TensorRT-LLM 0.17.0.post1 fails to run Whisper on 5080 GPU

System Info

• GPU 5080
• TensorRT-LLM: 0.17.0.post1
• PyTorch: 2.7.0.dev20250302+cu128
• CUDA: 12.8

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

tensorrt-llm 0.17.0.post1 requires torch<=2.6.0a0,>=2.5.1, but you have torch 2.7.0.dev20250302+cu128 which is incompatible.

Expected behavior

pass

actual behavior

torchaudio 2.6.0.dev20250302+cu128 requires torch==2.7.0.dev20250302, but you have torch 2.5.1 which is incompatible.

additional notes

Description:
When trying to run Whisper on an NVIDIA 5080 GPU using TensorRT-LLM 0.17.0.post1, the process fails. The issue may be related to compatibility with Torch 2.7.0.dev20250302+cu128 or specific support for the 5080 GPU in TensorRT-LLM.

Steps to Reproduce:

1. Install the following environment on a system with 5080 GPU:

• TensorRT-LLM: 0.17.0.post1
• PyTorch: 2.7.0.dev20250302+cu128
• CUDA: 12.8

2. Attempt to accelerate Whisper using TensorRT-LLM.
3. The execution fails with an error message (see below).
   Expected Behavior:
   Whisper should run successfully on the 5080 GPU with TensorRT-LLM acceleration.

Actual Behavior:
The execution fails, producing the following error (please provide the full error log):
tensorrt-llm 0.17.0.post1 requires torch<=2.6.0a0,>=2.5.1, but you have torch 2.7.0.dev20250302+cu128 which is incompatible.

System Information:

• GPU: NVIDIA 5080
• CUDA Version: 12.8
• TensorRT-LLM Version: 0.17.0.post1
• PyTorch Version: 2.7.0.dev20250302+cu128
• Whisper Model Version: (Please specify the exact model version)

Questions & Suggestions:

• Is there any plan to support Torch 2.7.0.dev20250302+cu128?
• Is there a workaround or patch available for 5080 GPU compatibility?

Thank you for your support! Looking forward to any guidance or a potential fix.

Start by reproducing the reported environment with NVIDIA 5080, CUDA 12.8, TensorRT-LLM 0.17.0.post1, PyTorch 2.7.0.dev20250302+cu128, and torchaudio, then inspect the dependency conflict and Whisper support. Done means identifying a supported compatible combination or documenting the specific limitation and workaround; the issue does not name files or tests.