initial commit

Dockerfile

@@ -0,0 +1,43 @@
+FROM rocm/pytorch:rocm6.2.3_ubuntu22.04_py3.10_pytorch_release_2.3.0 AS base-builder
+
+ENV DISTFILES_TAG=v1
+ENV GPU_ARCHS="gfx1100"
+ENV ROCM_TARGET="gfx1100"
+ENV HIP_PLATFORM=amd
+ENV DS_BUILD_CPU_ADAM=1 
+ENV TORCH_HIP_ARCH_LIST="gfx1100"
+
+RUN apt-get update
+RUN apt-get install -y wget git build-essential ninja-build git-lfs libaio-dev cmake
+RUN git lfs install
+RUN apt-get install -y --allow-change-held-packages vim curl nano rsync s3fs
+RUN apt-get install -y libjpeg-dev python3-dev python3-pip
+
+WORKDIR /workspace
+#RUN git clone --recurse https://github.com/ROCm/bitsandbytes.git
+RUN git clone --depth 1 -b multi-backend-refactor https://github.com/bitsandbytes-foundation/bitsandbytes.git
+WORKDIR /workspace/bitsandbytes
+#RUN git checkout rocm_enabled
+RUN pip install -r /workspace/bitsandbytes/requirements-dev.txt
+RUN cmake -DBNB_ROCM_ARCH="gfx1100" -DCOMPUTE_BACKEND=hip -S .
+RUN make
+RUN git tag 0.44.2
+RUN pip install .
+
+WORKDIR /workspace
+RUN pip install trl
+RUN pip install peft
+RUN pip install transformers datasets huggingface-hub scipy
+
+RUN git clone https://github.com/ggerganov/llama.cpp.git
+WORKDIR /workspace/llama.cpp
+RUN make -j32 GGML_HIPBLAS=1 AMDGPU_TARGETS=gfx1100
+RUN pip install -r requirements/requirements-convert_legacy_llama.txt
+
+WORKDIR /workspace
+COPY convert_to_gguf.sh /workspace/convert_to_gguf.sh
+RUN chmod +x /workspace/convert_to_gguf.sh
+COPY setup.sh /workspace/setup.sh
+RUN chmod +x /workspace/setup.sh
+
+ENTRYPOINT ["/workspace/setup.sh"]

README.md

@@ -0,0 +1 @@
+# lm-container

convert_to_gguf.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+if [[ $# -ne 2 ]] ; then
+    echo 'Arguments: <path> <model name>'
+    exit 1
+fi
+
+cd /workspace/llama.cpp
+./convert_hf_to_gguf.py --model-name $2 $1
+./llama-quantize $1/*-F16.gguf Q8_0
+./llama-quantize $1/*-F16.gguf Q6_K
+./llama-quantize $1/*-F16.gguf Q5_K_M
+./llama-quantize $1/*-F16.gguf Q4_K_M
+./llama-quantize $1/*-F16.gguf Q4_0_4_4
+./llama-quantize $1/*-F16.gguf IQ3_M

run.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+docker run --device=/dev/kfd --device=/dev/dri --privileged --security-opt seccomp=unconfined --group-add video --shm-size 20g --rm -it --name axolotl --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 --mount type=bind,src="/home/sb/lm/",target=/workspace/data --mount type=bind,src="/home/sb/svn/lm-finetuner/",target=/workspace/lm-finetuner pytorch-rocm-exo:v1

setup.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+echo Ready
+/bin/bash