ENH: Add multi-backend tests for bnb

.github/workflows/nightly-bnb.yml

@@ -18,7 +18,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-          docker-image-name: ["huggingface/peft-gpu-bnb-source:latest", "huggingface/peft-gpu-bnb-latest:latest"]
+          docker-image-name: ["huggingface/peft-gpu-bnb-source:latest", "huggingface/peft-gpu-bnb-latest:latest", "huggingface/peft-gpu-bnb-multi-source:latest"]
     runs-on: [self-hosted, single-gpu, nvidia-gpu, t4, ci]
     env:
       CUDA_VISIBLE_DEVICES: "0"
@@ -74,7 +74,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        docker-image-name: ["huggingface/peft-gpu-bnb-source:latest", "huggingface/peft-gpu-bnb-latest:latest"]
+        docker-image-name: ["huggingface/peft-gpu-bnb-source:latest", "huggingface/peft-gpu-bnb-latest:latest", "huggingface/peft-gpu-bnb-multi-source:latest"]
     runs-on: [self-hosted, multi-gpu, nvidia-gpu, t4, ci]
     env:
       CUDA_VISIBLE_DEVICES: "0,1"

docker/peft-gpu-bnb-multi-source/Dockerfile

@@ -0,0 +1,68 @@
+# Builds GPU docker image of PyTorch
+# Uses multi-staged approach to reduce size
+# Stage 1
+# Use base conda image to reduce time
+FROM continuumio/miniconda3:latest AS compile-image
+# Specify py version
+ENV PYTHON_VERSION=3.8
+# Install apt libs - copied from https://github.com/huggingface/accelerate/blob/main/docker/accelerate-gpu/Dockerfile
+RUN apt-get update && \
+    apt-get install -y curl git wget software-properties-common git-lfs && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists*
+
+# Install audio-related libraries 
+RUN apt-get update && \
+    apt install -y ffmpeg
+
+RUN apt install -y libsndfile1-dev
+RUN git lfs install
+
+# Create our conda env - copied from https://github.com/huggingface/accelerate/blob/main/docker/accelerate-gpu/Dockerfile
+RUN conda create --name peft python=${PYTHON_VERSION} ipython jupyter pip
+RUN python3 -m pip install --no-cache-dir --upgrade pip
+
+# Below is copied from https://github.com/huggingface/accelerate/blob/main/docker/accelerate-gpu/Dockerfile
+# We don't install pytorch here yet since CUDA isn't available
+# instead we use the direct torch wheel
+ENV PATH /opt/conda/envs/peft/bin:$PATH
+# Activate our bash shell
+RUN chsh -s /bin/bash
+SHELL ["/bin/bash", "-c"]
+
+# Stage 2
+FROM nvidia/cuda:12.1.0-devel-ubuntu22.04 AS build-image
+COPY --from=compile-image /opt/conda /opt/conda
+ENV PATH /opt/conda/bin:$PATH
+
+RUN chsh -s /bin/bash
+SHELL ["/bin/bash", "-c"]
+
+# Install apt libs
+RUN apt-get update && \
+    apt-get install -y curl git wget cmake && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists*
+
+# Activate the conda env and install transformers + accelerate from source
+# Also clone BNB and build it from source.
+RUN source activate peft && \
+    python3 -m pip install -U --no-cache-dir \
+    librosa \
+    "soundfile>=0.12.1" \
+    scipy \
+    git+https://github.com/huggingface/transformers \
+    git+https://github.com/huggingface/accelerate \
+    peft[test]@git+https://github.com/huggingface/peft \
+    optimum \
+    auto-gptq && \
+    git clone https://github.com/TimDettmers/bitsandbytes && cd bitsandbytes && git checkout multi-backend-refactor && \
+    cmake -B . -DCOMPUTE_BACKEND=cuda -S . && \
+    cmake --build . && \
+    pip install -e . && \ 
+    pip freeze | grep bitsandbytes
+
+RUN echo "source activate peft" >> ~/.profile
+
+# Activate the virtualenv
+CMD ["/bin/bash"]