macOS arm64 wheels for 0.3.35 / 0.3.36 fail their own CRC check (Bad CRC-32), so pip can't install them

Aberta
#2,402 0 comentários 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
3/5
Tempo estimado
1-2 dias
Facilidade para iniciantes
58/100
Tipo de issue
Bug
Clareza
Claramente especificada
Status de atividade
Ativa
Stack de tecnologia
python
Domínio
build-system

Direção de pesquisa

Start by locating the macOS arm64 wheel build and packaging steps for releases 0.3.35 and 0.3.36; the issue does not name specific files. Reproduce the failure by running Python's zipfile CRC check on a downloaded wheel, then trace whether any build step modifies files after the wheel is packed. Done means the affected wheels pass the CRC check and can be installed with pip.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

The prebuilt macOS arm64 wheels attached to recent releases fail their own zip CRC check, so pip install stops with zipfile.BadZipFile: Bad CRC-32. The downloaded files match the SHA-256 digests GitHub publishes for the release assets, so they appear to be corrupt as uploaded, not damaged in transit.

Release asset Size GitHub digest (sha256) First bad member
v0.3.35-metal / llama_cpp_python-0.3.35-py3-none-macosx_11_0_arm64.whl 18,164,825 faf9a2b38df5cbf9427a074f974c00abbb799185753c474672fd461c198b7a0a lib/cmake/ggml/ggml-config.cmake
v0.3.35 / llama_cpp_python-0.3.35-py3-none-macosx_11_0_arm64.whl 18,164,782 f16df6e77c2b9ced0ebe7b50de33a125b4d0463f758edf7b516e2b1296fbbcf5 lib/libggml-base.0.20.0.dylib
v0.3.36-metal / llama_cpp_python-0.3.36-py3-none-macosx_11_0_arm64.whl 24,928,420 — lib/cmake/ggml/ggml-config.cmake

For comparison, the v0.3.35 win_amd64 and both manylinux2014 wheels (x86_64, aarch64) pass the same check.

Reproduce (macOS 26, Apple M3, Python 3.13):

curl -sSLO https://github.com/abetlen/llama-cpp-python/releases/download/v0.3.35-metal/llama_cpp_python-0.3.35-py3-none-macosx_11_0_arm64.whl
shasum -a 256 llama_cpp_python-0.3.35-py3-none-macosx_11_0_arm64.whl   # matches GitHub's digest
python3 -c "import zipfile; print(zipfile.ZipFile('llama_cpp_python-0.3.35-py3-none-macosx_11_0_arm64.whl').testzip())"
# zipfile.BadZipFile: Bad CRC-32 for file 'lib/cmake/ggml/ggml-config.cmake'

and through pip:

pip install --no-cache-dir llama-cpp-python==0.3.35 --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/metal
# ... zipfile.BadZipFile: Bad CRC-32 for file 'lib/cmake/ggml/ggml-config.cmake'

Building from the sdist works (pip install --no-binary llama-cpp-python llama-cpp-python==0.3.35, about a minute on an M3, Metal enabled). So the workaround is easy for anyone with the Xcode command line tools, but it means the prebuilt Metal wheels can't be used at the moment.

Possibly the macOS build step changes a file after it has been written into the wheel (for example signing or install_name rewriting a dylib, or a cmake config rewritten after packing), so the stored CRCs no longer match. Re-packing the wheel after the last modification would fix it.

Thanks for maintaining these wheels.

Linguagem predominante
Python
Estrelas
10.7k
Forks
1.5k
Merge médio
1h 40min
PRs com merge (30d)
5

Preparar o ambiente

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de abetlen/llama-cpp-python

Todas as issues de abetlen/llama-cpp-python

Issues semelhantes

Mais issues de Python

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.