H2ogpte

Dockerfile

@@ -1,46 +1,128 @@
-# devel needed for bitsandbytes requirement of libcudart.so, otherwise runtime sufficient
-FROM nvidia/cuda:12.1.1-cudnn8-devel-ubuntu20.04
+FROM gcr.io/vorvan/h2oai/h2ogpt-oss-wolfi-base:4 AS base-stage
 
-ENV DEBIAN_FRONTEND=noninteractive
-
-ENV PATH="/h2ogpt_conda/envs/h2ogpt/bin:${PATH}"
-ARG PATH="/h2ogpt_conda/envs/h2ogpt/bin:${PATH}"
+USER root
 
 ENV HOME=/workspace
 ENV CUDA_HOME=/usr/local/cuda-12.1
 ENV VLLM_CACHE=/workspace/.vllm_cache
 ENV TIKTOKEN_CACHE_DIR=/workspace/tiktoken_cache
+ENV NUMBA_CACHE_DIR=/workspace/numba_cache
 ENV HF_HUB_ENABLE_HF_TRANSFER=1
 
 WORKDIR /workspace
 
-COPY . /workspace/
+FROM base-stage as intermediate-stage
+
+## copy code
+COPY .              /workspace/
 
+# copy build info
 COPY build_info.txt /workspace/
 
-RUN cd /workspace && ./docker_build_script_ubuntu.sh
+# copy install script
+COPY linux_install_wolfi.sh    /workspace/
+
+# run setup
+RUN cd /workspace && ./linux_install_wolfi.sh
+
+# mv to separate locations so can copy from under a enw docker layer
+RUN \
+    mkdir -p /docker_cache && \
+    mv /usr/lib/python3.10/site-packages/nvidia         /docker_cache/nvidia && \
+    mv /usr/lib/python3.10/site-packages/torch          /docker_cache/torch && \
+    mv /usr/lib/python3.10/site-packages/onnxruntime    /docker_cache/onnxruntime && \
+    mv /usr/lib/python3.10/site-packages/triton         /docker_cache/triton && \
+    mv /usr/lib/python3.10                              /docker_cache/python_data && \
+    cp -R /usr                                          /docker_cache/user_data
+
+# remove since already in base image and didn't change
+RUN \
+    rm -rf /docker_cache/user_data/lib && \
+    rm -rf /docker_cache/user_data/libexec && \
+    rm -rf /docker_cache/user_data/local && \
+    rm -rf /docker_cache/user_data/bin/pandoc && \
+    rm -rf /docker_cache/user_data/bin/node && \
+    rm -rf /docker_cache/user_data/bin/lto-dump-11 && \
+    rm -rf /docker_cache/user_data/bin/lto-dump && \
+    rm -rf /docker_cache/user_data/share/misc/magic.mgc && \
+    rm -rf /docker_cache/user_data/share/icu && \
+    rm -rf /docker_cache/user_data/x86_64-pc-linux-gnu && \
+    rm -rf /docker_cache/python_data/site-packages/future/backports/test
 
+# cleanup
+RUN rm -rf /workspace/.cache && \
+    rm -rf /workspace/spaces && \
+    rm -rf /workspace/benchmarks && \
+    rm -rf /workspace/data && \
+    rm -rf /workspace/cloud && \
+    rm -rf /workspace/docs && \
+    rm -rf /workspace/helm && \
+    rm -rf /workspace/notebooks && \
+    rm -rf /workspace/papers
+
+RUN mkdir -p /workspace/save
+
+# make main workspace writable
 RUN chmod -R a+rwx /workspace
 
-ARG user=h2ogpt
-ARG group=h2ogpt
-ARG uid=1000
-ARG gid=1000
+FROM base-stage as final-stage
+
+COPY --from=intermediate-stage    /docker_cache/user_data         /usr
+COPY --from=intermediate-stage    /docker_cache/python_data/      /usr/lib/python3.10/
+COPY --from=intermediate-stage    /docker_cache/nvidia/           /usr/lib/python3.10/site-packages/nvidia/
+COPY --from=intermediate-stage    /docker_cache/torch/            /usr/lib/python3.10/site-packages/torch/
+COPY --from=intermediate-stage    /docker_cache/onnxruntime/      /usr/lib/python3.10/site-packages/onnxruntime/
+COPY --from=intermediate-stage    /docker_cache/triton/           /usr/lib/python3.10/site-packages/triton/
 
-RUN groupadd -g ${gid} ${group} && useradd -u ${uid} -g ${group} -s /bin/bash ${user}
-# already exists in base image
-# RUN groupadd -g ${gid} docker && useradd -u ${uid} -g ${group} -m ${user}
+COPY --from=intermediate-stage    /workspace/build_info.txt       /build_info.txt
+COPY --from=intermediate-stage    /workspace                      /workspace
+RUN chmod a+rwx /workspace  # only for top dir, as docker COPY skips it.
 
-# Add the user to the docker group
-RUN usermod -aG docker ${user}
+# Install Node.js, npm, and other necessary dependencies
+RUN apk add --no-cache  \
+    nodejs \
+    npm \
+    fontconfig \
+    git \
+    curl \
+    wget \
+    pandoc \
+    vim \
+    poppler-utils \
+    autoconf \
+    libtool \
+    docker \
+    zip \
+    unzip \
+    htop \
+    tree \
+    tmux \
+    jq \
+    net-tools \
+    nmap \
+    mtr \
+    rsync \
+    build-base \
+    parallel \
+    bc \
+    cronie \
+    screen \
+    inotify-tools \
+    xmlstarlet \
+    openssh
 
-# Switch to the new user
-USER ${user}
+# Install Mermaid CLI globally
+RUN npm install -g @mermaid-js/mermaid-cli
+
+# Verify Mermaid CLI installation
+RUN mmdc --version
 
 EXPOSE 8888
 EXPOSE 7860
 EXPOSE 5000
 EXPOSE 5002
 EXPOSE 5004
 
+USER h2ogpt
+
 ENTRYPOINT ["python3.10"]

docs/linux_install.sh

@@ -44,10 +44,9 @@ else
 fi
 # Optional: PyMuPDF/ArXiv:
 #   Note!! that pymupdf is AGPL, requiring any source code be made available, but it's like GPL and too strong a constraint for general commercial use.
-if [ "${GPLOK}" -eq "1" ]
-then
-    pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt -c reqs_optional/reqs_constraints.txt
-fi
+# h2oGPTe has license:
+pip install -r reqs_optional/requirements_optional_langchain.gpllike.txt -c reqs_optional/reqs_constraints.txt
+
 # Optional: FAISS
 pip install -r reqs_optional/requirements_optional_gpu_only.txt -c reqs_optional/reqs_constraints.txt
 # Optional: Selenium/PlayWright:
@@ -279,3 +278,8 @@ if [[ -z "${WOLFI_OS}" ]]; then
       # pick version 11
   fi
 fi
+
+if [[ -n "${WOLFI_OS}" ]]; then
+ # mitigate CVE-2024-22195, CVE-2024-34064
+ pip install jinja2==3.1.4 --no-cache-dir -c reqs_optional/reqs_constraints.txt
+fi

helm/h2ogpt-chart/templates/deployment.yaml

@@ -94,7 +94,7 @@ spec:
           image: "{{ .Values.vllm.image.repository }}:{{ .Values.vllm.image.tag | default .Chart.AppVersion }}"
           imagePullPolicy: {{ .Values.vllm.image.pullPolicy }}
           command: ["python3"]
-          args: 
+          args:
             - "-m" 
             - "vllm.entrypoints.openai.api_server"
             - "--port"
@@ -631,7 +631,7 @@ spec:
           image: "{{ .Values.vllm.image.repository }}:{{ .Values.vllm.image.tag | default .Chart.AppVersion }}"
           imagePullPolicy: {{ .Values.vllm.image.pullPolicy }}
           command: ["python3"]
-          args: 
+          args:
             - "-m" 
             - "vllm.entrypoints.openai.api_server"
             - "--port"

helm/h2ogpt-chart/values.yaml

@@ -10,7 +10,7 @@ h2ogpt:
   replicaCount: 1
   imagePullSecrets: 
   image:
-    repository: gcr.io/vorvan/h2oai/h2ogpt-runtime
+    repository: gcr.io/vorvan/h2oai/h2oai-h2ogpt-runtime
     tag: 
     pullPolicy: IfNotPresent
   initImage:
@@ -183,8 +183,10 @@ vllm:
   replicaCount: 1
 
   image:
-    repository: vllm/vllm-openai
-    tag: latest
+    repository: gcr.io/vorvan/h2oai/h2oai-h2ogpt-vllm
+    tag:
+    #repository: vllm/vllm-openai
+    #tag: latest
     pullPolicy: IfNotPresent
 
   podAffinity:

linux_install_wolfi.sh

@@ -0,0 +1,49 @@
+#!/bin/bash
+set -o pipefail
+set -ex
+
+export WOLFI_OS=true
+
+unset LLAMA_CUBLAS
+unset CMAKE_ARGS
+unset FORCE_CMAKE
+
+export GGML_CUDA=1
+export CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=all -DCMAKE_CUDA_COMPILER=/usr/local/cuda-12.1/bin/nvcc"
+export FORCE_CMAKE=1
+
+bash ./docs/linux_install.sh
+
+# setup tiktoken cache
+python3.10 -c "
+import tiktoken
+from tiktoken_ext import openai_public
+# FakeTokenizer etc. needs tiktoken for general tasks
+for enc in openai_public.ENCODING_CONSTRUCTORS:
+    encoding = tiktoken.get_encoding(enc)
+model_encodings = [
+    'gpt-4',
+    'gpt-4-0314',
+    'gpt-4-32k',
+    'gpt-4-32k-0314',
+    'gpt-3.5-turbo',
+    'gpt-3.5-turbo-16k',
+    'gpt-3.5-turbo-0301',
+    'text-ada-001',
+    'ada',
+    'text-babbage-001',
+    'babbage',
+    'text-curie-001',
+    'curie',
+    'davinci',
+    'text-davinci-003',
+    'text-davinci-002',
+    'code-davinci-002',
+    'code-davinci-001',
+    'code-cushman-002',
+    'code-cushman-001'
+]
+for enc in model_encodings:
+    encoding = tiktoken.encoding_for_model(enc)
+print('Done!')
+"

openai_server/agent_tools/download_youtube_video.py

@@ -0,0 +1,30 @@
+import argparse
+import os
+
+
+def download_video(url, output_dir='.'):
+    # Ensure the output directory exists
+    os.makedirs(output_dir, exist_ok=True)
+
+    ydl_opts = {
+        'format': 'mp4',
+        'outtmpl': os.path.join(output_dir, '%(title)s.%(ext)s'),
+        'restrictfilenames': True,
+    }
+
+    import yt_dlp
+    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+        ydl.download([url])
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Download a video from a given URL (e.g. https://www.youtube.com/watch?v=2Njmx-UuU3M)")
+    parser.add_argument("--url", type=str, required=True, help="The URL of the video to download")
+    parser.add_argument("--output", type=str, default=".", help="The directory to save the downloaded video")
+    args = parser.parse_args()
+
+    download_video(url=args.url, output_dir=args.output)
+
+
+if __name__ == "__main__":
+    main()

reqs_optional/reqs_constraints.txt

@@ -1,9 +1,18 @@
 # ensure doesn't drift, e.g. Issue #1348
 torch==2.2.1; sys_platform != "darwin" and platform_machine != "arm64"
 torch==2.3.1; sys_platform == "darwin" and platform_machine == "arm64"
-gradio==4.44.0
-gradio_client==1.3.0
+
+gradio==3.50.2
+gradio_client==0.6.1
+
+# gradio==4.26.0
+# gradio_client==0.15.1
+
+# gradio==4.44.0
+# gradio_client==1.3.0
+
 transformers>=4.45.1
+
 # https://github.com/langchain-ai/langchain/issues/22972
 tenacity==8.3.0
 pydantic==2.7.0

reqs_optional/requirements_optional_langchain.txt

@@ -97,7 +97,7 @@ weaviate-client==3.26.2
 # vllm==0.2.2
 
 # only gradio>=4
-gradio_pdf>=0.0.7
+#gradio_pdf>=0.0.7
 
 gradio_tools>=0.0.9
 

requirements.txt

@@ -1,16 +1,20 @@
 # no websockets, more cloud friendly
 # able to make gradio clean-up states
 
+gradio_clean @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio_clean-3.50.2-py3-none-any.whl
+gradio_client==0.6.1
+
 # gradio @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio-4.25.0-py3-none-any.whl
 # gradio_client @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio_client-0.15.0-py3-none-any.whl
+
+# gradio @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio-4.20.1-py3-none-any.whl
+# gradio_client==0.11.0
+
 #gradio @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio-4.20.1-py3-none-any.whl
 #gradio_client==0.11.0
 # gradio @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio-4.26.0-py3-none-any.whl
 # gradio_client @ https://h2o-release.s3.amazonaws.com/h2ogpt/gradio_client-0.15.1-py3-none-any.whl
 
-gradio==4.44.0
-gradio_client==1.3.0
-
 uvicorn[standard]
 gunicorn
 fastapi-utils

src/version.py

@@ -1 +1 @@
-__version__ = "7a8e0c5236f6b39949223b70441212930953146e"
+__version__ = "d7b0fe7efd0f587deef882e3dab1e9bd3f72de2a"