Merged Model from Huggingface runs fine with fastchat CLI but not when using service worker
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 39.5k
- Forks
- 4.8k
- PR merge metrics
- No merged PRs in 30d
Description
I am running Fastchat on kubernetes. I have a worker for the controller, the fastchat api and a (gpu) worker for each of the models. When I pull this model from huggingface (downloaded using huggingface-cli) https://huggingface.co/Rmote6603/MedPrescription-FineTuning, I run the fastchat CLI command and type in my prompt, it works perfectly fine as expected:
`python3.9 -m fastchat.serve.cli --model-path MedPrescription-FineTuning`
However, when I use the fastchat.serve.model_worker, it does not work at all when I try to use chat completion API, it gives me an error, even though v1/models API works as shown in the photo below:
`python3.9 -m fastchat.serve.model_worker --model-path MedPrescription-FineTuning --worker-address http://localhost:21002 --port 21002`
When I run this POST request,
`curl http://localhost:8000/v1/chat/completions
-H "Content-Type: application/json" -H "Authorization:Bearer API-TOKEN"
-d '{
"model": "MedPrescription-FineTuning",
"messages": [{"role": "user", "content": "Hello! What is your name?"}]
}'
`
It first times out:
Then it subsequently gives me Network Error:
`{"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}`
I was wondering if anyone else has ran into this issue before. Does it have anything to do with Huggingface, models weights or something with FastChat limitations. I have only having issues with this merged mistral model.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the fastchat.serve.cli and fastchat.serve.model_worker entry points, then compare the working CLI invocation with the Kubernetes worker and the /v1/chat/completions request. Reproduce the timeout and probability-tensor error while checking the /v1/models response; done means identifying whether the failure is in FastChat serving or this merged Hugging Face model and documenting the evidence.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- huggingface, kubernetes, python
- Domain
- ai, api, backend, infrastructure
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100