[Bug]: AutoDeploy model-coverage hits NVRTC compilation failed across multiple models

System Info

• CPU architecture: x86_64
• GPU properties: not captured from CI artifact
• Libraries
  • TensorRT-LLM branch or tag: main (pipeline context)
  • Container used: AutoDeploy model-coverage CI container
• OS: CI environment
• Additional context: terminal pipeline 52785950

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

Pipeline 52785720 resolved to terminal model-coverage pipeline 52785950.

Affected registry entries currently disabled in examples/auto_deploy/model_registry/models.yaml include:

• zai-org/GLM-5-FP8
• MiniMaxAI/MiniMax-M2.5
• MiniMaxAI/MiniMax-M2.7
• deepseek-ai/DeepSeek-V3.2
• deepseek-ai/DeepSeek-V3.2-Speciale
• MiniMaxAI/MiniMax-M2
• deepseek-ai/DeepSeek-Prover-V2-671B
• deepseek-ai/DeepSeek-V3
• Qwen/Qwen3-0.6B-FP8

Representative evidence from per-test stdout logs ends with:

RuntimeError: NVRTC compilation failed

One concrete example is MiniMaxAI/MiniMax-M2.7, where the traceback tail runs through tensorrt_llm/_torch/auto_deploy/custom_ops/quantization/trtllm_quant.py into torch.ops.trtllm.fp8_block_scaling_gemm(...) before failing with RuntimeError: NVRTC compilation failed.

Expected behavior

These model-coverage entries should either compile and run successfully, or fail with a more specific unsupported-configuration signal if FP8 block-scaling GEMM is not supported in the current path.

actual behavior

TensorRT-LLM fails during AutoDeploy execution with RuntimeError: NVRTC compilation failed across multiple distinct models, suggesting a shared runtime or codegen failure rather than a model-specific problem.

additional notes

This looks like a TensorRT-LLM / AutoDeploy runtime issue, not a Hugging Face access problem. The same failure string appears across nine disabled entries, which makes it a good candidate for one general issue

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

Start with terminal pipeline 52785950 and the disabled entries in examples/auto_deploy/model_registry/models.yaml. Trace the MiniMaxAI/MiniMax-M2.7 failure through tensorrt_llm/_torch/auto_deploy/custom_ops/quantization/trtllm_quant.py and the fp8_block_scaling_gemm operation. Done means the affected entries run successfully or report a specific unsupported-configuration signal instead of the generic NVRTC compilation failure.