Issues

• llm_bench crashes on Gemma-4 engines due to missing PLE bindings (ple_token_embeds_0)

  bug
  NVIDIA/TensorRT-Edge-LLM#230 · Filter by author satabios opened on Oct 1, 2026
  1

• 0.11.0 regression: spec-base engine for Qwen3.8-27B fails to build at maxInputLen 196608 (Myelin autotune requests 582.5 GiB) and OOM-kills the device at maxInputLen 98304–131072 — 0.10.1 builds the identical config cleanly

  bug
  NVIDIA/TensorRT-Edge-LLM#228 · Filter by author wlarsong opened on Sep 30, 2026
  1

• OpenAI-compatible server cannot build MTP spec-base engine for Qwen3.8-27B: --speculative-config {"method":"mtp"} fails in the Python builder (Myelin tactic assertion)

  bug
  NVIDIA/TensorRT-Edge-LLM#223 · Filter by author wlarsong opened on Sep 27, 2026
  6

• llm_build fails with Myelin codegen error for maxInputLen >= 229376 (and 131072 at default batch size) on Qwen3.8-27B / SM110

  bug
  NVIDIA/TensorRT-Edge-LLM#222 · Filter by author wlarsong opened on Sep 26, 2026
  2

• feat: Add image-text datasets

  NVIDIA/TensorRT-Edge-LLM#221 · Filter by author FABallemand opened on Sep 25, 2026
  1

• Decoding cannot be constrained: no per-step hook in the sampling path, and logit_bias cannot substitute for one

  NVIDIA/TensorRT-Edge-LLM#218 · Filter by author filipemartinsubrobotics opened on Sep 17, 2026
  1

• Direct export/build fails for Qwen3 VL FP8 fine-grained 2D block-FP8 checkpoints

  bug
  NVIDIA/TensorRT-Edge-LLM#217 · Filter by author Sunk011 opened on Sep 17, 2026
  2

• tensorrt-edgellm-build --dense nvfp4-qdq SIGSEGVs in a libnvinfer compiler worker thread on Jetson Thor (CC 11.0) for Cosmos-Reason2-2B; --dense fp16 builds fine from the same NVFP4 checkpoint

  bug
  NVIDIA/TensorRT-Edge-LLM#216 · Filter by author TeramotoRYEC opened on Sep 17, 2026
  10

• llm_inference ignores temperature/top_p/top_k from the input JSON its own help tells you to set them in

  NVIDIA/TensorRT-Edge-LLM#211 · Filter by author filipemartinsubrobotics opened on Sep 12, 2026
  2

• Int4GroupwiseGemmPluginV2 fails silently when ENABLE_CUTE_DSL omits int4_fp16_gemm: "Failed to enqueue status -1" with no diagnostic

  NVIDIA/TensorRT-Edge-LLM#210 · Filter by author filipemartinsubrobotics opened on Sep 11, 2026
  1

• ONNX-less direct builder cannot resolve decoder tensors for multimodal cosmos3_edge, and renaming them yields a numerically wrong engine

  NVIDIA/TensorRT-Edge-LLM#208 · Filter by author filipemartinsubrobotics opened on Sep 11, 2026

• SM103 (B300 / Blackwell Ultra) is missing from five SM allowlists across the build and the attention path

  NVIDIA/TensorRT-Edge-LLM#206 · Filter by author filipemartinsubrobotics opened on Sep 11, 2026

• XQA NVRTC JIT passes no CUDA include path: cuda_fp16.h cannot find vector_types.h on x86 CUDA 12.x

  NVIDIA/TensorRT-Edge-LLM#204 · Filter by author filipemartinsubrobotics opened on Sep 11, 2026

• x86 CuteDSL cubins missing in the repo

  bug
  NVIDIA/TensorRT-Edge-LLM#203 · Filter by author victoroliv2 opened on Sep 10, 2026
  1

• feat: Add base64 image support to llm_inference

  NVIDIA/TensorRT-Edge-LLM#202 · Filter by author FABallemand opened on Sep 10, 2026

• Experimental OpenAI-compatible server fails to start with: ModuleNotFoundError: No module named 'tensorrt'

  NVIDIA/TensorRT-Edge-LLM#201 · Filter by author FABallemand opened on Sep 9, 2026
  1

• Qwen3-TTS CodePredictor (MTP) crashes with "No available kernel for the GQA" on Jetson AGX Orin

  NVIDIA/TensorRT-Edge-LLM#198 · Filter by author WonderW-hub opened on Sep 4, 2026

• Version mismatch while running engine files

  NVIDIA/TensorRT-Edge-LLM#194 · Filter by author abd-gang opened on Aug 25, 2026
  5

• Qwen3.8-27B NVFP4 base engine build fails: Myelin removeEmptyProducersFromSubgraph not implemented for PLUGIN_V3 (Jetson Thor, TRT 10.13.3.9 and 10.14.1.48)

  NVIDIA/TensorRT-Edge-LLM#192 · Filter by author ruiqi14312 opened on Aug 24, 2026
  4

• feat: Add base64 image support to the OpenAI-compatible server

  External PR Request
  NVIDIA/TensorRT-Edge-LLM#191 · Filter by author FABallemand opened on Aug 24, 2026
  1

• feat: add InternVLA-N1-DualVLN vision-language navigation support

  New Model
  NVIDIA/TensorRT-Edge-LLM#190 · Filter by author hungho77 opened on Aug 24, 2026

• feat: quantized runtime input sidecars for Gemma 4 E-models

  External PR Request
  NVIDIA/TensorRT-Edge-LLM#188 · Filter by author mihai-chiorean opened on Aug 22, 2026
  1

• [Jetson AGX Orin][v0.10.0] Native build may generate XQA kernels for SM80/86/89 instead of SM87

  bug
  NVIDIA/TensorRT-Edge-LLM#187 · Filter by author CaesarYP opened on Aug 21, 2026
  1

• Support multimodal calibration when quantizing only the LLM backbone

  NVIDIA/TensorRT-Edge-LLM#186 · Filter by author twjww opened on Aug 20, 2026
  1

• The discrepancy between the cosine similarity of the intermediate vectors in the VIT engine of Qwen3-8B-VL and the PyTorch-based visual part leads to a decrease in accuracy.

  bug
  NVIDIA/TensorRT-Edge-LLM#185 · Filter by author xiaohaiyo opened on Aug 19, 2026