Merge branch 'main' of https://github.com/h2oai/h2ogpt

.dockerignore

@@ -3,11 +3,15 @@
 .dockerignore
 .pytest_cache
 .cache
+.local
 .github
 .nv
 .benchmarks
+.bash_history
+.gitignore
 h2ogpt.egg-info
 venv
 build
 dist
+prebuilt_deps
 Dockerfile

.env

@@ -0,0 +1,17 @@
+
+# H2OGPT
+
+H2OGPT_PORT=7860
+H2OGPT_RUNTIME=gcr.io/vorvan/h2oai/h2ogpt-runtime:0.1.0
+H2OGPT_BASE_MODEL=h2oai/h2ogpt-4096-llama2-7b-chat
+H2OGPT_ARGS="/workspace/generate.py --base_model=${H2OGPT_BASE_MODEL} --use_safetensors=True --prompt_type=llama2 --save_dir=/workspace/save/ --use_gpu_id=False --score_model=None --max_max_new_tokens=2048 --max_new_tokens=1024"
+
+# VLLM
+
+VLLM_TOKENIZER=hf-internal-testing/llama-tokenizer
+H2OGPT_VLLM_ARGS="--model=${H2OGPT_BASE_MODEL} --tokenizer=${VLLM_TOKENIZER} --tensor-parallel-size=2 --seed=1234 --trust-remote-code --download-dir=/workspace/.cache/huggingface/hub"
+
+# CPU models
+
+MODEL_PATH_LLAMA=https://huggingface.co/TheBloke/Llama-2-7b-Chat-GGUF/resolve/main/llama-2-7b-chat.Q6_K.gguf
+H2OGPT_CPU_ARGS="/workspace/generate.py --base_model=llama --model_path_llama=${MODEL_PATH_LLAMA} --max_seq_len=4096"

.github/workflows/python-package-publish.yml

@@ -30,16 +30,16 @@ jobs:
           run: |
             python3.10 -m pip install --upgrade pip
             python3.10 -m pip install setuptools wheel twine --upgrade
-
-        - name: Build Wheel
-          run: make clean dist
         
         - name: Modify Version
           if: ${{ inputs.version != ''}}
           run: |
             echo ${{ inputs.version}} > version.txt
             echo "h2ogpt-wheel-version  = $(cat version.txt)"
         
+        - name: Build Wheel
+          run: make clean dist
+
         - name: Publish to Test-PyPI
           if: ${{ inputs.pypi-index == 'Test-PyPI' }}
           run: |

.gitignore

@@ -38,3 +38,7 @@ Dockerfile_deps
 
 # virtual envs
 venv
+
+# Mac one click installer
+Tesseract-OCR/
+poppler/

Dockerfile

@@ -1,90 +1,34 @@
 # devel needed for bitsandbytes requirement of libcudart.so, otherwise runtime sufficient
 FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu20.04
 
-ARG DEBIAN_FRONTEND=noninteractive
-
-RUN apt-get update && apt-get install -y \
-    git \
-    curl \
-    wget \
-    software-properties-common \
-    pandoc
+ENV DEBIAN_FRONTEND=noninteractive
 
 ENV PATH="/h2ogpt_conda/bin:${PATH}"
 ARG PATH="/h2ogpt_conda/bin:${PATH}"
 
-RUN wget https://repo.anaconda.com/miniconda/Miniconda3-py310_23.1.0-1-Linux-x86_64.sh && \
-    mkdir -p h2ogpt_conda && \
-    bash ./Miniconda3-py310_23.1.0-1-Linux-x86_64.sh -b -u -p /h2ogpt_conda && \
-    conda install python=3.10 pygobject weasyprint -c conda-forge -y
-
-WORKDIR /workspace
-
-RUN apt-get install -y libmagic-dev poppler-utils tesseract-ocr libtesseract-dev libreoffice autoconf libtool
-
-COPY requirements.txt requirements.txt
-COPY reqs_optional reqs_optional
-
-RUN python3.10 -m pip install -r requirements.txt --extra-index-url https://download.pytorch.org/whl/cu118
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_langchain.txt --extra-index-url https://download.pytorch.org/whl/cu118
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_gpt4all.txt --extra-index-url https://download.pytorch.org/whl/cu118
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt --extra-index-url https://download.pytorch.org/whl/cu118
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_langchain.urls.txt --extra-index-url https://download.pytorch.org/whl/cu118
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_faiss_cpu.txt --extra-index-url https://download.pytorch.org/whl/cu118
-
-RUN python3.10 -m pip install -r reqs_optional/requirements_optional_doctr.txt --extra-index-url https://download.pytorch.org/whl/cu118
-# go back to older onnx so Tesseract OCR still works
-RUN python3.10 -m pip install onnxruntime==1.15.0 onnxruntime-gpu==1.15.0 --extra-index-url https://download.pytorch.org/whl/cu118 && \
-    python3.10 -m pip uninstall -y weasyprint && \
-    python3.10 -m pip install weasyprint
-
+ENV HOME=/workspace
 ENV CUDA_HOME=/usr/local/cuda-11.8
-
-# Install prebuilt dependencies
-
-RUN python3.10 -m nltk.downloader all
-RUN python3.10 -m pip install https://github.com/PanQiWei/AutoGPTQ/releases/download/v0.4.2/auto_gptq-0.4.2+cu118-cp310-cp310-linux_x86_64.whl
-RUN python3.10 -m pip install https://github.com/jllllll/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.1.73+cu118-cp310-cp310-linux_x86_64.whl
-RUN python3.10 -m pip install https://github.com/jllllll/exllama/releases/download/0.0.13/exllama-0.0.13+cu118-cp310-cp310-linux_x86_64.whl --no-cache-dir
-RUN playwright install --with-deps
-
-COPY . .
-
 ENV VLLM_CACHE=/workspace/.vllm_cache
+ENV TIKTOKEN_CACHE_DIR=/workspace/tiktoken_cache
 
-RUN sp=`python3.10 -c 'import site; print(site.getsitepackages()[0])'` && sed -i 's/posthog\.capture/return\n            posthog.capture/' $sp/chromadb/telemetry/posthog.py
-# RUN sp=`python3.10 -c 'import site; print(site.getsitepackages()[0])'` && \
-#     sed -i 's/posthog\.capture/return\n            posthog.capture/' $sp/chromadb/telemetry/posthog.py && \
-#     cd $sp && \
-#     rm -rf openai_vllm* && \
-#     cp -a openai openai_vllm && \
-#     cp -a openai-0.27.8.dist-info openai_vllm-0.27.8.dist-info && \
-#     find openai_vllm -name '*.py' | xargs sed -i 's/from openai /from openai_vllm /g' && \
-#     find openai_vllm -name '*.py' | xargs sed -i 's/openai\./openai_vllm./g' && \
-#     find openai_vllm -name '*.py' | xargs sed -i 's/from openai\./from openai_vllm./g' && \
-#     find openai_vllm -name '*.py' | xargs sed -i 's/import openai/import openai_vllm/g' && \
-#     conda create -n vllm python=3.10 -y && \
-#     /h2ogpt_conda/envs/vllm/bin/python3.10 -m pip install vllm ray pandas --extra-index-url https://download.pytorch.org/whl/cu118 && \
-#     mkdir ${VLLM_CACHE}
+WORKDIR /workspace
 
-EXPOSE 8888
-EXPOSE 7860
-EXPOSE 5000
+COPY . /workspace/
 
-# /workspace/.cache is the equivalent to ~/.cache
-ENV HOME=/workspace
+RUN cd /workspace && ./docker_build_script_ubuntu.sh
 
-COPY build_info.txt* /build_info.txt
-RUN touch /build_info.txt
+RUN chmod -R a+rwx /workspace
 
 ARG user=h2ogpt
 ARG group=h2ogpt
 ARG uid=1000
 ARG gid=1000
 
 RUN groupadd -g ${gid} ${group} && useradd -u ${uid} -g ${group} -s /bin/bash ${user}
-RUN chmod -R a+rwx /workspace
-RUN chmod -R a+rwx /h2ogpt_conda
+
+EXPOSE 8888
+EXPOSE 7860
+EXPOSE 5000
 
 USER h2ogpt
 

Makefile

@@ -1,8 +1,7 @@
 all: clean dist
 
 PACKAGE_VERSION       := `cat version.txt | tr -d '\n'`
-BUILD_TAG_FILES       := requirements.txt Dockerfile `ls reqs_optional/*.txt | sort`
-BUILD_TAG             := $(shell md5sum $(BUILD_TAG_FILES) 2> /dev/null | sort | md5sum | cut -d' ' -f1)
+BUILD_TAG             := $(shell git describe --always --dirty)
 DOCKER_TEST_IMAGE     := harbor.h2o.ai/h2ogpt/test-image:$(BUILD_TAG)
 PYTHON_BINARY         ?= `which python`
 DEFAULT_MARKERS       ?= "not need_tokens and not need_gpu"
@@ -47,18 +46,44 @@ build_info.txt:
 
 # Deprecated for now, no 0.4.1 on pypi, use release binary wheel that has no CUDA errors anymore
 docker_build_deps:
-	@rm -rf Dockerfile_deps
-	@sed '/# Install prebuilt dependencies/,$$d' Dockerfile > Dockerfile_deps
-	@docker build -t h2ogpt-deps-builder -f Dockerfile_deps .
+	@cp docker_build_script_ubuntu.sh docker_build_script_ubuntu.sh.back
+	@sed -i '/# Install prebuilt dependencies/,$$d' docker_build_script_ubuntu.sh
+	@docker build -t h2ogpt-deps-builder -f Dockerfile .
+	@mv docker_build_script_ubuntu.sh.back docker_build_script_ubuntu.sh
+	@mkdir -p prebuilt_deps
 	@docker run \
-		--rm -it --entrypoint bash --runtime nvidia -v `pwd`:/dot \
-		h2ogpt-deps-builder -c " \
+		--rm \
+		-it \
+		--entrypoint bash \
+		--runtime nvidia \
+		-v `pwd`:/dot \
+		-v /etc/passwd:/etc/passwd:ro \
+		-v /etc/group:/etc/group:ro \
+		-u `id -u`:`id -g` \
+		h2ogpt-deps-builder  -c " \
 			mkdir -p /dot/prebuilt_deps && cd /dot/prebuilt_deps && \
 			GITHUB_ACTIONS=true python3.10 -m pip install auto-gptq==0.4.2 --no-cache-dir --use-deprecated=legacy-resolver && \
 			python3.10 -m pip wheel auto-gptq==0.4.2 \
 		"
-	s3cmd put prebuilt_deps/auto_gptq-0.4.2-cp310-cp310-linux_x86_64.whl s3://artifacts.h2o.ai/deps/h2ogpt/ && \
-	s3cmd setacl s3://artifacts.h2o.ai/deps/h2ogpt/auto_gptq-0.4.2-cp310-cp310-linux_x86_64.whl --acl-public
+	@docker run \
+		--rm \
+		-it \
+		--entrypoint bash \
+		-v `pwd`:/dot \
+		quay.io/pypa/manylinux2014_x86_64 -c " \
+			ln -s /usr/local/bin/python3.10 /usr/local/bin/python3 && cd /tmp && \
+			git clone https://github.com/h2oai/duckdb.git && \
+			cd duckdb && \
+			git checkout dcd8c1ffc53dd020623630efb99ba6a3a4cbc5ad && \
+			BUILD_PYTHON=1 make release && \
+			cd tools/pythonpkg  && \
+			python3.10 setup.py bdist_wheel  && \
+			cp dist/duckdb-0.*.whl /dot/prebuilt_deps \
+		"
+	s3cmd put prebuilt_deps/auto_gptq-0.4.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl s3://artifacts.h2o.ai/deps/h2ogpt/ && \
+	s3cmd setacl s3://artifacts.h2o.ai/deps/h2ogpt/auto_gptq-0.4.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl --acl-public
+	s3cmd put prebuilt_deps/duckdb-0.8.2.dev4026+gdcd8c1ffc5-cp310-cp310-linux_x86_64.whl s3://artifacts.h2o.ai/deps/h2ogpt/ && \
+	s3cmd setacl s3://artifacts.h2o.ai/deps/h2ogpt/duckdb-0.8.2.dev4026+gdcd8c1ffc5-cp310-cp310-linux_x86_64.whl --acl-public
 
 docker_build: build_info.txt
 ifeq ($(shell curl --connect-timeout 4 --write-out %{http_code} -sS --output /dev/null -X GET http://harbor.h2o.ai/api/v2.0/projects/h2ogpt/repositories/test-image/artifacts/$(BUILD_TAG)/tags),200)
@@ -68,6 +93,9 @@ else
 	docker push $(DOCKER_TEST_IMAGE)
 endif
 
+just_docker_build: build_info.txt
+	DOCKER_BUILDKIT=1 docker build -t $(DOCKER_TEST_IMAGE) -f Dockerfile .
+
 docker_build_runner: docker_build
 	-docker pull $(DOCKER_TEST_IMAGE)
 	docker tag $(DOCKER_TEST_IMAGE) gcr.io/vorvan/h2oai/h2ogpt-runtime:$(BUILD_TAG)