-
Notifications
You must be signed in to change notification settings - Fork 236
Pull requests: google/tunix
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Refactor tunix Gemma3-4b SFT script to use new config structure.
#1059
opened Feb 7, 2026 by
copybara-service
bot
Loading…
Log the computed score in GSM8K reward function
#1058
opened Feb 7, 2026 by
copybara-service
bot
Loading…
Skip softmax and sorting of probabilities when top_p == 1.0 and top_k is None.
#1056
opened Feb 6, 2026 by
copybara-service
bot
Loading…
[Tunix] Refactor DeepScaler training script to support different rollout engines and mesh configurations.
#1048
opened Feb 5, 2026 by
copybara-service
bot
Loading…
[Tiny Feat] add rollout_sglang_jax_log_level in RolloutConfig
#1041
opened Feb 3, 2026 by
aolemila
Loading…
6 tasks
feat: log rollout and train time at micro batch level.
#1038
opened Feb 3, 2026 by
copybara-service
bot
Loading…
[Tunix] Use compat.ModuleDict for Flax nnx.Dict compatibility.
#1033
opened Jan 31, 2026 by
copybara-service
bot
Loading…
Lazily import reward_manager in function_registry.
#1032
opened Jan 30, 2026 by
copybara-service
bot
Loading…
Add support for stop strings in vLLM sampler and rollout.
#1027
opened Jan 30, 2026 by
copybara-service
bot
Loading…
Add
max_context_tokens to trajectory engine.
#1005
opened Jan 27, 2026 by
copybara-service
bot
Loading…
Allow config_id as an alternative model_id to automodel
#1002
opened Jan 27, 2026 by
copybara-service
bot
Loading…
fix(rl): robust integer validation for utils.py (Fixes #953)
#1000
opened Jan 24, 2026 by
abdulwahabahmedkhanyusufzai
Loading…
Add GRPO natural language-to-SQL example with execution-based reward
#997
opened Jan 22, 2026 by
NP2241
Loading…
6 tasks done
remove pip install jax==0.8.1 flax==0.12.0 libtpu==0.0.24
#996
opened Jan 22, 2026 by
aolemila
Loading…
6 tasks
[Tunix] Improve the reshard logging. Current logging only polls on the results for on device finish time. We need to do the same thing for the inputs for accurate logging. This cl splits the total wait time to waiting for input time and output ready time.
#992
opened Jan 21, 2026 by
copybara-service
bot
Loading…
Previous Next
ProTip!
Add no:assignee to see everything that’s not assigned.