安装zero_shot_re遇numpy编译错误，寻求解决方案

运行以下关系抽取代码时，出现ModuleNotFoundError: No module named 'zero_shot_re'错误：

from transformers import AutoTokenizer
from zero_shot_re import RelTaggerModel, RelationExtractor

model = RelTaggerModel.from_pretrained("fractalego/fewrel-zero-shot")
tokenizer = AutoTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
relations = ['associated', 'interacts']
extractor = RelationExtractor(model, tokenizer, relations)

尝试执行pip install zero-shot-re安装模块时，遇到numpy==1.19.1编译失败，核心错误日志如下：

(biomedical-knowledge-graph-venv) C:\Users\User\Documents\GitHub\biomedical-knowledge-graph-nlp>pip install zero-shot-re
Collecting zero-shot-re
  Using cached zero_shot_re-0.0.4-py3-none-any.whl (6.6 kB)
Collecting numpy==1.19.1 (from zero-shot-re)
  Using cached numpy-1.19.1.zip (7.3 MB)
  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
  ╰─> [关键输出片段]
      Error compiling Cython file:
      ------------------------------------------------------------
      ...
              self._bitgen.next_uint64 = &mt19937_uint64
                                         ^
      ------------------------------------------------------------

      _mt19937.pyx:138:35: Cannot assign type 'uint64_t (*)(void *) except? -1 nogil' to 'uint64_t (*)(void *) noexcept nogil'. Exception values are incompatible. Suggest adding 'noexcept' to type 'uint64_t (void *) except? -1 nogil'.
      ...
      RuntimeError: Running cythonize failed!
      [end of output]

error: metadata-generation-failed

已尝试以下操作但问题仍未解决：

• pip install --upgrade pip
• pip install --upgrade cython
• pip install --upgrade setuptools wheel
• pip install --no-cache-dir -v numpy==1.19.1

方案1：绕过numpy版本限制安装 #

numpy==1.19.1与Python3.10+兼容性较差，先安装适配当前Python的numpy版本，再忽略依赖安装zero-shot-re，最后补全必要依赖：

# 安装兼容当前Python的numpy版本
pip install numpy>=1.21
# 忽略依赖安装zero-shot-re
pip install zero-shot-re --no-deps
# 手动补全所需核心依赖
pip install transformers torch

方案2：从GitHub源码安装最新版 #

PyPI上的zero-shot-re旧版本绑定了过时的numpy依赖，直接从源码安装可避免版本冲突：

pip install git+https://github.com/fractalego/zero-shot-re.git

方案3：降级Python版本 #

若上述方案无效，可将Python降级至3.8或3.9版本（numpy==1.19.1对这两个版本支持更好），之后再执行pip install zero-shot-re即可正常安装。

内容的提问来源于stack exchange，提问作者Le Dinh Tuan