From 432ca8b7af8b8d3a7ec30151ca38308470473292 Mon Sep 17 00:00:00 2001
From: "bryan.jia" <iccccing@gmail.com>
Date: Fri, 28 Feb 2025 16:00:20 +0800
Subject: [PATCH] fix: ollama model provider can not set 'num_ctx' etc.
parameter #2442
langchain-openai is not compatible with parameter Settings in ollama, such as num_ctx. Therefore, you need to create model instances using langchain-ollama
(cherry picked from commit 42ae7b443d7320ace464e9a12abcbb48d7fabc9c)
---
.../impl/ollama_model_provider/model/llm.py | 11 +++++------
pyproject.toml | 1 +
2 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/apps/setting/models_provider/impl/ollama_model_provider/model/llm.py b/apps/setting/models_provider/impl/ollama_model_provider/model/llm.py
index 7c98f7e5cef..6cd291ff3cc 100644
--- a/apps/setting/models_provider/impl/ollama_model_provider/model/llm.py
+++ b/apps/setting/models_provider/impl/ollama_model_provider/model/llm.py
@@ -10,7 +10,8 @@
from urllib.parse import urlparse, ParseResult
from langchain_core.messages import BaseMessage, get_buffer_string
-from langchain_openai.chat_models import ChatOpenAI
+from langchain_ollama.chat_models import ChatOllama
+
from common.config.tokenizer_manage_config import TokenizerManage
from setting.models_provider.base_model_provider import MaxKBBaseModel
@@ -24,7 +25,7 @@ def get_base_url(url: str):
return result_url[:-1] if result_url.endswith("/") else result_url
-class OllamaChatModel(MaxKBBaseModel, ChatOpenAI):
+class OllamaChatModel(MaxKBBaseModel, ChatOllama):
@staticmethod
def is_cache_model():
return False
@@ -33,12 +34,10 @@ def is_cache_model():
def new_instance(model_type, model_name, model_credential: Dict[str, object], **model_kwargs):
api_base = model_credential.get('api_base', '')
base_url = get_base_url(api_base)
- base_url = base_url if base_url.endswith('/v1') else (base_url + '/v1')
optional_params = MaxKBBaseModel.filter_optional_params(model_kwargs)
- return OllamaChatModel(model=model_name, openai_api_base=base_url,
- openai_api_key=model_credential.get('api_key'),
- stream_usage=True, **optional_params)
+ return OllamaChatModel(model=model_name, base_url=base_url,
+ stream=True, **optional_params)
def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
tokenizer = TokenizerManage.get_tokenizer()
diff --git a/pyproject.toml b/pyproject.toml
index a5d175b855b..8134688e758 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -28,6 +28,7 @@ pycryptodome = "^3.19.0"
beautifulsoup4 = "^4.12.2"
html2text = "^2024.2.26"
langchain-openai = "^0.1.8"
+langchain-ollama = "0.1.3"
django-ipware = "^6.0.4"
django-apscheduler = "^0.6.2"
pymupdf = "1.24.9"