Hi :slightly_smiling_face: Probably a very dumb qu...
# ask-for-help
p
Hi 🙂 Probably a very dumb question, trying to get the container to run properly for dbrx. This gives the error "ValueError: Failed to determine config class for 'databricks/dbrx-instruct'. Make sure databricks/dbrx-instruct is saved with openllm."
Copy code
docker run --rm -it -p 3000:3000 -e TRUST_REMOTE_CODE=True -e HUGGING_FACE_HUB_TOKEN=XXXXXXXXXXXXXXXXXXXX ghcr.io/bentoml/openllm start databricks/dbrx-instruct --backend pt
Which I assume is because in github readme it says
Copy code
Note: Dbrx requires to install with:
pip install "openllm[dbrx]"
But how do I add the install to the container? I didn't really find a documentation for the container besides the run commands in readme unfortunately 😕 EDIT: Actually seems i have a general problem as well. Tried to use a model that does not require install. But then I just get this.... O.o
Copy code
File "/openllm-python/src/openllm/serialisation/transformers/_helpers.py", line 33, in infer_autoclass_from_llm
raise ValueError(
ValueError: Invalid configuration for google/flan-t5-large. ``trust_remote_code=True`` requires `transformers.PretrainedConfig` to contain a `auto_map` mapping

2024-04-16T21:06:49+0000 [ERROR] [runner:llm-flan-t5-runner:1] Application startup failed. Exiting.
This is now simple "start google/flan-t5-large --backend pt" for the container. Doing this on an up2date debian 12
c
Hi @Panslothda -
Could you try without the backend pt flat?
@Aaron Pham could you help take a look?
@Panslothda it may has to do with the dbrx support has not been released yet
in the mean while, feel free to try out the example repo here for serving LLM using BentoML and VLLM async engine directly: https://github.com/bentoml/BentoVLLM - slightly lower level APIs but gives user more control
a
Will doo