Repository Issue Activity (beta)

#1585 Failed build: Tunix Nightly Regression Tests

#587 Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

#1580 Failed build: Tunix Nightly Regression Tests

#1538 Failed build: Tunix Package Tests

#1543 MM SFT of Gemma4

#1539 [bug] TypeError: Gemma3.__call__() got an unexpected keyword argument 'segment_ids'

#1490 add BEFT as a new SFT/PEFT method

#667 Failed build: Tunix Package Tests

#1205 [BUG] fail to run sglang_jax_lora_test.py with flax==0.12.5

#953 Feature Request: RLOO (REINFORCE Leave-One-Out) Advantage Estimator

#1207 [BUG] remove flax==0.12.4 constraint for sglang

#1399 RLCluster unconditionally imports vllm_rollout even when rollout_engine='vanilla'

#832 This repo is trying to do too much

#1378 Support configurable advantage estimation in AgenticRLLearner (RLOO, DrGRPO)

#1469 Configurable advantage estimator in AgenticGRPOLearner

#1386 [Bug] compute_entropy_from_logits uses softmax(log_probs) instead of exp(log_probs), producing incorrect entropy

#1474 [bug] batch iterator exception when running Qwen example

#1321 Qwen2 KV-cache updates do not cast projections to cache dtype before `dynamic_update_slice`

#1380 How to enable audio and vision modalities in tunix's gemma4?

#1196 Unsafe use of eval() in calculate_reward enables code execution if tasks are untrusted

#1224 Add eval datasets to tunix cli for grpo_main

#1384 Clarification on (external) contribution policy

#560 How to train on multi host tpu pod

#1259 TOCTOU race condition in concurrent safetensors loading bypasses duplicate key detection

#510 VLM support for DPO