Issues

• RewardMathFn gives full reward to an empty \boxed{} answer when the ground truth is empty or an empty \boxed{}, instead of returning unk_error_reward

  rllm-org/rllm#929 · Filter by author shaurya416 opened on Sep 25, 2026

• Modal uploads silently drop files when the remote path contains a space

  rllm-org/rllm#927 · Filter by author zhenga1 opened on Sep 23, 2026

• rllm-model-gateway: make the worker health-probe timeout and failure threshold configurable

  rllm-org/rllm#926 · Filter by author AshishKumar4 opened on Sep 23, 2026

• when will you release the code for the blog of realtime_rl.md

  rllm-org/rllm#923 · Filter by author yanan1116 opened on Sep 14, 2026

• reset_batch() wipes RejectionSamplingState every batch, so "episode" mode never accumulates across batches

  rllm-org/rllm#920 · Filter by author AmirF194 opened on Sep 10, 2026

• Bug: reinforce_plus_plus_baseline silently produces zero advantages when rollout.n = 1

  bug
  rllm-org/rllm#916 · Filter by author yanan1116 opened on Sep 9, 2026
  1

• check_correctness / lcb_check_correctness_v2 leave a zombie process after killing a timed-out test run

  rllm-org/rllm#915 · Filter by author AmirF194 opened on Sep 7, 2026
  1

• R1ToolParser.tool_output_end doesn't match DeepSeek's own chat_template tool-output token

  rllm-org/rllm#914 · Filter by author AmirF194 opened on Sep 6, 2026
  1

• No train/held-out split for claw_eval and skillsbench

  rllm-org/rllm#901 · Filter by author yanan1116 opened on Aug 24, 2026
  3

• async_save=true on the default FSDP strategy never writes the checkpoint tracker file, so resume_mode:auto silently restarts from scratch

  rllm-org/rllm#897 · Filter by author AmirF194 opened on Aug 21, 2026
  2

• OpenAIEngine.completion() retokenization fallback can desync completion_ids from logprobs

  rllm-org/rllm#816 · Filter by author AmirF194 opened on Aug 8, 2026

• rllm train crashes at first validation: 'FixedEvaluatorHooks' object has no attribute 'warm_queue'

  rllm-org/rllm#717 · Filter by author Chanbinski2 opened on Jun 30, 2026
  1

• How to run PPO for agents?

  question
  rllm-org/rllm#704 · Filter by author miaoyuchun opened on Jun 26, 2026
  1

• Example for training on SWE (agentic software-engineering) tasks?

  question
  rllm-org/rllm#625 · Filter by author dipta007 opened on Jun 7, 2026

• AgentWorkflowPPOTrainer: GRPO groups by trajectory.uid instead of prompt/agent key

  question
  rllm-org/rllm#605 · Filter by author TtLuckyyy opened on Jun 2, 2026
  1

• Sandboxed code execution for RL rollouts

  rllm-org/rllm#468 · Filter by author congwang-mk opened on Apr 1, 2026
  2

• Error encountered when using SDK mode

  rllm-org/rllm#465 · Filter by author hailuoS opened on Mar 31, 2026
  1

• Qwen3-VL training with flashattn+vllm

  question
  rllm-org/rllm#464 · Filter by author llexieguo opened on Mar 30, 2026
  1

• Qwen3.5 support

  question
  rllm-org/rllm#434 · Filter by author HaishuoFang opened on Mar 10, 2026
  13

• Bug: Multi-turn verl engine causing abnormal agent behavior

  rllm-org/rllm#331 · Filter by author WxxShirley opened on Dec 1, 2025
  6

• DPO support

  enhancement
  rllm-org/rllm#329 · Filter by author LianShuQuan opened on Nov 29, 2025

• Where can I find the ablation results for the 4k context length version?

  rllm-org/rllm#104 · Filter by author vHhee opened on Apr 21, 2025
  1

• Feasibility of Reproducing DeepCoder-1.5B-Preview 16k and 24k Tasks on a Single A100 8×80GB Machine

  rllm-org/rllm#103 · Filter by author Go4miii opened on Apr 20, 2025
  1

• Can I use the latest vLLM 0.8.x?

  rllm-org/rllm#102 · Filter by author Superskyyy opened on Apr 20, 2025
  3

• Could you provide a Dockerfile?

  rllm-org/rllm#99 · Filter by author xjw-nlp opened on Apr 18, 2025
  1