Diwank Singh Tomer

[Jamba](https://huggingface.co/ai21labs/Jamba-v0.1) is a really interesting model. It’s an MoE+Transformer+Mamba hybrid so I’m not sure how that would work with vllm. I’d love to add support for it. thoughts / pointers?...

AFAIK, not really possible to do at the generation level but might be a good idea to tune a classifier on top of vector data stores. Openai cookbook includes some...

it throws an error from time to time `IndexError`. Noticed it while using the `QAGenerateChain`

Example code: ```python texts = essay_texts + wiki_texts + webpages examples = [] num_batches = 10 batches = np.array_split(texts, num_batches) for batch in tqdm_notebook(batches): docs = [ {"doc": t} for...

For instance, I'm using it like so: ```python class FinishReasonChain(LLMChain): prefer_finish_on: Optional[str] = None def apply(self, input_list: List[Dict[str, Any]]) -> List[Dict[str, str]]: """Utilize the LLM generate method for speed gains."""...

IMHO, prefer outputs with `finish_reson=='stop'` should be part of the default implementation. OpenAI itself recommends it. Do you think it makes sense to patch the `OpenAIChain` with that. Something like...

gotcha! feel free to close this until then. Also, can you point me to a source file / function def where this is implemented?

Added implementation for a `MultiVectorDBQA` class that can query multiple vector stores for a query. Ran lint and format. Happy to expand further.

Thanks for the correction. :)

Yes I agree. Is there some dataset that we could use as a good benchmark?