Merge pull request #4773 from oobabooga/dev

Merge dev branch

.gitignore

@@ -26,9 +26,10 @@
 .DS_Store
 .eslintrc.js
 .idea
-.env
 .venv
 venv
+.envrc
+.direnv
 .vscode
 *.bak
 *.ipynb
@@ -40,3 +41,9 @@ package.json
 package-lock.json
 Thumbs.db
 wandb
+
+# ignore user docker config and top level links to docker files
+/docker-compose.yaml
+/docker-compose.yml
+/Dockerfile
+.env

README.md

@@ -163,9 +163,12 @@ The requirments*.txt above contain various precompiled wheels. If you wish to co
 ### Alternative: Docker
 
 ```
-ln -s docker/{Dockerfile,docker-compose.yml,.dockerignore} .
+ln -s docker/{nvidia/Dockerfile,docker-compose.yml,.dockerignore} .
 cp docker/.env.example .env
-# Edit .env and set TORCH_CUDA_ARCH_LIST based on your GPU model
+# Edit .env and set: 
+#   TORCH_CUDA_ARCH_LIST based on your GPU model
+#   APP_RUNTIME_GID      your host user's group id (run `id -g` in a terminal)
+#   BUILD_EXTENIONS      optionally add comma separated list of extensions to build
 docker compose up --build
 ```
 

docker/.env.example

@@ -2,19 +2,21 @@
 # however for me to work i had to specify the exact version for my card ( 2060 ) it was 7.5
 # https://developer.nvidia.com/cuda-gpus you can find the version for your card here
 TORCH_CUDA_ARCH_LIST=7.5
-
 # your command-line flags go here:
-CLI_ARGS=
+CLI_ARGS=--listen
-
 # the port the webui binds to on the host
 HOST_PORT=7860
 # the port the webui binds to inside the container
 CONTAINER_PORT=7860
-
 # the port the api binds to on the host
 HOST_API_PORT=5000
 # the port the api binds to inside the container
 CONTAINER_API_PORT=5000
-
+# Comma separated extensions to build
-# the version used to install text-generation-webui from
+BUILD_EXTENSIONS=""
-WEBUI_VERSION=HEAD
+# Set APP_RUNTIME_GID to an appropriate host system group to enable access to mounted volumes 
+# You can find your current host user group id with the command `id -g`
+APP_RUNTIME_GID=6972
+# override default app build permissions (handy for deploying to cloud)
+#APP_GID=6972
+#APP_UID=6972

docker/Dockerfile

@@ -1,77 +0,0 @@
-FROM nvidia/cuda:12.1.0-devel-ubuntu22.04 as builder
-
-RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,rw apt-get update && \
-    apt-get install --no-install-recommends -y git vim build-essential python3-dev python3-venv && \
-    rm -rf /var/lib/apt/lists/*
-
-RUN git clone --depth=1 https://github.com/oobabooga/GPTQ-for-LLaMa /build
-
-WORKDIR /build
-
-RUN --mount=type=cache,target=/root/.cache/pip,rw \
-    python3 -m venv /build/venv && \
-    . /build/venv/bin/activate && \
-    pip3 install --upgrade pip setuptools wheel ninja && \
-    pip3 install torch torchvision torchaudio xformers --index-url https://download.pytorch.org/whl/cu121 && \
-    pip3 install -r requirements.txt
-
-# https://developer.nvidia.com/cuda-gpus
-# for a rtx 2060: ARG TORCH_CUDA_ARCH_LIST="7.5"
-ARG TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5;5.0;6.0;6.1;7.0;7.5;8.0;8.6+PTX}"
-RUN . /build/venv/bin/activate && \
-    python3 setup_cuda.py bdist_wheel -d .
-
-FROM nvidia/cuda:12.1.0-runtime-ubuntu22.04
-
-LABEL maintainer="Your Name <your.email@example.com>"
-LABEL description="Docker image for GPTQ-for-LLaMa and Text Generation WebUI"
-
-RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,rw apt-get update && \
-    apt-get install --no-install-recommends -y python3-dev libportaudio2 libasound-dev git python3 python3-pip make g++ ffmpeg && \
-    rm -rf /var/lib/apt/lists/*
-
-RUN --mount=type=cache,target=/root/.cache/pip,rw pip3 install virtualenv
-
-RUN mkdir /app
-
-WORKDIR /app
-
-ARG WEBUI_VERSION
-RUN test -n "${WEBUI_VERSION}" && git reset --hard ${WEBUI_VERSION} || echo "Using provided webui source"
-
-# Create virtualenv
-RUN virtualenv /app/venv
-RUN --mount=type=cache,target=/root/.cache/pip,rw \
-    . /app/venv/bin/activate && \
-    pip3 install --upgrade pip setuptools wheel ninja && \
-    pip3 install torch xformers --index-url https://download.pytorch.org/whl/cu121 && \
-    pip3 install torchvision torchaudio sentence_transformers
-
-# Copy and install GPTQ-for-LLaMa
-COPY --from=builder /build /app/repositories/GPTQ-for-LLaMa
-RUN --mount=type=cache,target=/root/.cache/pip,rw \
-    . /app/venv/bin/activate && \
-    pip3 install /app/repositories/GPTQ-for-LLaMa/*.whl
-
-# Install main requirements
-COPY requirements.txt /app/requirements.txt
-RUN --mount=type=cache,target=/root/.cache/pip,rw \
-    . /app/venv/bin/activate && \
-    pip3 install -r requirements.txt
-
-COPY . /app/
-
-RUN cp /app/venv/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cuda121.so /app/venv/lib/python3.10/site-packages/bitsandbytes/libbitsandbytes_cpu.so
-
-# Install extension requirements
-RUN --mount=type=cache,target=/root/.cache/pip,rw \
-    . /app/venv/bin/activate && \
-    for ext in /app/extensions/*/requirements.txt; do \
-    cd "$(dirname "$ext")"; \
-    pip3 install -r requirements.txt; \
-    done
-
-ENV CLI_ARGS=""
-
-EXPOSE ${CONTAINER_PORT:-7860} ${CONTAINER_API_PORT:-5000}
-CMD . /app/venv/bin/activate && python3 server.py ${CLI_ARGS}

docker/docker-compose.yml

@@ -6,27 +6,30 @@ services:
       args:
         # specify which cuda version your card supports: https://developer.nvidia.com/cuda-gpus
         TORCH_CUDA_ARCH_LIST: ${TORCH_CUDA_ARCH_LIST:-7.5} 
-        WEBUI_VERSION: ${WEBUI_VERSION:-HEAD}
+        BUILD_EXTENSIONS: ${BUILD_EXTENSIONS:-}
+        APP_GID: ${APP_GID:-6972} 
+        APP_UID: ${APP_UID-6972} 
     env_file: .env
+    user: "${APP_RUNTIME_UID:-6972}:${APP_RUNTIME_GID:-6972}"
     ports:
       - "${HOST_PORT:-7860}:${CONTAINER_PORT:-7860}"
       - "${HOST_API_PORT:-5000}:${CONTAINER_API_PORT:-5000}"
     stdin_open: true
     tty: true
     volumes:
-      - ./characters:/app/characters
+      - ./characters:/home/app/text-generation-webui/characters
-      - ./extensions:/app/extensions
+      - ./extensions:/home/app/text-generation-webui/extensions
-      - ./loras:/app/loras
+      - ./loras:/home/app/text-generation-webui/loras
-      - ./models:/app/models
+      - ./models:/home/app/text-generation-webui/models
-      - ./presets:/app/presets
+      - ./presets:/home/app/text-generation-webui/presets
-      - ./prompts:/app/prompts
+      - ./prompts:/home/app/text-generation-webui/prompts
-      - ./softprompts:/app/softprompts
+      - ./softprompts:/home/app/text-generation-webui/softprompts
-      - ./training:/app/training
+      - ./training:/home/app/text-generation-webui/training
       - ./cloudflared:/etc/cloudflared
     deploy:
       resources:
         reservations:
           devices:
             - driver: nvidia
-              device_ids: ['0']
+              count: all
               capabilities: [gpu]

docker/nvidia/Dockerfile

@@ -0,0 +1,56 @@
+# BUILDER
+FROM nvidia/cuda:12.1.1-devel-ubuntu22.04 as builder
+WORKDIR /builder
+ARG TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5;5.0;6.0;6.1;7.0;7.5;8.0;8.6+PTX}"
+ARG BUILD_EXTENSIONS="${BUILD_EXTENSIONS:-}"
+ARG APP_UID="${APP_UID:-6972}"
+ARG APP_GID="${APP_GID:-6972}"
+# create / update build env
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,rw \
+    apt update && \
+    apt install --no-install-recommends -y git vim build-essential python3-dev pip && \
+    rm -rf /var/lib/apt/lists/*
+RUN --mount=type=cache,target=/root/.cache/pip,rw \
+    pip3 install --global --upgrade pip wheel setuptools && \
+    # make shared builder & runtime app user
+    addgroup --gid $APP_GID app_grp && \
+    useradd -m -u $APP_UID --gid app_grp app
+USER app:app_grp
+# build wheels for runtime
+WORKDIR /home/app/build
+COPY --chown=app:app_grp requirements.txt /home/app/build
+COPY --chown=app:app_grp extensions /home/app/build/extensions
+RUN --mount=type=cache,target=/root/.cache/pip,rw \
+    # build all requirements files as wheel dists
+    pip3 wheel -w wheels -r requirements.txt `echo "$BUILD_EXTENSIONS" | sed -r 's/([^,]+)\s*,?\s*/ -r \/home\/app\/build\/extensions\/\1\/requirements.txt/g'`
+    # drop wheel and setuptools .whl to avoid install issues
+RUN rm wheels/setuptools*.whl
+
+# RUNTIME
+FROM nvidia/cuda:12.1.1-runtime-ubuntu22.04
+ARG TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST:-3.5;5.0;6.0;6.1;7.0;7.5;8.0;8.6}"
+ARG APP_UID="${APP_UID:-6972}"
+ARG APP_GID="${APP_GID:-6972}"
+ENV CLI_ARGS=""
+# create / update runtime env
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,rw \
+    apt update && \
+    apt install --no-install-recommends -y git python3 pip && \
+    rm -rf /var/lib/apt/lists/* && \
+    pip3 install --global --no-cache --upgrade pip wheel setuptools && \
+    # make shared builder & runtime app user
+    addgroup --gid $APP_GID app_grp && \
+    useradd -m -u $APP_UID --gid app_grp app
+USER app:app_grp
+# install locally built wheels for app
+WORKDIR /home/app/wheels
+COPY --from=builder /home/app/build/wheels /home/app/wheels
+COPY --chown=app:app_grp . /home/app/text-generation-webui
+RUN umask 0002 && \
+    chmod g+rwX /home/app/text-generation-webui && \
+    pip3 install --global --no-build-isolation --no-cache --no-index ./*.whl && \
+    rm -r /home/app/wheels
+WORKDIR /home/app/text-generation-webui
+EXPOSE ${CONTAINER_PORT:-7860} ${CONTAINER_API_PORT:-5000} ${CONTAINER_API_STREAM_PORT:-5005}
+# set umask to ensure group read / write at runtime
+CMD umask 0002 && export HOME=/home/app && python3 server.py ${CLI_ARGS}

extensions/openai/script.py

@@ -364,7 +364,7 @@ def run_server():
 
         logger.info(f'OpenAI API key:\n\n{shared.args.api_key}\n')
 
-    if shared.args.admin_key:
+    if shared.args.admin_key and shared.args.admin_key != shared.args.api_key:
         logger.info(f'OpenAI API admin key (for loading/unloading models):\n\n{shared.args.admin_key}\n')
 
     uvicorn.run(app, host=server_addr, port=port, ssl_certfile=ssl_certfile, ssl_keyfile=ssl_keyfile)

models/config.yaml

@@ -182,3 +182,7 @@
   instruction_template: 'Alpaca'
 .*orca-2-(13|7)b:
   instruction_template: 'ChatML'
+.*openhermes.*mistral:
+  instruction_template: 'ChatML'
+.*Yi-34B-Chat:
+  instruction_template: 'ChatML'

modules/llamacpp_model.py

@@ -46,7 +46,7 @@ class LlamaCppModel:
         self.grammar = None
 
     def __del__(self):
-        self.model.__del__()
+        del self.model
 
     @classmethod
     def from_pretrained(self, path):
@@ -64,7 +64,8 @@ class LlamaCppModel:
             else:
                 cache_capacity = int(shared.args.cache_capacity)
 
-        logger.info("Cache capacity is " + str(cache_capacity) + " bytes")
+        if cache_capacity > 0:
+            logger.info("Cache capacity is " + str(cache_capacity) + " bytes")
 
         if shared.args.tensor_split is None or shared.args.tensor_split.strip() == '':
             tensor_split_list = None
@@ -118,9 +119,7 @@ class LlamaCppModel:
                 self.grammar = None
 
     def generate(self, prompt, state, callback=None):
-
         LogitsProcessorList = llama_cpp_lib().LogitsProcessorList
-
         prompt = prompt if type(prompt) is str else prompt.decode()
 
         # Handle truncation
@@ -163,6 +162,7 @@ class LlamaCppModel:
         for completion_chunk in completion_chunks:
             if shared.stop_everything:
                 break
+
             text = completion_chunk['choices'][0]['text']
             output += text
             if callback:

modules/loaders.py

@@ -27,68 +27,6 @@ loaders_and_params = OrderedDict({
         'disable_exllama',
         'transformers_info'
     ],
-    'ExLlama_HF': [
-        'gpu_split',
-        'max_seq_len',
-        'alpha_value',
-        'rope_freq_base',
-        'compress_pos_emb',
-        'cfg_cache',
-        'no_use_fast',
-        'exllama_HF_info',
-    ],
-    'ExLlamav2_HF': [
-        'gpu_split',
-        'max_seq_len',
-        'cfg_cache',
-        'no_flash_attn',
-        'cache_8bit',
-        'alpha_value',
-        'compress_pos_emb',
-        'no_use_fast',
-    ],
-    'ExLlama': [
-        'gpu_split',
-        'max_seq_len',
-        'alpha_value',
-        'rope_freq_base',
-        'compress_pos_emb',
-        'exllama_info',
-    ],
-    'ExLlamav2': [
-        'gpu_split',
-        'max_seq_len',
-        'no_flash_attn',
-        'cache_8bit',
-        'alpha_value',
-        'compress_pos_emb',
-    ],
-    'AutoGPTQ': [
-        'triton',
-        'no_inject_fused_attention',
-        'no_inject_fused_mlp',
-        'no_use_cuda_fp16',
-        'wbits',
-        'groupsize',
-        'desc_act',
-        'disable_exllama',
-        'gpu_memory',
-        'cpu_memory',
-        'cpu',
-        'disk',
-        'auto_devices',
-        'trust_remote_code',
-        'no_use_fast',
-        'autogptq_info',
-    ],
-    'GPTQ-for-LLaMa': [
-        'wbits',
-        'groupsize',
-        'model_type',
-        'pre_layer',
-        'no_use_fast',
-        'gptq_for_llama_info',
-    ],
     'llama.cpp': [
         'n_ctx',
         'n_gpu_layers',
@@ -125,14 +63,43 @@ loaders_and_params = OrderedDict({
         'logits_all',
         'llamacpp_HF_info',
     ],
-    'ctransformers': [
+    'ExLlama_HF': [
-        'n_ctx',
+        'gpu_split',
-        'n_gpu_layers',
+        'max_seq_len',
-        'n_batch',
+        'alpha_value',
-        'threads',
+        'rope_freq_base',
-        'model_type',
+        'compress_pos_emb',
-        'no_mmap',
+        'cfg_cache',
-        'mlock'
+        'no_use_fast',
+        'exllama_HF_info',
+    ],
+    'ExLlamav2_HF': [
+        'gpu_split',
+        'max_seq_len',
+        'cfg_cache',
+        'no_flash_attn',
+        'cache_8bit',
+        'alpha_value',
+        'compress_pos_emb',
+        'no_use_fast',
+    ],
+    'AutoGPTQ': [
+        'triton',
+        'no_inject_fused_attention',
+        'no_inject_fused_mlp',
+        'no_use_cuda_fp16',
+        'wbits',
+        'groupsize',
+        'desc_act',
+        'disable_exllama',
+        'gpu_memory',
+        'cpu_memory',
+        'cpu',
+        'disk',
+        'auto_devices',
+        'trust_remote_code',
+        'no_use_fast',
+        'autogptq_info',
     ],
     'AutoAWQ': [
         'cpu_memory',
@@ -142,7 +109,40 @@ loaders_and_params = OrderedDict({
         'no_inject_fused_attention',
         'trust_remote_code',
         'no_use_fast',
-    ]
+    ],
+    'GPTQ-for-LLaMa': [
+        'wbits',
+        'groupsize',
+        'model_type',
+        'pre_layer',
+        'no_use_fast',
+        'gptq_for_llama_info',
+    ],
+    'ExLlama': [
+        'gpu_split',
+        'max_seq_len',
+        'alpha_value',
+        'rope_freq_base',
+        'compress_pos_emb',
+        'exllama_info',
+    ],
+    'ExLlamav2': [
+        'gpu_split',
+        'max_seq_len',
+        'no_flash_attn',
+        'cache_8bit',
+        'alpha_value',
+        'compress_pos_emb',
+    ],
+    'ctransformers': [
+        'n_ctx',
+        'n_gpu_layers',
+        'n_batch',
+        'threads',
+        'model_type',
+        'no_mmap',
+        'mlock'
+    ],
 })
 
 loaders_samplers = {

modules/text_generation.py

@@ -78,10 +78,9 @@ def _generate_reply(question, state, stopping_strings=None, is_chat=False, escap
 
     # Generate
     for reply in generate_func(question, original_question, seed, state, stopping_strings, is_chat=is_chat):
+        reply, stop_found = apply_stopping_strings(reply, all_stop_strings)
         if escape_html:
             reply = html.escape(reply)
-
-        reply, stop_found = apply_stopping_strings(reply, all_stop_strings)
         if is_stream:
             cur_time = time.time()