Repository Issue Activity (beta)

#188 Training collapsion

#187 infer.py: get_query() on full sequence may match <search> block from previous turns

#185 请问如果想用您的语料库和索引完全的复现的话，需要一台多少内存、显存的机器？以及推荐的配置

#186 LLDS: A solution to stablize the search R1 training

#157 Retrieval Service Fails with Faiss CUBLAS Error During PPO Training：CUBLAS_STATUS_EXECUTION_FAILED

#167 Why is {do_search} False in the last turn

#177 About benchmark

#171 Test dataset

#182 请问 sm_120框架的 blackwell gpu可以使用吗

#97 Abnormal Reward Curve When Reproducing train_ppo.sh

#173 Support for "Qwen3" architecture and vLLM versions >= 0.11.0

#175 训练稳定性与信用分配：针对长检索序列实现“跳过检索内容”的 GAE 机制

#43 An error while running the infer.py

#67 Getting error while running inference

#129 verl/third_party/vllm modifications and support for newer vLLM versions

#88 Question about GRPO and sample indexing

#50 retrieval_server load wiki-18.jsonl error

#166 Urgent, training encountered an error: batch size must be positive

#152 The training time

#127 OBSERVATION TOO LONG, CONSIDER CHANGING YOUR CONFIG

#170 Quick question: Has anyone solved “NCCL communicator was aborted on rank 0” error? Need some help here, thanks!

#151 Why not use </search> for early stopping during rollout?

#169 Inference using the VLLM-based OpenAI interface fails to align with the output of the offline interface.

#98 Qwen3 support

#162 Response become garbled in later stages of GRPO training