llms.py — langchain Source File

Source Code

"""Ollama large language models."""

from __future__ import annotations

from collections.abc import AsyncIterator, Iterator, Mapping
from typing import Any, Literal

from langchain_core.callbacks import (
    AsyncCallbackManagerForLLMRun,
    CallbackManagerForLLMRun,
)
from langchain_core.language_models import BaseLLM, LangSmithParams
from langchain_core.outputs import GenerationChunk, LLMResult
from ollama import AsyncClient, Client, Options
from pydantic import PrivateAttr, model_validator
from typing_extensions import Self

from langchain_ollama._utils import (
    merge_auth_headers,
    parse_url_with_auth,
    validate_model,
)


class OllamaLLM(BaseLLM):
    """Ollama large language models.

    Setup:
        Install `langchain-ollama` and install/run the Ollama server locally:

        ```bash
        pip install -U langchain-ollama
        # Visit https://ollama.com/download to download and install Ollama
        # (Linux users): start the server with `ollama serve`
        ```

        Download a model to use:

        ```bash
        ollama pull llama3.1
        ```

    Key init args — generation params:
        model: str
            Name of the Ollama model to use (e.g. `'llama4'`).
        temperature: float | None
            Sampling temperature. Higher values make output more creative.
        num_predict: int | None
            Maximum number of tokens to predict.
        top_k: int | None
            Limits the next token selection to the K most probable tokens.
        top_p: float | None
            Nucleus sampling parameter. Higher values lead to more diverse text.
        mirostat: int | None
            Enable Mirostat sampling for controlling perplexity.
        seed: int | None
            Random number seed for generation reproducibility.

    Key init args — client params:
        base_url:
// ... (490 more lines)