f1349ec455
百炼（alibaba provider）的 OpenAI 兼容流经 LangGraph v3 event stream 累积工具调用时会丢失 tool_call.id，空 id 写入 checkpoint 后导致工具结果 无法按 tool_call_id 关联、工具状态永远停留在"进行中"。 实测确认 ainvoke 与 stream_mode=messages 累积均正常，仅 astream_events v3 协议丢失 id；升级 langchain-core 至 1.4.4 仍未修复（属上游问题 langchain#37420 / langchainjs#10937 / langgraphjs#2496）。 与已有的 SiliconFlow MiniMax 修复同源，将 alibaba 加入禁流式名单， 工具调用阶段禁用流式（正文回答仍流式）以拿到完整 tool_call。
74 lines
2.8 KiB
Python
74 lines
2.8 KiB
Python
from langchain.chat_models import BaseChatModel
|
||
from pydantic import SecretStr
|
||
|
||
from yuxi.models.providers.cache import model_cache
|
||
from yuxi.utils import get_docker_safe_url
|
||
from yuxi.utils.logging_config import logger
|
||
|
||
|
||
# 这些提供商的 OpenAI 兼容流式接口，在 LangGraph v3 事件流累积工具调用时会丢失
|
||
# tool_call 的关键字段（siliconflow MiniMax 丢 name、alibaba 百炼丢 id），空值会被写入
|
||
# checkpoint，导致工具结果无法按 tool_call_id 关联、工具状态永远停留在“进行中”。
|
||
# 对这些提供商在工具调用阶段禁用流式（正文回答仍流式）以拿到完整 tool_call。
|
||
# 该缺陷属 LangChain v3 流式协议上游问题（langchain#37420 / langchainjs#10937 /
|
||
# langgraphjs#2496），截至 langchain-core 1.4.4 仍未修复，待上游修复后可移除本处理。
|
||
_NON_STREAMING_TOOL_CALL_PROVIDERS = ("siliconflow", "alibaba")
|
||
|
||
|
||
def _requires_non_streaming_tool_calls(provider_id: str, model_id: str) -> bool:
|
||
return provider_id.startswith(_NON_STREAMING_TOOL_CALL_PROVIDERS)
|
||
|
||
|
||
def load_chat_model(fully_specified_name: str, **kwargs) -> BaseChatModel:
|
||
if not fully_specified_name:
|
||
raise ValueError("model spec 不能为空")
|
||
|
||
info = model_cache.get_model_info(fully_specified_name)
|
||
if not info:
|
||
available_specs = model_cache.get_all_specs("chat")
|
||
available_ids = [item.spec for item in available_specs[:10]]
|
||
raise ValueError(
|
||
f"Unknown model spec: '{fully_specified_name}'. "
|
||
f"Available chat models ({len(available_specs)}): {available_ids}"
|
||
)
|
||
|
||
if info.model_type != "chat":
|
||
raise ValueError(f"Model {fully_specified_name} is not a chat model (type={info.model_type})")
|
||
|
||
api_key = info.api_key
|
||
base_url = get_docker_safe_url(info.base_url)
|
||
|
||
logger.debug(f"Loading model {fully_specified_name} with provider_type={info.provider_type}")
|
||
|
||
if info.provider_type == "anthropic":
|
||
from langchain_anthropic import ChatAnthropic
|
||
|
||
return ChatAnthropic(
|
||
model=info.model_id,
|
||
api_key=SecretStr(api_key),
|
||
base_url=base_url,
|
||
**kwargs,
|
||
)
|
||
if info.provider_type == "gemini":
|
||
from langchain_google_genai import ChatGoogleGenerativeAI
|
||
|
||
return ChatGoogleGenerativeAI(
|
||
model=info.model_id,
|
||
google_api_key=SecretStr(api_key),
|
||
**kwargs,
|
||
)
|
||
|
||
from langchain_openai import ChatOpenAI
|
||
|
||
openai_kwargs = dict(kwargs)
|
||
if _requires_non_streaming_tool_calls(info.provider_id, info.model_id):
|
||
openai_kwargs.setdefault("disable_streaming", "tool_calling")
|
||
|
||
return ChatOpenAI(
|
||
model=info.model_id,
|
||
api_key=SecretStr(api_key),
|
||
base_url=base_url,
|
||
stream_usage=True,
|
||
**openai_kwargs,
|
||
)
|