fixes for new dockerfile

.dockerignore

@@ -0,0 +1 @@
+seamless_server/models/*.pt

.gitattributes

@@ -34,3 +34,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 *.mp3 filter=lfs diff=lfs merge=lfs -text
+*.whl filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -1,2 +1,5 @@
 .vscode/settings.json
 __pycache__/
+*.pt
+*.model
+venv/

Dockerfile

@@ -1,7 +1,25 @@
-# TODO: This doesn't work, copied over from M4T but needs an update
+# build frontend with node
+FROM node:20-alpine AS frontend
+RUN apk add --no-cache libc6-compat
+WORKDIR /app
+
+COPY streaming-react-app .
+RUN \
+    if [ -f yarn.lock ]; then yarn --frozen-lockfile; \
+    elif [ -f package-lock.json ]; then npm ci; \
+    elif [ -f pnpm-lock.yaml ]; then yarn global add pnpm && pnpm i --frozen-lockfile; \
+    else echo "Lockfile not found." && exit 1; \
+    fi
+
+RUN npm run build
+
+# build backend on CUDA 
+FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 AS backend
+WORKDIR /app
 
-FROM nvidia/cuda:11.7.1-cudnn8-devel-ubuntu22.04
 ENV DEBIAN_FRONTEND=noninteractive
+ENV NODE_MAJOR=20
+
 RUN apt-get update && \
     apt-get upgrade -y && \
     apt-get install -y --no-install-recommends \
@@ -23,6 +41,7 @@ RUN apt-get update && \
     libxmlsec1-dev \
     libffi-dev \
     liblzma-dev \
+    sox libsox-fmt-all \
     # gradio dependencies \
     ffmpeg \
     # fairseq2 dependencies \
@@ -33,30 +52,39 @@ RUN apt-get update && \
 RUN useradd -m -u 1000 user
 USER user
 ENV HOME=/home/user \
-    PATH=/home/user/.local/bin:${PATH}
-WORKDIR ${HOME}/app
+    PATH=/home/user/.local/bin:$PATH
+WORKDIR $HOME/app
 
 RUN curl https://pyenv.run | bash
-ENV PATH=${HOME}/.pyenv/shims:${HOME}/.pyenv/bin:${PATH}
+ENV PATH=$HOME/.pyenv/shims:$HOME/.pyenv/bin:$PATH
 ARG PYTHON_VERSION=3.10.12
-RUN pyenv install ${PYTHON_VERSION} && \
-    pyenv global ${PYTHON_VERSION} && \
+RUN pyenv install $PYTHON_VERSION && \
+    pyenv global $PYTHON_VERSION && \
     pyenv rehash && \
     pip install --no-cache-dir -U pip setuptools wheel
 
-COPY --chown=1000 ./requirements.txt /tmp/requirements.txt
-RUN pip install --no-cache-dir --upgrade -r /tmp/requirements.txt
-
-COPY --chown=1000 . ${HOME}/app
-ENV PYTHONPATH=${HOME}/app \
-    PYTHONUNBUFFERED=1 \
-    GRADIO_ALLOW_FLAGGING=never \
-    GRADIO_NUM_PORTS=1 \
-    GRADIO_SERVER_NAME=0.0.0.0 \
-    GRADIO_THEME=huggingface \
-    SYSTEM=spaces
-
-# gradio instead of python for reload on file save with mountin pwd volume:
-# docker run -p 7860:7860 -v $(pwd):/home/user/app seamless_m4t_text
-CMD ["gradio", "app.py"]
-# CMD ["python", "app.py"]
+COPY --chown=user:user ./seamless_server ./seamless_server
+# change dir since pip needs to seed whl folder
+RUN cd seamless_server && pip install --no-cache-dir --upgrade -r requirements.txt
+COPY --from=frontend /app/dist ./streaming-react-app/dist
+
+WORKDIR $HOME/app/seamless_server
+# temporary hack to link SeamlessStreaming models
+ARG HF_TOKEN
+RUN huggingface-cli download meta-private/SeamlessStreaming seamless_streaming_unity.pt spm_char_lang38_tc.model seamless_streaming_monotonic_decoder.pt --local-dir ./models/ && \
+    huggingface-cli download meta-private/SeamlessExpressive pretssel_melhifigan_wm-final.pt  --local-dir ./models/ 
+USER root
+RUN mkdir -p /large_experiments/seamless/ust/krs/fairseq2_checkpoints/ && \
+    mkdir -p /large_experiments/seamless/workstream/expressivity/oss/checkpoints/ && \
+    mkdir -p /checkpoint/krs/unity2/ && \
+    chown -R user:user /large_experiments/ && \
+    chown -R user:user /checkpoint/ && \
+    ln -s $(readlink -f models/seamless_streaming_unity.pt) /large_experiments/seamless/ust/krs/fairseq2_checkpoints/seamless_streaming_unity.pt && \
+    ln -s $(readlink -f models/seamless_streaming_monotonic_decoder.pt) /large_experiments/seamless/ust/krs/fairseq2_checkpoints/seamless_streaming_monotonic_decoder.pt && \
+    ln -s $(readlink -f models/pretssel_melhifigan_wm-final.pt) /large_experiments/seamless/workstream/expressivity/oss/checkpoints/pretssel_melhifigan_wm-final.pt && \
+    ln -s $(readlink -f models/spm_char_lang38_tc.model) /checkpoint/krs/unity2/
+
+USER user
+CMD [ "uvicorn", "app_pubsub:app", "--host", "0.0.0.0", "--port", "7860" ]
+
+

README.md

@@ -1,5 +1,5 @@
 ---
-title: Seamless M4T
+title: Seamless Streaming Backend/Frontend
 emoji: 📞
 colorFrom: blue
 colorTo: yellow

seamless_server/app_pubsub.py

@@ -12,6 +12,10 @@ import sys
 import time
 import random
 import string
+from starlette.applications import Starlette
+from starlette.routing import Mount, Route
+from starlette.staticfiles import StaticFiles
+
 
 from src.room import Room, Member
 from src.simuleval_agent_directory import NoAvailableAgentException
@@ -86,7 +90,15 @@ sio = socketio.AsyncServer(
     # engineio_logger=logger,
 )
 # sio.logger.setLevel(logging.DEBUG)
-app = socketio.ASGIApp(sio, static_files=static_files)
+socketio_app = socketio.ASGIApp(sio)
+
+app_routes = [
+    Mount("/ws", app=socketio_app),  # Mount Socket.IO server under /app
+    Mount(
+        "/", app=StaticFiles(directory=CLIENT_BUILD_PATH, html=True)
+    ),  # Serve static files from root
+]
+app = Starlette(debug=True, routes=app_routes)
 
 # rooms is indexed by room_id
 rooms: Dict[str, Room] = {}

seamless_server/requirements.txt

@@ -1,5 +1,7 @@
-# fairseq2
+--pre --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/nightly/pt2.1.1/cu118
+git+https://github.com/facebookresearch/SimulEval.git
 # seamless_communication
+./whl/seamless_communication-1.0.0-py3-none-any.whl
 Flask==2.1.3
 Flask_Sockets==0.2.1
 g2p_en==2.1.0
@@ -12,7 +14,7 @@ protobuf==4.24.2
 psola==0.0.1
 pydub==0.25.1
 silero==0.4.1
-simuleval==1.1.1
+# simuleval==1.1.1
 soundfile==0.11.0
 stable_ts==1.4.0
 torch  # specific torch version depends on fairseq2 installation
@@ -23,3 +25,6 @@ python-socketio==5.9.0
 uvicorn[standard]==0.23.2
 parallel-wavegan==0.5.5
 python-jose[cryptography]==3.3.0
+starlette==0.32.0.post1 
+hf_transfer==0.1.4
+huggingface_hub==0.19.4

seamless_server/whl/seamless_communication-1.0.0-py3-none-any.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3380dc7d6613c4dc9ef4d78fd3dcf1a50f7c6a659b8ba0b37ad5237533d002e
+size 234811

streaming-react-app/src/SocketWrapper.tsx

@@ -11,8 +11,9 @@ import {getURLParams} from './URLParams';
 
 // The time to wait before showing a "disconnected" screen upon initial app load
 const INITIAL_DISCONNECT_SCREEN_DELAY = 2000;
-const SERVER_URL_DEFAULT = "localhost:8000"
-
+const SERVER_URL_DEFAULT = `${window.location.protocol === "https:" ? "wss" : "ws"
+                    }://${window.location.host}`;
+                    
 export default function SocketWrapper({children}) {
   const [socket, setSocket] = useState<Socket | null>(null);
   const [connected, setConnected] = useState<boolean | null>(null);
@@ -63,6 +64,7 @@ export default function SocketWrapper({children}) {
       // want that because that'd mean awful performance. It'd be better for the app
       // to simply break in that case and not connect.
       transports: ['websocket'],
+      path: '/ws/socket.io'
     });
 
     const onServerID = (serverID: string) => {

streaming-react-app/vite.config.ts

@@ -1,4 +1,4 @@
-import {defineConfig} from 'vite';
+import { defineConfig } from 'vite';
 import react from '@vitejs/plugin-react';
 // import {resolve} from 'path';
 
@@ -7,10 +7,18 @@ import react from '@vitejs/plugin-react';
 // const typesDir = resolve(__dirname, 'types');
 
 // https://vitejs.dev/config/
-export default defineConfig(({command}) => {
+export default defineConfig(({ command }) => {
   let define = {};
   return {
     plugins: [react()],
     define: define,
-  };
+    server: {
+      proxy: {
+        '/ws': {
+          target: 'ws://localhost:7860',
+          ws: true
+        }
+      },
+    },
+  }
 });