Issues

• where is your seekworld_test.parquet and data_thinklite_reasoning_acc.parquet

  Visual-Agent/DeepEyes#146 · Filter by author Smiling-Weeping-zhr opened on Jul 7, 2026

• no multi-turn results

  Visual-Agent/DeepEyes#143 · Filter by author mosAI25 opened on Apr 14, 2026

• Could you please provide an inference script along with the minimum GPU requirements for 32B, 7B, and 3B models respectively?

  Visual-Agent/DeepEyes#142 · Filter by author chfyfr opened on Apr 7, 2026

• Question: bbox coordinate space in eval scripts

  Visual-Agent/DeepEyes#141 · Filter by author Blackteaxx opened on Apr 2, 2026
  1

• ReasonSeg 使用的是什么evaluation metric？Box or Mask？

  Visual-Agent/DeepEyes#140 · Filter by author bryanzhou008 opened on Mar 27, 2026

• increasing GPU memory and finally OOM

  Visual-Agent/DeepEyes#139 · Filter by author whq-pku1 opened on Mar 20, 2026

• Issue starting vLLM serving and training for Qwen-2.5-72B

  Visual-Agent/DeepEyes#138 · Filter by author Amark-cheey opened on Feb 27, 2026

• 请问下huggingface的模型怎么输出思维链呢？

  Visual-Agent/DeepEyes#136 · Filter by author Tristan-Liu-13 opened on Jan 21, 2026
  2

• RL数据

  Visual-Agent/DeepEyes#135 · Filter by author wuzhaodongaipython opened on Dec 15, 2025

• Release artifacts (DeepEyesV2 model, datasets) on Hugging Face

  Visual-Agent/DeepEyes#132 · Filter by author NielsRogge opened on Nov 11, 2025

• Using DeepEyes checkpoint fine-tuning, many <im_start> are repeatedly output when rollout.

  Visual-Agent/DeepEyes#131 · Filter by author ckx506772099 opened on Nov 3, 2025
  2

• how to deploy to vllm and the deepspeed version unknown

  Visual-Agent/DeepEyes#130 · Filter by author Cyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy opened on Oct 31, 2025

• can you provide your checkpoint for evaluation and show demo?

  Visual-Agent/DeepEyes#129 · Filter by author Cyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy opened on Oct 29, 2025

• Evaluation on HR-Bench

  Visual-Agent/DeepEyes#127 · Filter by author XLearning-SCU opened on Oct 12, 2025
  1

• Tool call can't stop after generating </tool_call> during training

  Visual-Agent/DeepEyes#126 · Filter by author pspdada opened on Oct 8, 2025

• Some questions about the implementation of interleaved reasoning.

  Visual-Agent/DeepEyes#124 · Filter by author andwizard opened on Sep 29, 2025

• How to get the tool_call curve if wandb is offline?

  Visual-Agent/DeepEyes#123 · Filter by author yuyi0312 opened on Sep 19, 2025

• Qwen2.5VL 3B training with lora OOM on H100

  Visual-Agent/DeepEyes#122 · Filter by author Liac-li opened on Sep 11, 2025
  2

• How can I determine when to stop training

  Visual-Agent/DeepEyes#120 · Filter by author ckx506772099 opened on Sep 4, 2025
  1

• ImportError: cannot import name 'Qwen2_5_VLFlashAttention2' from 'transformers.models.qwen2_5_vl.modeling_qwen2_5_vl

  Visual-Agent/DeepEyes#119 · Filter by author yuyi0312 opened on Sep 4, 2025
  3

• Consider add a warning for stopping rollout

  Visual-Agent/DeepEyes#118 · Filter by author agoyang opened on Sep 3, 2025

• 模型推理慢

  Visual-Agent/DeepEyes#117 · Filter by author HZWHH opened on Sep 3, 2025
  4

• 输出思维链中出现一些异常，请问什么导致的？

  Visual-Agent/DeepEyes#116 · Filter by author HZWHH opened on Sep 2, 2025

• ray cpu memory OOM/cpu memory leak

  Visual-Agent/DeepEyes#113 · Filter by author woshitff opened on Aug 25, 2025
  9

• Question about the code of tool.execute

  Visual-Agent/DeepEyes#112 · Filter by author w-zhih opened on Aug 25, 2025
  1