Build docker container for ROCm (#1595)

* Dockerfile changes to build for ROCm * Adjust linker flags for ROCm * Update conda env for diffusers and transformers to use ROCm pytorch * Update transformers conda env for ROCm * ci: build hipblas images * fixup rebase * use self-hosted Signed-off-by: mudler <mudler@localai.io> * specify LD_LIBRARY_PATH only when BUILD_TYPE=hipblas --------- Signed-off-by: mudler <mudler@localai.io> Co-authored-by: mudler <mudler@localai.io>
2024-12-18 20:27:57 +00:00 · 2024-02-16 15:08:50 +01:00 · 2024-02-16 15:08:50 +01:00 · fb0a4c5d9a
commit fb0a4c5d9a
parent e690bf387a
9 changed files with 233 additions and 5 deletions
--- a/.github/workflows/image-pr.yml
+++ b/.github/workflows/image-pr.yml
@ -51,6 +51,14 @@ jobs:
            image-type: 'extras'
            runs-on: 'arc-runner-set'
            base-image: "ubuntu:22.04"
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
  core-image-build:
    uses: ./.github/workflows/image_build.yml
    with:
--- a/.github/workflows/image.yml
+++ b/.github/workflows/image.yml
@ -103,6 +103,22 @@ jobs:
            image-type: 'extras'
            base-image: "ubuntu:22.04"
            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg'
+            ffmpeg: 'true'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas'
+            ffmpeg: 'false'
+            image-type: 'extras'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
  core-image-build:
    uses: ./.github/workflows/image_build.yml
    with:
@ -124,6 +140,22 @@ jobs:
    strategy:
      matrix:
        include:
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-ffmpeg-core'
+            ffmpeg: 'true'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
+          - build-type: 'hipblas'
+            platforms: 'linux/amd64'
+            tag-latest: 'false'
+            tag-suffix: '-hipblas-core'
+            ffmpeg: 'false'
+            image-type: 'core'
+            base-image: "rocm/dev-ubuntu-22.04:6.0-complete"
+            runs-on: 'arc-runner-set'
          - build-type: ''
            platforms: 'linux/amd64'
            tag-latest: 'false'
--- a/7
+++ b/7
@ -1,4 +1,3 @@
-ARG GO_VERSION=1.21
 ARG IMAGE_TYPE=extras
 ARG BASE_IMAGE=ubuntu:22.04

@ -42,8 +41,12 @@ RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
    apt-get install -y cuda-nvcc-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcurand-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcublas-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusparse-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION} libcusolver-dev-${CUDA_MAJOR_VERSION}-${CUDA_MINOR_VERSION}  && apt-get clean \
    ; fi

+# Cuda
 ENV PATH /usr/local/cuda/bin:${PATH}

+# HipBLAS requirements
+ENV PATH /opt/rocm/bin:${PATH}
+
 # OpenBLAS requirements and stable diffusion
 RUN apt-get install -y \
    libopenblas-dev \
@ -70,7 +73,9 @@ RUN curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmo
    apt-get install -y conda && apt-get clean

 ENV PATH="/root/.cargo/bin:${PATH}"
+RUN apt-get install -y python3-pip && apt-get clean
 RUN pip install --upgrade pip
+
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
 RUN apt-get install -y espeak-ng espeak && apt-get clean

--- a/4
+++ b/4
@ -97,6 +97,8 @@ endif

 ifeq ($(BUILD_TYPE),hipblas)
 	ROCM_HOME ?= /opt/rocm
+	ROCM_PATH ?= /opt/rocm
+	LD_LIBRARY_PATH ?= /opt/rocm/lib:/opt/rocm/llvm/lib
 	export CXX=$(ROCM_HOME)/llvm/bin/clang++
 	export CC=$(ROCM_HOME)/llvm/bin/clang
 	# llama-ggml has no hipblas support, so override it here.
@ -105,7 +107,7 @@ ifeq ($(BUILD_TYPE),hipblas)
 	GPU_TARGETS ?= gfx900,gfx90a,gfx1030,gfx1031,gfx1100
 	AMDGPU_TARGETS ?= "$(GPU_TARGETS)"
 	CMAKE_ARGS+=-DLLAMA_HIPBLAS=ON -DAMDGPU_TARGETS="$(AMDGPU_TARGETS)" -DGPU_TARGETS="$(GPU_TARGETS)"
-	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link
+	CGO_LDFLAGS += -O3 --rtlib=compiler-rt -unwindlib=libgcc -lhipblas -lrocblas --hip-link -L${ROCM_HOME}/lib/llvm/lib
 endif

 ifeq ($(BUILD_TYPE),metal)
--- a/backend/python/common-env/transformers/Makefile
+++ b/backend/python/common-env/transformers/Makefile
@ -4,6 +4,10 @@ ifeq ($(BUILD_TYPE), cublas)
 	CONDA_ENV_PATH = "transformers-nvidia.yml"
 endif

+ifeq ($(BUILD_TYPE), hipblas)
+	CONDA_ENV_PATH = "transformers-rocm.yml"
+endif
+
 .PHONY: transformers
 transformers:
 	@echo "Installing $(CONDA_ENV_PATH)..."
--- a/backend/python/common-env/transformers/transformers-rocm.yml
+++ b/backend/python/common-env/transformers/transformers-rocm.yml
@ -0,0 +1,109 @@
+name: transformers
+channels:
+  - defaults
+dependencies:
+  - _libgcc_mutex=0.1=main
+  - _openmp_mutex=5.1=1_gnu
+  - bzip2=1.0.8=h7b6447c_0
+  - ca-certificates=2023.08.22=h06a4308_0
+  - ld_impl_linux-64=2.38=h1181459_1
+  - libffi=3.4.4=h6a678d5_0
+  - libgcc-ng=11.2.0=h1234567_1
+  - libgomp=11.2.0=h1234567_1
+  - libstdcxx-ng=11.2.0=h1234567_1
+  - libuuid=1.41.5=h5eee18b_0
+  - ncurses=6.4=h6a678d5_0
+  - openssl=3.0.11=h7f8727e_2
+  - pip=23.2.1=py311h06a4308_0
+  - python=3.11.5=h955ad1f_0
+  - readline=8.2=h5eee18b_0
+  - setuptools=68.0.0=py311h06a4308_0
+  - sqlite=3.41.2=h5eee18b_0
+  - tk=8.6.12=h1ccaba5_0
+  - wheel=0.41.2=py311h06a4308_0
+  - xz=5.4.2=h5eee18b_0
+  - zlib=1.2.13=h5eee18b_0
+  - pip:
+      - --pre
+      - --extra-index-url https://download.pytorch.org/whl/nightly/
+      - accelerate==0.23.0
+      - aiohttp==3.8.5
+      - aiosignal==1.3.1
+      - async-timeout==4.0.3
+      - attrs==23.1.0
+      - bark==0.1.5
+      - boto3==1.28.61
+      - botocore==1.31.61
+      - certifi==2023.7.22
+      - TTS==0.22.0
+      - charset-normalizer==3.3.0
+      - datasets==2.14.5
+      - sentence-transformers==2.2.2
+      - sentencepiece==0.1.99
+      - dill==0.3.7
+      - einops==0.7.0
+      - encodec==0.1.1
+      - filelock==3.12.4
+      - frozenlist==1.4.0
+      - fsspec==2023.6.0
+      - funcy==2.0
+      - grpcio==1.59.0
+      - huggingface-hub
+      - idna==3.4
+      - jinja2==3.1.2
+      - jmespath==1.0.1
+      - markupsafe==2.1.3
+      - mpmath==1.3.0
+      - multidict==6.0.4
+      - multiprocess==0.70.15
+      - networkx
+      - numpy==1.26.0
+      - packaging==23.2
+      - pandas
+      - peft==0.5.0
+      - protobuf==4.24.4
+      - psutil==5.9.5
+      - pyarrow==13.0.0
+      - python-dateutil==2.8.2
+      - pytz==2023.3.post1
+      - pyyaml==6.0.1
+      - regex==2023.10.3
+      - requests==2.31.0
+      - rouge==1.0.1
+      - s3transfer==0.7.0
+      - safetensors==0.3.3
+      - scipy==1.11.3
+      - six==1.16.0
+      - sympy==1.12
+      - tokenizers
+      - torch
+      - torchaudio
+      - tqdm==4.66.1
+      - triton==2.1.0
+      - typing-extensions==4.8.0
+      - tzdata==2023.3
+      - auto-gptq==0.6.0
+      - urllib3==1.26.17
+      - xxhash==3.4.1
+      - yarl==1.9.2
+      - soundfile
+      - langid
+      - wget
+      - unidecode
+      - pyopenjtalk-prebuilt
+      - pypinyin
+      - inflect
+      - cn2an
+      - jieba
+      - eng_to_ipa
+      - openai-whisper
+      - matplotlib
+      - gradio==3.41.2
+      - nltk
+      - sudachipy
+      - sudachidict_core
+      - vocos
+      - vllm==0.2.7
+      - transformers>=4.36.0 # Required for Mixtral.
+      - xformers==0.0.23.post1
+prefix: /opt/conda/envs/transformers
--- a/backend/python/diffusers/Makefile
+++ b/backend/python/diffusers/Makefile
@ -1,4 +1,8 @@
-CONDA_ENV_PATH = "diffusers.yml"
+export CONDA_ENV_PATH = "diffusers.yml"
+
+ifeq ($(BUILD_TYPE), hipblas)
+export CONDA_ENV_PATH = "diffusers-rocm.yml"
+endif

 .PHONY: diffusers
 diffusers:
@ -12,4 +16,4 @@ run:
 	@echo "Diffusers run."

 test:
-	bash test.sh
+	bash test.sh
--- a/backend/python/diffusers/diffusers-rocm.yml
+++ b/backend/python/diffusers/diffusers-rocm.yml
@ -0,0 +1,64 @@
+name: diffusers
+channels:
+  - defaults
+dependencies:
+  - _libgcc_mutex=0.1=main
+  - _openmp_mutex=5.1=1_gnu
+  - bzip2=1.0.8=h7b6447c_0
+  - ca-certificates=2023.08.22=h06a4308_0
+  - ld_impl_linux-64=2.38=h1181459_1
+  - libffi=3.4.4=h6a678d5_0
+  - libgcc-ng=11.2.0=h1234567_1
+  - libgomp=11.2.0=h1234567_1
+  - libstdcxx-ng=11.2.0=h1234567_1
+  - libuuid=1.41.5=h5eee18b_0
+  - ncurses=6.4=h6a678d5_0
+  - openssl=3.0.11=h7f8727e_2
+  - pip=23.2.1=py311h06a4308_0
+  - python=3.11.5=h955ad1f_0
+  - readline=8.2=h5eee18b_0
+  - setuptools=68.0.0=py311h06a4308_0
+  - sqlite=3.41.2=h5eee18b_0
+  - tk=8.6.12=h1ccaba5_0
+  - tzdata=2023c=h04d1e81_0
+  - wheel=0.41.2=py311h06a4308_0
+  - xz=5.4.2=h5eee18b_0
+  - zlib=1.2.13=h5eee18b_0
+  - pip:
+      - --pre
+      - --extra-index-url https://download.pytorch.org/whl/nightly/
+      - accelerate>=0.11.0
+      - certifi==2023.7.22
+      - charset-normalizer==3.3.0
+      - compel==2.0.2
+      - diffusers==0.24.0
+      - filelock==3.12.4
+      - fsspec==2023.9.2
+      - grpcio==1.59.0
+      - huggingface-hub>=0.19.4
+      - idna==3.4
+      - importlib-metadata==6.8.0
+      - jinja2==3.1.2
+      - markupsafe==2.1.3
+      - mpmath==1.3.0
+      - networkx==3.1
+      - numpy==1.26.0
+      - omegaconf
+      - packaging==23.2
+      - pillow==10.0.1
+      - protobuf==4.24.4
+      - psutil==5.9.5
+      - pyparsing==3.1.1
+      - pyyaml==6.0.1
+      - regex==2023.10.3
+      - requests==2.31.0
+      - safetensors==0.4.0
+      - sympy==1.12
+      - tqdm==4.66.1
+      - transformers>=4.25.1
+      - triton==2.1.0
+      - typing-extensions==4.8.0
+      - urllib3==2.0.6
+      - zipp==3.17.0
+      - torch
+prefix: /opt/conda/envs/diffusers
--- a/backend/python/diffusers/diffusers.yml
+++ b/backend/python/diffusers/diffusers.yml
@ -71,4 +71,4 @@ dependencies:
      - typing-extensions==4.8.0
      - urllib3==2.0.6
      - zipp==3.17.0
-prefix: /opt/conda/envs/diffusers
+prefix: /opt/conda/envs/diffusers