Error message is not helpful when the input to a model exceeds the limit.

[gamer-mitsuha]

Context: I tried to send chunked texts to an embedding model using Ollama and got the following error:

.venv/lib/python3.12/site-packages/ollama/_client.py", line 120, in _request_raw
    raise ResponseError(e.response.text, e.response.status_code) from None
ollama._types.ResponseError: {}

It was not clear what happened and I had to check the failed batch of examples and examine what happened. Eventually I was able to figure out the reason is that the chunk limit is higher than the allowed input size, and thus the server returns error for such long inputs.

I'm wondering if it's possible to produce more helpful error message in such cases?

[ParthSareen]

Hey @gamer-mitsuha - sorry about that. Seems like the Ollama server didn't give an output back or we should have captured that. Thanks for raising - will take a look!