Issues

• test_cast_fp4_impl_gpu_mem is flaky on Jetson Thor: mem_get_info reports system-wide memory on integrated GPUs

  NVIDIA/Model-Optimizer#2706 · Filter by author LinCanNerd opened on Oct 8, 2026

• test_cuda_pack_reconstruction_matches_pytorch_at_scale[iq1_m] fails on Jetson Thor: exact ties exceed the blocks // 1000 bound

  NVIDIA/Model-Optimizer#2705 · Filter by author LinCanNerd opened on Oct 8, 2026

• Support activation-only scale search for PyTorch AWQ calibration

  NVIDIA/Model-Optimizer#2674 · Filter by author thuiaaa opened on Oct 6, 2026
  

• Entropy calibration mutates the stored NumPy histogram

  NVIDIA/Model-Optimizer#2654 · Filter by author sylvesterkaczmarek opened on Oct 4, 2026
  

• warnings.warn(f"RealQuantLinear: No real-quant GEMM found: {self}.")

  bug
  NVIDIA/Model-Optimizer#2593 · Filter by author zewenli98 opened on Sep 29, 2026
  3

  

• [Bug] qdq_to_dq cannot convert per-tensor quantized weights (scalar scale raises IndexError, (1,) scale raises shape mismatch)

  NVIDIA/Model-Optimizer#2561 · Filter by author SID-6921 opened on Sep 27, 2026
  3

  

• [Bug] modelopt-mcp: all launcher subprocesses hang when the server runs as a stdio MCP server — missing stdin=DEVNULL

  NVIDIA/Model-Optimizer#2551 · Filter by author lucky20260806 opened on Sep 26, 2026
  1

  

• [Puzzletron] KV-cache memory computes head size as n_embd // n_head instead of reading head_dim

  bug
  NVIDIA/Model-Optimizer#2482 · Filter by author soksof opened on Sep 20, 2026
  1

  

• Your project is on StackMap — a curated map of the AI stack

  NVIDIA/Model-Optimizer#2481 · Filter by author hoghweed opened on Sep 19, 2026

• calling mtq.quantize causes AttributeError: 'tuple' object has no attribute 'dim'

  question
  NVIDIA/Model-Optimizer#2424 · Filter by author ccarmatic opened on Sep 13, 2026
  2

  

• Only 1 sample from AutoCast calibration data set is truly used

  bug
  NVIDIA/Model-Optimizer#2337 · Filter by author lashgar opened on Sep 4, 2026
  2

  

• After applying NVFP4 quantization followed by compression, RealQuantLinear fails to locate the corresponding GEMM implementation during inference.

  bug
  NVIDIA/Model-Optimizer#2330 · Filter by author Lee-YNU opened on Sep 4, 2026
  4

  

• ONNX PTQ: no NVFP4 path for convolutional models

  NVIDIA/Model-Optimizer#2281 · Filter by author geoffrey-delhomme opened on Aug 28, 2026
  1

  

• hf_ptq.py: deprecated --auto_quantize_bits CLI flag silently no-ops instead of enabling AutoQuantize

  NVIDIA/Model-Optimizer#2230 · Filter by author wyattearp opened on Aug 22, 2026
  1

  

• [Feature Request] DeepSeek-V4-Flash-0731 NVFP4 checkpoint

  feature request
  NVIDIA/Model-Optimizer#2220 · Filter by author nicole-lihui opened on Aug 20, 2026
  2

• LUT-B Support

  question
  NVIDIA/Model-Optimizer#2204 · Filter by author brian-dellabetta opened on Aug 17, 2026
  1

  

• [Bug] init_quantized_weights / --low_memory_mode exports numerically broken NVFP4 (quantizes meta tensors before weights load)

  NVIDIA/Model-Optimizer#2160 · Filter by author spped2000 opened on Aug 12, 2026
  1

  

• Can auto_quant calculate the score for kv-cache separately?

  question
  NVIDIA/Model-Optimizer#2158 · Filter by author zcfh opened on Aug 12, 2026
  1

  

• Support MiniMax-H3 Visual VAE quantization in ModelOpt

  feature request
  NVIDIA/Model-Optimizer#2137 · Filter by author baonudesifeizhai opened on Aug 10, 2026
  

• # [ONNX][Autotune] Integrated quantization does not preserve AutoTune Q/DQ placement on ViT

  bug
  NVIDIA/Model-Optimizer#2123 · Filter by author HannisLee opened on Aug 10, 2026
  1

  

• [ONNX PTQ] Support for third-party custom ORT/TRT plugins in static calibration

  feature request
  NVIDIA/Model-Optimizer#2016 · Filter by author e-said opened on Jul 24, 2026
  8

  

• Recommended FP8 recipe for Blackwell (B200)? Per-tensor FP8_DEFAULT_CFG underperforms BF16 at prefill; NVFP4 much faster. Also: exported FP8 checkpoint has no KV q/k/v_scale

  bug
  NVIDIA/Model-Optimizer#2015 · Filter by author Francesco-Carlucci opened on Jul 24, 2026
  4

  

• NVFP4 support for Qwen3_5MoeExperts (fused MoE): quantizer registration + export serializer

  NVIDIA/Model-Optimizer#2011 · Filter by author frischeDaten opened on Jul 23, 2026
  10

  

• [QAD] Request for exact training configurations used in arXiv:2601.20088

  NVIDIA/Model-Optimizer#2002 · Filter by author Sekri0 opened on Jul 21, 2026
  5

  

• Public preset quant_cfg changed from dict to list in 0.44 (breaking, undocumented?)

  NVIDIA/Model-Optimizer#2001 · Filter by author sayakpaul opened on Jul 21, 2026
  8