Langchain: Update Vectara templates #15363
Conversation
added self-query template
dosubot
bot
added
the
size:L
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎ 1 Ignored Deployment
|
dosubot
bot
added
Ɑ: models
There was a problem hiding this comment.
Hey @ofermend ! The multiquery changes look great.
I don't think the self query one has been tested. I think the chain is missing a prompt and LLM call potentially, and let me know if that's incorrect! I run into a KeyError on -1 when running it because the object passed into the lambda is a dict.
|
|
||
| chain = ( | ||
| RunnableParallel({"context": retriever, "question": RunnablePassthrough()}) | ||
| | (lambda res: res[-1]) |
There was a problem hiding this comment.
yes, you're right.
I've spent some time looking at this more closely and I think it's more complex than I had originally thought. Not this bug specifically, I mean the whole self-query RAG. I wanted to make it work such that it uses Vectara summarizer, but it seems quite complex in the current setting, as SelfQueryRetriever seems to be built around retreiving documents and then using a separate LLM to do the final summarization.
- I could probably still make this template follow this path - use Vectara as purely a document retriever (without summarization) - what other example is best to follow in this case?
- Potentially later we can refactor this to make SelfQuery also support accepting a summarization component.
WDYT?
There was a problem hiding this comment.
@efriis due to not having too much bandwidth next few weeks to make this work, let me just clean up (remove) the SQ stuff and will just leave the good fixes for MQ. We can tackle the SQ at some later time. SG?
| python = ">=3.8.1,<4.0" | ||
| langchain = ">=0.0.325" | ||
| openai = "<2" | ||
| tiktoken = "^0.5.1" |
dosubot
bot
added
size:S
|
Updated w/o SelfQuery |
|
@efriis - update as discussed. Pls let me know if anything else missing. |
dosubot
bot
added
the
lgtm
fixed multi-query template for Vectara
added self-query template for Vectara
Also added prompt_name parameter to summarization
CC @efriis
Twitter handle: @ofermend