feat(parler-tts): Add new backend

.github/workflows/test-extra.yml

@@ -104,6 +104,35 @@ jobs:
            make --jobs=5 --output-sync=target -C backend/python/diffusers
            make --jobs=5 --output-sync=target -C backend/python/diffusers test
 
+  tests-parler-tts:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Clone
+        uses: actions/checkout@v4
+        with: 
+          submodules: true
+      - name: Dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential ffmpeg
+          curl https://repo.anaconda.com/pkgs/misc/gpgkeys/anaconda.asc | gpg --dearmor > conda.gpg && \
+             sudo install -o root -g root -m 644 conda.gpg /usr/share/keyrings/conda-archive-keyring.gpg && \
+              gpg --keyring /usr/share/keyrings/conda-archive-keyring.gpg --no-default-keyring --fingerprint 34161F5BF5EB1D4BFBBB8F0A8AEB4F8B29D82806 && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" > /etc/apt/sources.list.d/conda.list' && \
+             sudo /bin/bash -c 'echo "deb [arch=amd64 signed-by=/usr/share/keyrings/conda-archive-keyring.gpg] https://repo.anaconda.com/pkgs/misc/debrepo/conda stable main" | tee -a /etc/apt/sources.list.d/conda.list' && \
+             sudo apt-get update && \
+             sudo apt-get install -y conda
+          sudo apt-get install -y ca-certificates cmake curl patch python3-pip
+          sudo apt-get install -y libopencv-dev
+          pip install --user grpcio-tools
+
+          sudo rm -rfv /usr/bin/conda || true
+
+      - name: Test parler-tts
+        run: |
+           export PATH=$PATH:/opt/conda/bin
+           make --jobs=5 --output-sync=target -C backend/python/parler-tts
+           make --jobs=5 --output-sync=target -C backend/python/parler-tts test
 
   tests-transformers-musicgen:
     runs-on: ubuntu-latest

Dockerfile

@@ -15,7 +15,7 @@ ARG TARGETVARIANT
 
 ENV BUILD_TYPE=${BUILD_TYPE}
 ENV DEBIAN_FRONTEND=noninteractive
-ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,petals:/build/backend/python/petals/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh"
+ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,huggingface-embeddings:/build/backend/python/sentencetransformers/run.sh,petals:/build/backend/python/petals/run.sh,transformers:/build/backend/python/transformers/run.sh,sentencetransformers:/build/backend/python/sentencetransformers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,exllama:/build/backend/python/exllama/run.sh,vall-e-x:/build/backend/python/vall-e-x/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,transformers-musicgen:/build/backend/python/transformers-musicgen/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
 
 ARG GO_TAGS="stablediffusion tinydream tts"
 
@@ -275,6 +275,9 @@ RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
     make -C backend/python/transformers-musicgen \
     ; fi
+RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
+    make -C backend/python/parler-tts \
+    ; fi
 RUN if [ "${IMAGE_TYPE}" = "extras" ]; then \
     make -C backend/python/coqui \
     ; fi

Makefile

@@ -439,10 +439,10 @@ protogen-go-clean:
 	$(RM) bin/*
 
 .PHONY: protogen-python
-protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama-protogen exllama2-protogen mamba-protogen petals-protogen sentencetransformers-protogen transformers-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen
+protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama-protogen exllama2-protogen mamba-protogen petals-protogen sentencetransformers-protogen transformers-protogen parler-tts-protogen transformers-musicgen-protogen vall-e-x-protogen vllm-protogen
 
 .PHONY: protogen-python-clean
-protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama-protogen-clean exllama2-protogen-clean mamba-protogen-clean petals-protogen-clean sentencetransformers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean
+protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama-protogen-clean exllama2-protogen-clean mamba-protogen-clean petals-protogen-clean sentencetransformers-protogen-clean transformers-protogen-clean transformers-musicgen-protogen-clean parler-tts-protogen-clean vall-e-x-protogen-clean vllm-protogen-clean
 
 .PHONY: autogptq-protogen
 autogptq-protogen:
@@ -524,6 +524,14 @@ transformers-protogen:
 transformers-protogen-clean:
 	$(MAKE) -C backend/python/transformers protogen-clean
 
+.PHONY: parler-tts-protogen
+parler-tts-protogen:
+	$(MAKE) -C backend/python/parler-tts protogen
+
+.PHONY: parler-tts-protogen-clean
+parler-tts-protogen-clean:
+	$(MAKE) -C backend/python/parler-tts protogen-clean
+
 .PHONY: transformers-musicgen-protogen
 transformers-musicgen-protogen:
 	$(MAKE) -C backend/python/transformers-musicgen protogen
@@ -560,6 +568,7 @@ prepare-extra-conda-environments: protogen-python
 	$(MAKE) -C backend/python/sentencetransformers
 	$(MAKE) -C backend/python/transformers
 	$(MAKE) -C backend/python/transformers-musicgen
+	$(MAKE) -C backend/python/parler-tts
 	$(MAKE) -C backend/python/vall-e-x
 	$(MAKE) -C backend/python/exllama
 	$(MAKE) -C backend/python/petals

backend/python/parler-tts/Makefile

@@ -0,0 +1,39 @@
+export CONDA_ENV_PATH = "parler.yml"
+SKIP_CONDA?=0
+ifeq ($(BUILD_TYPE), cublas)
+export CONDA_ENV_PATH = "parler-nvidia.yml"
+endif
+
+# Intel GPU are supposed to have dependencies installed in the main python
+# environment, so we skip conda installation for SYCL builds.
+# https://github.com/intel/intel-extension-for-pytorch/issues/538
+ifneq (,$(findstring sycl,$(BUILD_TYPE)))
+export SKIP_CONDA=1
+endif
+
+.PHONY: parler-tts
+parler-tts: protogen
+	@echo "Installing $(CONDA_ENV_PATH)..."
+	bash install.sh $(CONDA_ENV_PATH)
+
+.PHONY: run
+run: protogen
+	@echo "Running transformers..."
+	bash run.sh
+	@echo "transformers run."
+
+.PHONY: test
+test: protogen
+	@echo "Testing transformers..."
+	bash test.sh
+	@echo "transformers tested."
+
+.PHONY: protogen
+protogen: backend_pb2_grpc.py backend_pb2.py
+
+.PHONY: protogen-clean
+protogen-clean:
+	$(RM) backend_pb2_grpc.py backend_pb2.py
+
+backend_pb2_grpc.py backend_pb2.py:
+	python3 -m grpc_tools.protoc -I../.. --python_out=. --grpc_python_out=. backend.proto

backend/python/parler-tts/install.sh

@@ -0,0 +1,39 @@
+#!/bin/bash
+set -ex
+
+SKIP_CONDA=${SKIP_CONDA:-0}
+
+# Check if environment exist
+conda_env_exists(){
+    ! conda list --name "${@}" >/dev/null 2>/dev/null
+}
+
+if [ $SKIP_CONDA -eq 1 ]; then
+    echo "Skipping conda environment installation"
+else
+    export PATH=$PATH:/opt/conda/bin
+    if conda_env_exists "parler" ; then
+        echo "Creating virtual environment..."
+        conda env create --name parler --file $1
+        echo "Virtual environment created."
+    else 
+        echo "Virtual environment already exists."
+    fi
+fi
+
+if [ $SKIP_CONDA -ne 1 ]; then
+    # Activate conda environment
+    source activate parler
+    # https://github.com/descriptinc/audiotools/issues/101
+    # incompatible protobuf versions.
+    curl -L https://raw.githubusercontent.com/protocolbuffers/protobuf/main/python/google/protobuf/internal/builder.py -o $CONDA_PREFIX/lib/python3.11/site-packages/google/protobuf/internal/builder.py
+fi
+
+if [ "$PIP_CACHE_PURGE" = true ] ; then
+    if [ $SKIP_CONDA -ne 1 ]; then
+        # Activate conda environment
+        source activate parler
+    fi
+
+    pip cache purge
+fi

backend/python/parler-tts/parler-nvidia.yml

@@ -0,0 +1,48 @@
+name: parler
+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:
+      - accelerate>=0.11.0
+      - grpcio==1.59.0
+      - numpy==1.26.0
+      - nvidia-cublas-cu12==12.1.3.1
+      - nvidia-cuda-cupti-cu12==12.1.105
+      - nvidia-cuda-nvrtc-cu12==12.1.105
+      - nvidia-cuda-runtime-cu12==12.1.105
+      - nvidia-cudnn-cu12==8.9.2.26
+      - nvidia-cufft-cu12==11.0.2.54
+      - nvidia-curand-cu12==10.3.2.106
+      - nvidia-cusolver-cu12==11.4.5.107
+      - nvidia-cusparse-cu12==12.1.0.106
+      - nvidia-nccl-cu12==2.18.1
+      - nvidia-nvjitlink-cu12==12.2.140
+      - nvidia-nvtx-cu12==12.1.105
+      - torch==2.1.0
+      - transformers>=4.34.0
+      - descript-audio-codec
+      - sentencepiece
+      - git+https://github.com/huggingface/parler-tts.git@10016fb0300c0dc31a0fb70e26f3affee7b62f16
+prefix: /opt/conda/envs/diffusers

backend/python/parler-tts/parler.yml

@@ -0,0 +1,36 @@
+name: parler
+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:
+      - accelerate>=0.11.0
+      - numpy==1.26.0
+      - grpcio==1.59.0
+      - torch==2.1.0
+      - transformers>=4.34.0
+      - descript-audio-codec
+      - sentencepiece
+      - git+https://github.com/huggingface/parler-tts.git@10016fb0300c0dc31a0fb70e26f3affee7b62f16
+prefix: /opt/conda/envs/parler

backend/python/parler-tts/parler_tts_server.py

@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+"""
+Extra gRPC server for MusicgenForConditionalGeneration models.
+"""
+from concurrent import futures
+
+import argparse
+import signal
+import sys
+import os
+
+import time
+import backend_pb2
+import backend_pb2_grpc
+
+import grpc
+
+from scipy.io.wavfile import write as write_wav
+
+from parler_tts import ParlerTTSForConditionalGeneration
+from transformers import AutoTokenizer
+import soundfile as sf  
+import torch
+
+_ONE_DAY_IN_SECONDS = 60 * 60 * 24
+
+# If MAX_WORKERS are specified in the environment use it, otherwise default to 1
+MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1'))
+
+# Implement the BackendServicer class with the service methods
+class BackendServicer(backend_pb2_grpc.BackendServicer):
+    """
+    A gRPC servicer for the backend service.
+
+    This class implements the gRPC methods for the backend service, including Health, LoadModel, and Embedding.
+    """
+    def Health(self, request, context):
+        """
+        A gRPC method that returns the health status of the backend service.
+
+        Args:
+            request: A HealthRequest object that contains the request parameters.
+            context: A grpc.ServicerContext object that provides information about the RPC.
+
+        Returns:
+            A Reply object that contains the health status of the backend service.
+        """
+        return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
+
+    def LoadModel(self, request, context):
+        """
+        A gRPC method that loads a model into memory.
+
+        Args:
+            request: A LoadModelRequest object that contains the request parameters.
+            context: A grpc.ServicerContext object that provides information about the RPC.
+
+        Returns:
+            A Result object that contains the result of the LoadModel operation.
+        """
+        model_name = request.Model
+        device = "cuda:0" if torch.cuda.is_available() else "cpu"
+        try:
+            self.model = ParlerTTSForConditionalGeneration.from_pretrained(model_name).to(device)
+            self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+        except Exception as err:
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+
+        return backend_pb2.Result(message="Model loaded successfully", success=True)
+
+    def TTS(self, request, context):
+        model_name = request.model
+        voice = request.voice
+        if voice == "":
+            voice = "A female speaker with a slightly low-pitched voice delivers her words quite expressively, in a very confined sounding environment with clear audio quality. She speaks very fast."
+        if model_name == "":
+            return backend_pb2.Result(success=False, message="request.model is required")
+        try:
+            device = "cuda:0" if torch.cuda.is_available() else "cpu"
+            input_ids = self.tokenizer(voice, return_tensors="pt").input_ids.to(device)
+            prompt_input_ids = self.tokenizer(request.text, return_tensors="pt").input_ids.to(device)
+
+            generation = self.model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids)
+            audio_arr = generation.cpu().numpy().squeeze()
+            print("[parler-tts] TTS generated!", file=sys.stderr)
+            sf.write(request.dst, audio_arr, self.model.config.sampling_rate)
+            print("[parler-tts] TTS saved to", request.dst, file=sys.stderr)
+            print("[parler-tts] TTS for", file=sys.stderr)
+            print(request, file=sys.stderr)
+        except Exception as err:
+            return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
+        return backend_pb2.Result(success=True)
+
+
+def serve(address):
+    server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS))
+    backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server)
+    server.add_insecure_port(address)
+    server.start()
+    print("[parler-tts] Server started. Listening on: " + address, file=sys.stderr)
+
+    # Define the signal handler function
+    def signal_handler(sig, frame):
+        print("[parler-tts] Received termination signal. Shutting down...")
+        server.stop(0)
+        sys.exit(0)
+
+    # Set the signal handlers for SIGINT and SIGTERM
+    signal.signal(signal.SIGINT, signal_handler)
+    signal.signal(signal.SIGTERM, signal_handler)
+
+    try:
+        while True:
+            time.sleep(_ONE_DAY_IN_SECONDS)
+    except KeyboardInterrupt:
+        server.stop(0)
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Run the gRPC server.")
+    parser.add_argument(
+        "--addr", default="localhost:50051", help="The address to bind the server to."
+    )
+    args = parser.parse_args()
+    print(f"[parler-tts] startup: {args}", file=sys.stderr)
+    serve(args.addr)