=> ERROR [xinference 5/8] RUN pip install flash-attn

[flynow0103]

I hope you can support resolve this issue or let me know if there is any environment setting mistake from my computer end ?

D:\Xinference>docker compose up -d
[+] Running 1/1
! xinference Warning pull access denied for xinference, repository does not exist or may r... 3.8s
[+] Building 4.8s (9/12) docker:desktop-linux
=> [xinference internal] load build definition from Dockerfile 0.0s
=> => transferring dockerfile: 284B 0.0s
=> [xinference internal] load metadata for docker.io/xprobe/xinference:latest 0.0s
=> [xinference internal] load .dockerignore 0.0s
=> => transferring context: 2B 0.0s
=> [xinference 1/8] FROM docker.io/xprobe/xinference:latest 0.0s
=> [xinference internal] load build context 0.0s
=> => transferring context: 2.07kB 0.0s
=> CACHED [xinference 2/8] WORKDIR /opt/conda/lib/python3.10/ 0.0s
=> CACHED [xinference 3/8] RUN pip install huggingface_hub 0.0s
=> CACHED [xinference 4/8] RUN pip install gradio==3.50.2 0.0s
=> ERROR [xinference 5/8] RUN pip install flash-attn 4.6s

[xinference 5/8] RUN pip install flash-attn:
1.157 Collecting flash-attn
1.429 Downloading flash_attn-2.6.3.tar.gz (2.6 MB)
2.161 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.6/2.6 MB 4.2 MB/s eta 0:00:00
2.561 Preparing metadata (setup.py): started
4.553 Preparing metadata (setup.py): finished with status 'error'
4.557 error: subprocess-exited-with-error
4.557
4.557 × python setup.py egg_info did not run successfully.
4.557 │ exit code: 1
4.557 ╰─> [23 lines of output]
4.557 No CUDA runtime is found, using CUDA_HOME='/usr/local/cuda'
4.557 fatal: not a git repository (or any of the parent directories): .git
4.557 Traceback (most recent call last):
4.557 File "", line 2, in
4.557 File "", line 34, in
4.557 File "/tmp/pip-install-znz9avtq/flash-attn_00b363935c7343058f80451bbf754322/setup.py", line 158, in
4.557 _, bare_metal_version = get_cuda_bare_metal_version(CUDA_HOME)
4.557 File "/tmp/pip-install-znz9avtq/flash-attn_00b363935c7343058f80451bbf754322/setup.py", line 82, in get_cuda_bare_metal_version
4.557 raw_output = subprocess.check_output([cuda_dir + "/bin/nvcc", "-V"], universal_newlines=True)
4.557 File "/usr/lib/python3.10/subprocess.py", line 421, in check_output
4.557 return run(*popenargs, stdout=PIPE, timeout=timeout, check=True,
4.557 File "/usr/lib/python3.10/subprocess.py", line 503, in run
4.557 with Popen(*popenargs, **kwargs) as process:
4.557 File "/usr/lib/python3.10/subprocess.py", line 971, in init
4.557 self._execute_child(args, executable, preexec_fn, close_fds,
4.557 File "/usr/lib/python3.10/subprocess.py", line 1863, in _execute_child
4.557 raise child_exception_type(errno_num, err_msg, err_filename)
4.557 FileNotFoundError: [Errno 2] No such file or directory: '/usr/local/cuda/bin/nvcc'
4.557
4.557
4.557 torch.version = 2.3.1+cu121
4.557
4.557
4.557 [end of output]
4.557
4.557 note: This error originates from a subprocess, and is likely not a problem with pip.
4.565 error: metadata-generation-failed
4.565
4.565 × Encountered error while generating package metadata.
4.565 ╰─> See above for output.
4.565
4.565 note: This is an issue with the package mentioned above, not pip.
4.565 hint: See above for details.

---

failed to solve: process "/bin/sh -c pip install flash-attn" did not complete successfully: exit code: 1