RAG app example #118
RAG app example #118
Conversation
heyjustinai
requested review from
ashwinb,
yanxi0830,
hardikjshah,
dltn and
raghotham
as code owners
This reverts commit 7b00a1c.
| @@ -0,0 +1,181 @@ | |||
| import argparse | |||
There was a problem hiding this comment.
I think we should simply call this directory DocQA with the organization:
- DocQA
- app.py
- README.md
- scripts/
- data/
Could we also avoid prefixing files 01_ or 02_ etc.
|
|
||
| ### How to run the pipeline: | ||
|
|
||
|  |
There was a problem hiding this comment.
any chance we can simplify this diagram a lot? actually, I think a simpler inline Mermaid diagram which shows just the basic high-level flow would be more useful. docker, etc. should be completely avoided.
| } | ||
| ``` | ||
|
|
||
| 2. Inside of docker folder, `run_RAG.sh` is the main script that can create `.env` file for compose.yaml and then actually start the `docker compose` process to launch all the pipelines in our dockers. `compose.yaml` is the main docker yaml that specifies all the mount option and docker configs, change the mounts if needed. |
There was a problem hiding this comment.
these details are unnecessary I think. these scripts are very simple and self-documenting in a way.
examples/E2E-RAG-App/README.md
Outdated
|
|
||
| 4. ChromaDB docker will also start. This docker will host the chroma database that can interact with llama-stack. | ||
|
|
||
| 5. Lastly, Llama-stack docker will start. The `llama_stack_start.sh` control the docker startup behavior, change it if needed. It will first run the ingestion pipeline to convert all the documents into MarkDown files. Then, it will run llama-stack server based on the `llama_stack_run.yaml` config. Once the server is ready, then it will run the `gradio_interface.py` which will insert document chunks into memory_bank and start the UI for user interaction. |
There was a problem hiding this comment.
we have scripts within scripts, is it possible to inline some of them perhaps?
| ], | ||
| "query_generator_config": {"type": "default", "sep": " "}, | ||
| "max_tokens_in_context": 300, | ||
| "max_chunks": 5, |
There was a problem hiding this comment.
I think we should just elide this and leave it to default. because we don't want people to be thinking about all these pieces when they first look at the stack (and even later if we do a good job)
There was a problem hiding this comment.
sounds good, will leave out "query_generator_config": {"type": "default", "sep": " "},
but since we are running ollama locally, we will need to keep the max_tokens_in_context and max_chunks for it to run at resonable speed
… a ragservice template
What does this PR do?
Creating a E2E RAG example that is able to do retrieval on documents and answer user questions. Components included:
Inference (with llama-stack)
Memory (with llama-stack)
Agent (with llama-stack)
Frontend (with Gradio)
Feature/Issue validation/testing/test plan
1120.mov
Before submitting
Pull Request section?
to it if that's the case.
Thanks for contributing 🎉!