Jetson Orin Nano Super無法安裝Whisper

Robotics & Edge Computing Jetson Systems Jetson Orin Nano
1

設備資訊
caesar@ubuntu: $ nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Wed_Aug_14_10:14:07_PDT_2024
Cuda compilation tools, release 12.6, V12.6.68
Build cuda_12.6.r12.6/compiler.34714021_0

caesar@ubuntu: $ cat /etc/nv_tegra_release

# KERNEL_VARIANT: oot

R36 (release), REVISION: 4.4, GCID: 41062509, BOARD: generic, EABI: aarch64, DATE: Mon Jun 16 16:07:13 UTC 2025
TARGET_USERSPACE_LIB_DIR=nvidia
TARGET_USERSPACE_LIB_DIR_PATH=usr/lib/aarch64-linux-gnu/nvidia

nvidia-smi
NVIDIA-SMI 540.4.0
Drive Version:540.4.0
Cuda Version 12.6
Orin(nvgpu)

python3 --version
pip 22.0.2
pip3 --version
Python 3.10.12

=======================================================================
安裝torch
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

檢查
Python3

import torch
print(torch.cuda.is_available())
True
print(torch.cuda.get_device_name(0))
Orin

=======================================================================
pip3 install openai-whisper
pip3 install huggingface_hub

這裡報錯
python3 -c “import whisper; whisper.load_model(‘base’)”

Traceback (most recent call last):
File “”, line 1, in
File “/home/caesar/.local/lib/python3.10/site-packages/whisper/init_.py”, line 155,in load_model
model.load_state_dict(checkpoint[“model_state_dict”])
File “/home/caesar/.local/lib/python3.10/site-packages/torch/nn/modules/module.py”, line in load_state_dict
raise RuntimeError(
RuntimeErrorг: Еггог(s) in loading state_dict for Whisper:
While copying the parameter named “encoder.blocks.0.attn.query.weight”, whose dimens in the model are torch. Size([512, 512]) and whose dimensions in the checkpoint are torch.Siz 2, 512]), an exception occurred: ('CUDA error: no kernel image is available for execution o device\nCUDA kernel errors might be asynchronously reported at some other API call, so the trace below might be incorrect.\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\nComp ith TORCH_USE_CUDA_DSA to enable device-side assertions.\n’,)

嘗試幾個PYTORCH版本都不行，求各位大大指點!

2

Hello,

Thanks for visiting the NVIDIA Developer forums.

Your topic will be best served in the Jetson category, I have moved this post for better visibility.

Cheers,
Tom

5

Hi,

Could you try to build it from the source to see if it can work?
You can follow the steps in the below Dockerfile:

Thanks.

6

Thank you for your reply. I compared the two files and found no discrepancies.

#---
# name: whisper
# group: audio
# depends: [numba, numpy, pytorch, torchaudio, jupyterlab]
# requires: '>=34.1.0'
# docs: docs.md
#---
ARG BASE_IMAGE
FROM ${BASE_IMAGE}

WORKDIR /opt

RUN apt-get update && \
    apt-get install -y --no-install-recommends \
                  ffmpeg \
    && rm -rf /var/lib/apt/lists/* \
    && apt-get clean

RUN pip3 install pandas scipy jiwer ipywebrtc

ADD https://api.github.com/repos/openai/whisper/git/refs/heads/main /tmp/whisper_version.json

# Clone the repository:
RUN git clone https://github.com/openai/whisper/ && \
    cd whisper && \
    sed -i 's/==/>=/g' requirements.txt && \
    sed -i 's/~=/>=/g' requirements.txt && \
    cat requirements.txt && \
    pip3 wheel --wheel-dir=dist --no-deps --verbose .

RUN cp whisper/dist/openai_whisper*.whl /opt && \
    pip3 install /opt/openai_whisper*.whl

WORKDIR /opt/whisper/

COPY record-and-transcribe.ipynb /opt/whisper/notebooks

RUN openssl req \
        -new \
        -newkey rsa:4096 \
        -days 365 \
        -nodes \
        -x509 \
        -keyout mykey.key \
        -out mycert.pem \
        -subj '/CN=localhost'

CMD /bin/bash -c "jupyter lab --ip 0.0.0.0 --port 8888  --certfile=mycert.pem --keyfile mykey.key --allow-root &> /var/log/jupyter.log" & \
        echo "allow 10 sec for JupyterLab to start @ https://$(hostname -I | cut -d' ' -f1):8888 (password nvidia)" && \
        echo "JupterLab logging location:  /var/log/jupyter.log  (inside the container)" && \
        /bin/bash
7

Hi,

Do you build it from the source or install it from pip3?
The prebuilt might not include the Orin GPU architecture so the CUDA function cannot work correctly.

Thanks.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.