Issues

• Myelin exec_with_offset failure at run time with builder_optimization_level=4 for part of a dynamic-batch profile (TensorRT 10.16, H100)

  NVIDIA/TensorRT#4898 · Filter by author pythongiant opened on Oct 27, 2026

• Publish onnx-graphsurgeon 0.6.2

  Module:GraphSurgeon
  NVIDIA/TensorRT#4895 · Filter by author ConnorBaker opened on Oct 26, 2026

• OSS release exports drop pyproject.toml files; Polygraphy can't be built from source since 11.3

  Module:Polygraphy
  NVIDIA/TensorRT#4894 · Filter by author ConnorBaker opened on Oct 26, 2026

• Wrong results of TensorRT 10.14 / 11.4 static-shape engines when a batch-half swap (Slice+Slice+Concat) is fused with two reductions (FP32 too; dynamic shapes correct)

  Module:Accuracy
  NVIDIA/TensorRT#4893 · Filter by author lualxand-hash opened on Oct 24, 2026

• Python 3.15 bindings: OSS BUILD_PYTHON build works, but wheel packaging fails (missing bindings_wheel/pyproject.toml) and 3.15 is not in the version list

  Module:OSS Build
  NVIDIA/TensorRT#4890 · Filter by author KRRT7 opened on Oct 24, 2026

• Polygraphy --load-inputs and inspect data fail with ValueError when a tensor is named Tensor or ndarray

  Module:Polygraphy
  NVIDIA/TensorRT#4889 · Filter by author mottopanikeiku opened on Oct 22, 2026
  4

• Build failure on CUDA < 12.8 due to ungated default CMAKE_CUDA_ARCHITECTURES containing SM 100/120

  Module:OSS Build
  NVIDIA/TensorRT#4887 · Filter by author Deeven-Seru opened on Oct 19, 2026

• buildWtsEngine assertion when an MXFP8 block scale (E8M0) is refittable under REFIT_INDIVIDUAL

  Module:Engine Build
  NVIDIA/TensorRT#4883 · Filter by author gafrom opened on Oct 15, 2026

• Build failure (Myelin layouts.cpp:2180) or wrong plugin input of TensorRT 10.16 when an IPluginV3 with tactics reads a kHWC8 network output, on Jetson Thor

  Module:Engine Build
  NVIDIA/TensorRT#4882 · Filter by author adujardin opened on Oct 15, 2026

• Performance: fused MHA reaches 77-125 TFLOPS for BERT-large attention on RTX PRO 6000 Blackwell, 2-4x below FlashAttention-2 on the same GPU

  Module:Performance
  NVIDIA/TensorRT#4881 · Filter by author gkswns0531 opened on Oct 15, 2026

• Performance: NVFP4 encoder slower than FP8 with TensorRT 11.4 on RTX PRO 6000 Blackwell, from copies and unfused quantize kernels around the FP4 GEMMs

  Module:Performance
  NVIDIA/TensorRT#4880 · Filter by author gkswns0531 opened on Oct 15, 2026

• Wrong NVFP4 results of TensorRT 11.4 at batch > 1 when Q/K/V MatMuls share a TRT_FP4DynamicQuantize input, on RTX PRO 6000 Blackwell

  Module:Accuracy
  NVIDIA/TensorRT#4879 · Filter by author gkswns0531 opened on Oct 15, 2026

• [TensorRT 11.x][Blackwell SM120] Internal DDS compiler assertion when IPluginV3 scalar output controls dynamic ONNX Slice

  Module:Engine Build
  NVIDIA/TensorRT#4878 · Filter by author Axlef opened on Oct 14, 2026

• EmbLayerNormVarSeqlenPlugin: dangling pointer and out-of-bounds read during serialization

  Module:Plugins
  NVIDIA/TensorRT#4876 · Filter by author gabortar opened on Oct 12, 2026

• onnx-graphsurgeon: Graph.toposort() hits RecursionError when a graph has a local function and an If/Loop node

  Module:GraphSurgeon
  NVIDIA/TensorRT#4874 · Filter by author gabortar opened on Oct 12, 2026

• Polygraphy inspect sparsity misreads BF16 initializers: shadowed group index and a half-read int32_data group

  Module:Polygraphy
  NVIDIA/TensorRT#4873 · Filter by author v0ropaev opened on Oct 12, 2026
  2

• Polygraphy cosine comparison changes verdict under rescaling

  Module:Polygraphy
  NVIDIA/TensorRT#4868 · Filter by author sylvesterkaczmarek opened on Oct 4, 2026

• Engine build failure of TensorRT 10.14.1 and 11.3.0 when running S2M2 stereo matching above ~1 MPix input on Jetson AGX Thor (sm_110)

  Module:Engine Build
  NVIDIA/TensorRT#4866 · Filter by author george-auradine opened on Oct 4, 2026
  1

• Polygraphy parse_meta accepts a shape with a missing bracket

  Module:Polygraphy
  NVIDIA/TensorRT#4864 · Filter by author SashaMIT opened on Sep 30, 2026

• ScatterND in a dynamic-shape engine silently drops updates beyond ~N when the update count M exceeds the output size N (static engine correct)

  NVIDIA/TensorRT#4861 · Filter by author braindevices opened on Sep 29, 2026

• TRT 11.3: Attention does not compile inside If, even with decomposable=True

  Module:ONNX
  NVIDIA/TensorRT#4860 · Filter by author vient opened on Sep 28, 2026
  1

• [Bug]: trtllm-serve renders a client-supplied chat_template server-side, a string-blow-up expression OOM-kills whole server (unauthenticated crash DoS)

  NVIDIA/TensorRT#4859 · Filter by author Yunzez opened on Sep 27, 2026

• onnx_graphsurgeon: Constant with float8_e4m3fnuz values is exported as FLOAT8E4M3FN

  Module:GraphSurgeon
  NVIDIA/TensorRT#4856 · Filter by author luicarus opened on Sep 21, 2026

• trtllm-serve accepts response_format and never enforces it

  NVIDIA/TensorRT#4855 · Filter by author Yunzez opened on Sep 17, 2026

• Wrong FP16 results when a fused convolution, activation and residual add reads a strided concat view

  Module:Accuracy
  NVIDIA/TensorRT#4854 · Filter by author Y-T-G opened on Sep 17, 2026
  1