Open ViktorooReps opened 1 week ago
Hi @ViktorooReps, thanks for reaching out.
We are planning some performance improvement in the next days/week. If you are willing to contribute, it will for sure be appreciated.
Performance will be addressed in three ways
The initial thought about 3 are
We have ran multiple time a RAG pipeline with included examples here with Milvus ... even with scaling on NVIDIA GPU L4 machines and it worked very well. Was a bit challenging to compile the docker image for that, but it seems to perform better Not really did a performance metrics, but from the observations is at least 2x faster. Looking forward to v2 implementation, thank you guys for that great job.
FYI is the start of the Dockerfile for getting the things run in the docker image. Ensure, that drivers are proper activated to docker with gpu capabilities, that GPU is really used:
##########
FROM nvidia/cuda:12.6.1-runtime-ubuntu24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
python3 python3-venv python3-dev python3-pip cron git curl && \
rm -rf /var/lib/apt/lists/*
WORKDIR /app
RUN python3 -m venv /opt/venv
# installing pytorch with GPU support (CUDA 12.1)
RUN /opt/venv/bin/pip install --upgrade pip && \
/opt/venv/bin/pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
.......
###########
@leviataniac thanks for sharing this!
Hello Deep Search Team!
Thank you for this contribution to open source!
We are considering using your library to parse PDF files for LLM training, so we will potentially need to scale things up. Do you have any updates on GPU/multi-GPU support? Maybe some directions on where to start if we were to work on GPU support ourselves?