Pull requests: google/tunix
Pull requests list
Make weight synchronization optional in distributed GRPO
#2024
opened Aug 28, 2026 by
copybara-service
Bot
Loading…
Upgrade the distributed GSM8K GRPO example to support real datasets and robust math rewards
#2023
opened Aug 28, 2026 by
copybara-service
Bot
Loading…
Add GSM8KEnv and GSM8KAgent for Tunix agentic RL
#2022
opened Aug 28, 2026 by
andytwigg
Collaborator
Loading…
Modify Qwen3 build file to include vllm jax mappings.
#2020
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Modify Llama3 build file to include vllm jax mappings.
#2019
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Modify
bfloat16 handling in safetensors_loader.py to properly convert Torch bfloat16 views into numpy/JAX arrays.
#2016
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Add MaxText and Raiden optional dependencies to Dockerfile
#2013
opened Aug 27, 2026 by
SurbhiJainUSC
Collaborator
Loading…
6 tasks done
feat(rollout): add VllmSamplerAdapter with Raiden weight sync support
#2012
opened Aug 27, 2026 by
SurbhiJainUSC
Collaborator
Loading…
6 tasks done
Add Pyrefly type ignore in FrozenLake data.
#2011
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Support
base_agentic_config.yaml relative path in CLI config loader.
#2008
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Fix Gemma 4 KV cache initialization and add params dependency.
#2007
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Add FrozenLake and PPO GSM8k configs to
xm_launch data dependencies.
#2006
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
Use
int type for num_input_tokens in SamplingState.
#2005
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
[Tunix] Replace legacy TrainExample with RLTrainerPayload in experimental orchestrator and workers.
#2000
opened Aug 27, 2026 by
copybara-service
Bot
Loading…
feat: add YAML template for multihost vllm sampler.
#1995
opened Aug 26, 2026 by
copybara-service
Bot
Loading…
[Tunix] Shift checkpoint restoration control from the trainer to the RL orchestrator.
#1994
opened Aug 26, 2026 by
copybara-service
Bot
Loading…
[Tunix] Refactor VanillaRollout update_params into Sampler.
#1992
opened Aug 26, 2026 by
copybara-service
Bot
Loading…
feat(examples): integrate MaxText trainer backend, multi-rollout, and batched RPA into distributed GSM8K example
#1991
opened Aug 26, 2026 by
SurbhiJainUSC
Collaborator
Loading…
6 tasks done
feat(weight_sync): Raiden weight synchronization with progressive memory reclamation
#1990
opened Aug 26, 2026 by
SurbhiJainUSC
Collaborator
Loading…
6 tasks done
- Merge trajectory lineage contexts into hardware-sized microbatch payloads using LineageContext.merge().
#1986
opened Aug 26, 2026 by
copybara-service
Bot
Loading…
Rollout Ingestion, Auto-Stamping & Worker Telemetry for Distributed RL Lineage.
#1985
opened Aug 26, 2026 by
copybara-service
Bot
Loading…
[Tunix] Inject RaidenWeightSyncDelegate into Vanilla and InprocessVllm Sampler Adapters.
#1975
opened Aug 25, 2026 by
copybara-service
Bot
Loading…
fix(scripts): fix checkpoint sharding restore, precompile CLI flag, and rollout sync in grpo demo
#1962
opened Aug 23, 2026 by
Adityakk9031
Loading…
ProTip!
Filter pull requests by the default branch with base:main.