading…
Reject an undersized packing budget in build_maxtext_config
#2351
opened Sep 21, 2026 by
A9isha
Collaborator
Loading…
[Tunix] Update Gemma4 vLLM JAX mappings for tpu-inference >= 0.28 and add E2E test.
#2350
opened Sep 21, 2026 by
copybara-service
Bot
Loading…
Share MaxText configs across all launcher scripts
#2349
opened Sep 21, 2026 by
SurbhiJainUSC
Collaborator
Loading…
Fix kv_einsum indexing and dynamic FSDP dummy input batch sizing in Tunix Gemma 4.
ready-to-submit
#2341
opened Sep 21, 2026 by
copybara-service
Bot
Loading…
Emit vLLM engine stats from the RL rollout sampler
#2336
opened Sep 21, 2026 by
wenxindongwork
Collaborator
Loading…
Correct the TPU and GPU install steps in the launching guide
#2335
opened Sep 21, 2026 by
dkling-it
Loading…
6 tasks done
fix(maxtext_utils): robustly normalize and unquote trainable_parameters_mask
#2325
opened Sep 20, 2026 by
niting
Collaborator
Loading…
Support Ray multi-host rollout backend
#2322
opened Sep 20, 2026 by
knightwh
Collaborator
Loading…
6 tasks
feat(mlperf): add MLPerf RCP logging support for distributed DeepSWE GRPO
#2317
opened Sep 20, 2026 by
susanbao
Collaborator
Loading…
[Tunix] Support DeepSWE evaluation pipeline with MaxText and vLLM on TPU
#2307
opened Sep 19, 2026 by
susanbao
Collaborator
Loading…
6 tasks done
Fix concurrency issue in InprocessVllmSamplerAdapter and VllmSampler.
ready-to-submit
#2304
opened Sep 19, 2026 by
copybara-service
Bot
Loading…
ProTip!
Type g p on any issue or pull request to go back to the pull request listing page.