Invalid response object from API: '{"object":"error","message":"**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**\\n\\n(probability tensor contains either `inf`, `nan` or element < 0)","code":50001}'

Nobody has claimed this yet.

I encountered an strange issue. I used the solutionin this issue （https://github.com/lm-sys/FastChat/issues/412）, and the curl command is working fine, and the web interface is also good, I can have a normal chat on gradio web server. but when I create an agent by langchain，It runs with error:
> Retrying langchain.chat_models.openai.ChatOpenAI.completion_with_retry.._completion_with_retry in 10.0 seconds as it raised APIError: Invalid response object from API: '{"object":"error","message":"**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**\\n\\n(probability tensor contains either `inf`, `nan` or element < 0)","code":50001}' (HTTP response code was 400).
Error in StdOutCallbackHandler.on_retry callback: 'StdOutCallbackHandler' object has no attribute 'on_retry'

My langchain code:
```python
from langchain.chat_models import ChatOpenAI
from langchain.llms import OpenAI
from langchain.document_loaders import TextLoader
from langchain.embeddings import OpenAIEmbeddings
from langchain.indexes import VectorstoreIndexCreator
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler

llm = ChatOpenAI(model="chatglm2-6b",temperature=0.0001,verbose=True)
from langchain.base_language import BaseLanguageModel
from langchain.chains.llm import LLMChain
from langchain.prompts.few_shot import FewShotPromptTemplate
from langchain.prompts.prompt import PromptTemplate
prompt='''\
what's the answer of {expr}。
'''
chain = LLMChain(llm=llm,prompt=PromptTemplate.from_template(prompt))
chain.run("3 times 4")

```

Start by reproducing the LangChain ChatOpenAI example in the issue and compare its request with the working curl and Gradio requests. Investigate the API response and callback error together; done means the shown agent request completes without the HTTP 400 or invalid-response error.

Written by the indexing model from the issue text.