清尘
LangGraph流式输出问题记录
前情
我克隆项目代码安装环境后，跑项目中间碰到的问题如下：
对图设置流式输出的模式stream_mode有下面这几种，项目里面配置了messages和updates，原理上在配置了messages后会把图里面有关大模型输出的内容全部吐出来，但是我这边实际上并没有输出messages，但是updates是正常的，一开始以为是代码哪里写的有问题，但是回过神来我是刚拷的公司项目代码什么都没改，同事本地也能跑，就定位到是环境的问题了（langgraph包的问题）。
https://langchain-ai.github.io/langgraph/reference/types/#langgraph.types.StreamMode
StreamMode = Literal[ "values", "updates", "checkpoints", "tasks", "debug", "messages", "custom", ]
async for stream_mode, chunk in graph.astream(
{"some_text": "original text"},
stream_mode=["values", "updates", "messages"],
config=config
):
print(f"Stream mode: {stream_mode}，chunk：{chunk}")
print("\n")
目前的解决办法
降低langgraph的版本
测试下来0.5.x和0.6.x都不行
最后安装0.4.5解决
pip install langgraph==0.4.5
pip install langgraph-prebuilt==0.5.0
我看到相关issue已经有人提出来了:在这里
插曲
发现不同平台提供的不同模型的流式返回也不一样：
比如gemini和硅基流动这两个，我让他们给我一个4字词语二者流式返回如下：
# gemini
Stream mode: messages，chunk：(AIMessageChunk(content='运转如常', additional_kwargs={}, response_metadata={'safety_ratings': [], 'usage_metadata': {}, 'finish_reason': 'STOP', 'model_name': 'gemini-2.5-flash'}, id='run--8d9060e0-7f53-4813-953d-77bf79525b97', usage_metadata={'input_tokens': 22, 'output_tokens': 3, 'total_tokens': 1101, 'input_token_details': {'cache_read': 0}, 'output_token_details': {'reasoning': 1076}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:c516e5ad-20d4-58da-24ce-ce6524c362cf|task_node:c8f26ac3-6e77-e501-9dd7-5e9cdada09f2', 'checkpoint_ns': 'human_node:c516e5ad-20d4-58da-24ce-ce6524c362cf|task_node:c8f26ac3-6e77-e501-9dd7-5e9cdada09f2', 'ls_provider': 'google_vertexai', 'ls_model_name': 'gemini-2.5-flash', 'ls_model_type': 'chat', 'ls_temperature': None})
# siliconflow
Stream mode: messages，chunk：(AIMessageChunk(content='', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 284, 'total_tokens': 312, 'input_token_details': {}, 'output_token_details': {'reasoning': 284}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
...
Stream mode: messages，chunk：(AIMessageChunk(content='\n\n', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 288, 'total_tokens': 316, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
Stream mode: messages，chunk：(AIMessageChunk(content='按', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 289, 'total_tokens': 317, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
Stream mode: messages，chunk：(AIMessageChunk(content='部', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 290, 'total_tokens': 318, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
Stream mode: messages，chunk：(AIMessageChunk(content='就', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 291, 'total_tokens': 319, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
Stream mode: messages，chunk：(AIMessageChunk(content='班', additional_kwargs={}, response_metadata={}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 292, 'total_tokens': 320, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
Stream mode: messages，chunk：(AIMessageChunk(content='', additional_kwargs={}, response_metadata={'finish_reason': 'stop', 'model_name': 'Qwen/Qwen3-32B'}, id='run--9a4bb426-9da3-4307-8001-708b6b653556', usage_metadata={'input_tokens': 28, 'output_tokens': 292, 'total_tokens': 320, 'input_token_details': {}, 'output_token_details': {'reasoning': 287}}), {'langgraph_step': 1, 'langgraph_node': 'task_node', 'langgraph_triggers': ('__pregel_push',), 'langgraph_path': ('__pregel_push', ('__pregel_pull', 'human_node'), 0, True), 'langgraph_checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'checkpoint_ns': 'human_node:1c4f6ac7-4460-ed5d-f773-8ec231b97b59|task_node:2a668166-512c-249b-bb01-9226ae6e9aff', 'ls_provider': 'openai', 'ls_model_name': 'Qwen/Qwen3-32B', 'ls_model_type': 'chat', 'ls_temperature': 0.9})
siliconflow一直返回空，真是害人呐！！！