* Adds prompting scripts for the pipeline

* Adds infer_gemini

* Adds llava inference

* Adds requirements for new scripts

* Minor fixes

* Adds batched inference for llava

* Minor fixes

* Adds inference_blip

* Adds retrieval part for e2e script

* Adds bash script

added_candidate_lookup

Added configurable k and zeroshot text prompt

add qid to llm output, skip images not in query dataset

This cl makes the following changes to the bash file:

get the image count from ls
increase the batch size to 512
minor clean up

tokenize before score eval and add eval for retriever

For llava and blip where we batch the inference, qid should be stored in a list, similar to prompts

…needed (#13)

* caption generation with image-text pairs for llava and gpt models

* updated eval script

* Minor modifications with eval

* Black formatting changes

---------

Co-authored-by: sahel <[email protected]>