15308555518/ MaxKB
1
0
Fork 0
mirror of https://github.com/1Panel-dev/MaxKB.git synced 2025-12-25 17:22:55 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix: ollama model provider can not set 'num_ctx' etc. parameter #2442 (#2444) 

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 42ae7b443d)
This commit is contained in:
itaa 2025-02-28 17:45:48 +08:00 committed by GitHub
parent a64adc2504
commit ed8f8f8a3e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
apps/setting/models_provider/impl/ollama_model_provider/model/llm.py
View File

@ -10,7 +10,8 @@ from typing import List, Dict
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 @@ class OllamaChatModel(MaxKBBaseModel, ChatOpenAI):
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()

1
pyproject.toml
View File

@ -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"