Issues

• Converting TranslateGemma to Q4_K takes an incredibly long time.

  plan_nim
  foldl/chatllm.cpp#129 · Filter by author ssechi2 opened on May 18, 2026
  16

• Gemme-4: support use_bidirectional_attention==vision

  bug
  foldl/chatllm.cpp#128 · Filter by author foldl opened on May 15, 2026

• When generating a voice, Qwen3-TTS often adds a strange "word" or "words" to the end.

  foldl/chatllm.cpp#123 · Filter by author ssechi2 opened on Apr 3, 2026
  1

  

• Support proposal : Nvidia's Music Flamingo (AudioFlamingo3ForConditionalGeneration)

  TODO
  foldl/chatllm.cpp#118 · Filter by author dorpxam opened on Mar 16, 2026
  

• dots ocr, result of some pages are wrong, some are correct.

  foldl/chatllm.cpp#105 · Filter by author nissansz opened on Jan 24, 2026
  31

• fix: std::unique_ptr<BaseTensorPosHelper>

  TODO
  foldl/chatllm.cpp#101 · Filter by author foldl opened on Jan 13, 2026

• vision performance of GLM-4.6V-Flash is poor when using native resolution

  bug
  foldl/chatllm.cpp#96 · Filter by author foldl opened on Dec 24, 2025

• share tensors within the same backend device to save memory

  enhancement
  foldl/chatllm.cpp#88 · Filter by author foldl opened on Oct 28, 2025

• Janus: quality of generated images is poor with GPU

  bugcudavukan
  foldl/chatllm.cpp#82 · Filter by author foldl opened on Oct 11, 2025

• sparsemixer function in Phi-3.5-MoE-insturct

  bugenhancement
  foldl/chatllm.cpp#33 · Filter by author JustQJ opened on Aug 29, 2024
  1