torch

torchvision

transformers>=4.49.0

accelerate

bitsandbytes

fastapi

pydantic

uvicorn

pdfminer.six

opencv-python

pymupdf # for unstructured fast mode

tantivy==0.25.1

tiktoken

gpt4all

chromadb

# vllm # only supported via pip for cuda machines, otherwise requires a custom build

# llama-cpp-python==0.3.1 # can be built for CPU only

huggingface_hub

rank_bm25

unstructured[docx]

unstructured[pptx]

unstructured[xlsx]

unstructured[pdf]

unstructured-inference # needed for non fast partitioning

langchain-core

langchain-community

langchain-chroma

langchain-huggingface

langchain-openai

langchain-ollama

langchain-unstructured

faiss-gpu-cu12

ujson

GitPython

ninja

qwen_vl_utils

wheel

#flash-attn; --global-option="--no-build-isolation"

pdf2image

pypandoc_binary

accelerate

python-doctr

diffusers