uv pip install fails to install llama-cpp-python

[alonsosilvaallende]

Thank you very much for this package. I'm encountering the following issue. In an emptly virtual environment, when I run

uv pip install llama-cpp-python --verbose

I obtain

DEBUG uv 0.2.35
DEBUG Searching for Python interpreter in system path
DEBUG Found `cpython-3.10.12-linux-x86_64-gnu` at `/home/asilva/test-llama-cpp/.venv/bin/python3` (active virtual environment)
DEBUG Using Python 3.10.12 environment at test-llama-cpp/.venv/bin/python3
DEBUG Acquired lock for `test-llama-cpp/.venv`
DEBUG At least one requirement is not satisfied: llama-cpp-python
DEBUG Using request timeout of 30s
DEBUG Solving with installed Python version: 3.10.12
DEBUG Adding direct dependency: llama-cpp-python*
DEBUG Found fresh response for: https://pypi.org/simple/llama-cpp-python/
DEBUG Searching for a compatible version of llama-cpp-python (*)
DEBUG Selecting: llama-cpp-python==0.2.87 [compatible] (llama_cpp_python-0.2.87.tar.gz)
DEBUG Acquired lock for `/home/asilva/.cache/uv/built-wheels-v3/pypi/llama-cpp-python/0.2.87`
DEBUG No cache entry for: https://files.pythonhosted.org/packages/a2/63/7b9fe4c6b9d52e1ed2c2689711cdbb60314cf13474c626accd4213cf937c/llama_cpp_python-0.2.87.tar.gz
DEBUG Downloading source distribution: llama-cpp-python==0.2.87
error: Failed to download and build `llama-cpp-python==0.2.87`
  Caused by: Failed to extract archive
  Caused by: failed to unpack `/home/asilva/.cache/uv/built-wheels-v3/.tmpvYPWA5/llama_cpp_python-0.2.87/vendor/llama.cpp/spm-headers/ggml-alloc.h`
  Caused by: File exists (os error 17) when symlinking ../ggml/include/ggml-alloc.h to /home/asilva/.cache/uv/built-wheels-v3/.tmpvYPWA5/llama_cpp_python-0.2.87/vendor/llama.cpp/spm-headers/ggml-alloc.h

Current uv version: uv 0.2.35

Current platform:
PRETTY_NAME="Ubuntu 22.04.4 LTS"
NAME="Ubuntu"
VERSION_ID="22.04"
VERSION="22.04.4 LTS (Jammy Jellyfish)"
VERSION_CODENAME=jammy
ID=ubuntu
ID_LIKE=debian
HOME_URL="https://www.ubuntu.com/"
SUPPORT_URL="https://help.ubuntu.com/"
BUG_REPORT_URL="https://bugs.launchpad.net/ubuntu/"
PRIVACY_POLICY_URL="https://www.ubuntu.com/legal/terms-and-policies/privacy-policy"
UBUNTU_CODENAME=jammy

[baggiponte]

Uhm, I cannot decipher the last line:

Caused by: File exists (os error 17) when symlinking ../ggml/include/ggml-alloc.h to /home/asilva/.cache/uv/built-wheels-v3/.tmpvYPWA5/llama_cpp_python-0.2.87/vendor/llama.cpp/spm-headers/ggml-alloc.h

There might be another version of ggml-alloc somewhere in the machine?

[alonsosilvaallende]

Uhm, I cannot decipher the last line:

Caused by: File exists (os error 17) when symlinking ../ggml/include/ggml-alloc.h to /home/asilva/.cache/uv/built-wheels-v3/.tmpvYPWA5/llama_cpp_python-0.2.87/vendor/llama.cpp/spm-headers/ggml-alloc.h

There might be another version of ggml-alloc somewhere in the machine?

Thank you very much for your answer @baggiponte
Any pointers on how to check if that is the case and how to remove it?

[alonsosilvaallende]

I tested other versions of llama-cpp-python. Up to version llama-cpp-python=0.2.85 it works without any problem. The last two versions llama-cpp-python=0.2.86 and llama-cpp-python=0.2.87 do not work with uv anymore.

[charliermarsh]

I can take a look. My guess is we're refusing to overwrite a file with a symlink, or something like that.

[charliermarsh]

I think there might be something wrong with this source distribution. If you untar it, every entry is duplicated: tar -tf llama_cpp_python-0.2.87.tar.gz:

...
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-1-bpe.cpp
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-1-bpe.cpp
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-1-spm.cpp
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-1-spm.cpp
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-random.py
llama_cpp_python-0.2.87/vendor/llama.cpp/tests/test-tokenizer-random.py
...

I don't see the same issue with the prior release. Do you mind reporting this on the library itself?

[charliermarsh]

Nevermind, I commented on abetlen/llama-cpp-python#1670.

[henryiii]

scikit-build-core 0.10.2 should fix this (thanks to @abetlen)!

[alonsosilvaallende]

Great. Thank you very much. Should I close it then or leave it open until fixed in case other people are having the same issue?

[zanieb]

Best to close it so we know we don't need to do anything here. Thanks!

[alonsosilvaallende]

For others looking at this issue, this has been solved in llama-cpp-python==0.2.88