Source code for langchain_community.embeddings.llamafile

import logging
from typing import List, Optional

import requests
from langchain_core.embeddings import Embeddings
from langchain_core.pydantic_v1 import BaseModel

logger = logging.getLogger(__name__)


class LlamafileEmbeddings(BaseModel, Embeddings):
    """Llamafile可以让您使用单个文件分发和运行大型语言模型。

要开始，请参阅：https://github.com/Mozilla-Ocho/llamafile

要使用这个类，您首先需要：

1. 下载一个llamafile。
2. 使下载的文件可执行：`chmod +x path/to/model.llamafile`
3. 以启用嵌入的服务器模式启动llamafile：

    `./path/to/model.llamafile --server --nobrowser --embedding`

示例：
    .. code-block:: python

        from langchain_community.embeddings import LlamafileEmbeddings
        embedder = LlamafileEmbeddings()
        doc_embeddings = embedder.embed_documents(
            [
                "Alpha is the first letter of the Greek alphabet",
                "Beta is the second letter of the Greek alphabet",
            ]
        )
        query_embedding = embedder.embed_query(
            "What is the second letter of the Greek alphabet"
        )"""

    base_url: str = "http://localhost:8080"
    """LLamafile服务器正在监听的基本URL。"""

    request_timeout: Optional[int] = None
    """服务器请求的超时时间"""

    def _embed(self, text: str) -> List[float]:
        try:
            response = requests.post(
                url=f"{self.base_url}/embedding",
                headers={
                    "Content-Type": "application/json",
                },
                json={
                    "content": text,
                },
                timeout=self.request_timeout,
            )
        except requests.exceptions.ConnectionError:
            raise requests.exceptions.ConnectionError(
                f"Could not connect to Llamafile server. Please make sure "
                f"that a server is running at {self.base_url}."
            )

        # Raise exception if we got a bad (non-200) response status code
        response.raise_for_status()

        contents = response.json()
        if "embedding" not in contents:
            raise KeyError(
                "Unexpected output from /embedding endpoint, output dict "
                "missing 'embedding' key."
            )

        embedding = contents["embedding"]

        # Sanity check the embedding vector:
        # Prior to llamafile v0.6.2, if the server was not started with the
        # `--embedding` option, the embedding endpoint would always return a
        # 0-vector. See issue:
        # https://github.com/Mozilla-Ocho/llamafile/issues/243
        # So here we raise an exception if the vector sums to exactly 0.
        if sum(embedding) == 0.0:
            raise ValueError(
                "Embedding sums to 0, did you start the llamafile server with "
                "the `--embedding` option enabled?"
            )

        return embedding

    def embed_documents(self, texts: List[str]) -> List[List[float]]:
        """使用在`self.base_url`上运行的llamafile服务器嵌入文档。
在调用此方法之前，应该在单独的进程中启动llamafile服务器。

参数：
    texts: 要嵌入的文本列表。

返回：
    每个文本的嵌入列表。
"""
        doc_embeddings = []
        for text in texts:
            doc_embeddings.append(self._embed(text))
        return doc_embeddings

    def embed_query(self, text: str) -> List[float]:
        """在运行在`self.base_url`上的llamafile服务器中嵌入一个查询。
在调用此方法之前，应该在单独的进程中启动llamafile服务器。

参数：
    text: 要嵌入的文本。

返回：
    文本的嵌入。
"""
        return self._embed(text)