Issues

• scannet_4frames评测中部分样本生成重复循环，触发max_new_tokens截断并导致JSON解析失败

  LaVi-Lab/VG-LLM#71 · Filter by author NONO997 opened on Jun 3, 2026

• 能用vllm部署zd11024/vgllm-qa-vggt-8b吗

  LaVi-Lab/VG-LLM#70 · Filter by author lsm2842035890 opened on May 19, 2026

• Orientation convention change

  LaVi-Lab/VG-LLM#69 · Filter by author danqu130 opened on Mar 31, 2026

• 训练日志的记录的loss和grad_norm与使用的GPU数量有关，这是Bug吗？

  LaVi-Lab/VG-LLM#68 · Filter by author 0x4e8 opened on Mar 10, 2026
  9

• About vlm-3r dataset performance

  LaVi-Lab/VG-LLM#67 · Filter by author Chongyu-hub opened on Mar 3, 2026
  1

• Reproduction and lmms-eval Differences in VSI-Bench Results

  LaVi-Lab/VG-LLM#66 · Filter by author TUe-Ray opened on Feb 25, 2026
  1

• AttributeError: group_by_modality_length_auto not found in TrainingArguments

  LaVi-Lab/VG-LLM#62 · Filter by author Wuuu3511 opened on Jan 16, 2026
  1

• Performance reproduce

  LaVi-Lab/VG-LLM#61 · Filter by author ChengJade opened on Dec 30, 2025
  3

• loss issues

  LaVi-Lab/VG-LLM#60 · Filter by author Dlonly904 opened on Dec 27, 2025

• custom data test

  LaVi-Lab/VG-LLM#58 · Filter by author wolf943134497 opened on Dec 24, 2025

• Performance differences on visual grounding

  LaVi-Lab/VG-LLM#57 · Filter by author Chongyu-hub opened on Dec 19, 2025
  2

• Weight with other geometry models

  LaVi-Lab/VG-LLM#54 · Filter by author NIneeeeeem opened on Nov 25, 2025

• Training loss difference

  LaVi-Lab/VG-LLM#53 · Filter by author Germany321 opened on Nov 15, 2025
  2

• Something wrong when installing VG-LLM

  LaVi-Lab/VG-LLM#48 · Filter by author starshine360 opened on Oct 20, 2025
  6

• about underfitting

  LaVi-Lab/VG-LLM#47 · Filter by author lhj-git opened on Oct 14, 2025
  3

• Performance gap when enabling tuning_mm_mlp (src vs. adding VGGT)

  LaVi-Lab/VG-LLM#42 · Filter by author Terry-Xu-666 opened on Sep 18, 2025
  1

• Results on ScanRefer with proposal refinement

  LaVi-Lab/VG-LLM#40 · Filter by author flyyyyer opened on Sep 9, 2025
  1

• How about the result of using combination of 3D scene understanding and spatial reasoning training set

  LaVi-Lab/VG-LLM#35 · Filter by author pear-blossoms opened on Aug 27, 2025