python安装tokenizers包@FreeBSD

tokenizers提供了当今最常用的tokenizers的实现，重点关注性能和多功能性。官网：GitHub - huggingface/tokenizers: 💥 Fast State-of-the-Art Tokenizers optimized for Research and Production

在FreeBSD系统python3.10下直接pip安装tokenizers报错，

pip install tokenizers
Looking in indexes: https://mirror.baidu.com/pypi/simple
Collecting tokenizers
  Using cached https://mirror.baidu.com/pypi/packages/59/9a/7ba038f101d74fea0861c8f82e188441fe99b2a26fb0991da35f2850f9f3/tokenizers-0.19.0.tar.gz (320 kB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... -


error
  error: subprocess-exited-with-error
  
  × Preparing metadata (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [93 lines of output]
          Updating crates.io index
       Downloading crates ...
        Downloaded darling_macro v0.20.8
        Downloaded windows-targets v0.52.5
        Downloaded derive_builder_macro v0.20.0
        Downloaded monostate-impl v0.1.12
        Downloaded pyo3-macros v0.21.2
        Downloaded fastrand v2.0.2
        Downloaded itoa v1.0.11
        Downloaded macro_rules_attribute-proc_macro v0.2.0
        Downloaded memoffset v0.9.1
        Downloaded rayon-cond v0.3.0
        Downloaded autocfg v1.2.0
        Downloaded env_filter v0.1.0
        Downloaded macro_rules_attribute v0.2.0
        Downloaded indoc v2.0.5
        Downloaded monostate v0.1.12
        Downloaded thiserror-impl v1.0.58
        Downloaded anstyle v1.0.6
        Downloaded either v1.11.0
        Downloaded num-integer v0.1.46
        Downloaded thiserror v1.0.58
        Downloaded pkg-config v0.3.30
        Downloaded darling v0.20.8
        Downloaded derive_builder_core v0.20.0
        Downloaded env_logger v0.11.3
        Downloaded pyo3-build-config v0.21.2
        Downloaded num-complex v0.4.5
        Downloaded anstream v0.6.13
        Downloaded quote v1.0.36
        Downloaded getrandom v0.2.14
        Downloaded derive_builder v0.20.0
        Downloaded tempfile v3.10.1
        Downloaded bitflags v2.5.0
        Downloaded proc-macro2 v1.0.81
        Downloaded ryu v1.0.17
        Downloaded serde_derive v1.0.198
        Downloaded log v0.4.21
        Downloaded num-traits v0.2.18
        Downloaded pyo3-macros-backend v0.21.2
        Downloaded matrixmultiply v0.3.8
        Downloaded darling_core v0.20.8
        Downloaded indicatif v0.17.8
        Downloaded pyo3-ffi v0.21.2
        Downloaded serde v1.0.198
        Downloaded cc v1.0.94
        Downloaded numpy v0.21.0
        Downloaded memchr v2.7.2
        Downloaded unicode-segmentation v1.11.0
        Downloaded aho-corasick v1.1.3
        Downloaded serde_json v1.0.116
        Downloaded esaxx-rs v0.1.10
        Downloaded rayon v1.10.0
        Downloaded regex v1.10.4
        Downloaded syn v2.0.59
      warning: spurious network error (3 tries remaining): [28] Timeout was reached (download of `regex-automata v0.4.6` failed to transfer more than 10 bytes in 30s)

于是改为手动安装

下载源代码

git clone https://github.com/huggingface/tokenizers

进入目录并编译安装

cd tokenizers/bindings/python

pip install -e .

一般这样编译安装之后，后面即使碰到pip安装不同版本的tokenizers，也会顺利通过。