mirror of
https://github.com/mudler/LocalAI.git
synced 2025-05-20 02:24:59 +00:00
feat(transformers): merge sentencetransformers backend (#4624)
* merge sentencetransformers Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Add alias to silently redirect sentencetransformers to transformers Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Add alias also for transformers-musicgen Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Drop from makefile Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Move tests from sentencetransformers Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Remove sentencetransformers Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Remove tests from CI (part of transformers) Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Do not always try to load the tokenizer Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Adapt tests Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Fix typo Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Tiny adjustments Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto
GitHub
parent
4bd8434ae0
commit
1e9bf19c8d
27 changed files with 104 additions and 354 deletions
|
|
@ -1,31 +0,0 @@
|
|||
.PHONY: sentencetransformers
|
||||
sentencetransformers: protogen
|
||||
bash ./install.sh
|
||||
|
||||
|
||||
.PHONY: run
|
||||
run: protogen
|
||||
@echo "Running sentencetransformers..."
|
||||
bash run.sh
|
||||
@echo "sentencetransformers run."
|
||||
|
||||
# It is not working well by using command line. It only6 works with IDE like VSCode.
|
||||
.PHONY: test
|
||||
test: protogen
|
||||
@echo "Testing sentencetransformers..."
|
||||
bash test.sh
|
||||
@echo "sentencetransformers 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
|
||||
|
||||
.PHONY: clean
|
||||
clean: protogen-clean
|
||||
rm -rf venv __pycache__
|
||||
|
|
@ -1,5 +0,0 @@
|
|||
# Creating a separate environment for the sentencetransformers project
|
||||
|
||||
```
|
||||
make sentencetransformers
|
||||
```
|
||||
|
|
@ -1,114 +0,0 @@
|
|||
#!/usr/bin/env python3
|
||||
"""
|
||||
Extra gRPC server for HuggingFace SentenceTransformer 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 sentence_transformers import SentenceTransformer
|
||||
|
||||
_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
|
||||
try:
|
||||
self.model = SentenceTransformer(model_name, trust_remote_code=request.TrustRemoteCode)
|
||||
except Exception as err:
|
||||
return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}")
|
||||
|
||||
# Implement your logic here for the LoadModel service
|
||||
# Replace this with your desired response
|
||||
return backend_pb2.Result(message="Model loaded successfully", success=True)
|
||||
|
||||
def Embedding(self, request, context):
|
||||
"""
|
||||
A gRPC method that calculates embeddings for a given sentence.
|
||||
|
||||
Args:
|
||||
request: An EmbeddingRequest object that contains the request parameters.
|
||||
context: A grpc.ServicerContext object that provides information about the RPC.
|
||||
|
||||
Returns:
|
||||
An EmbeddingResult object that contains the calculated embeddings.
|
||||
"""
|
||||
# Implement your logic here for the Embedding service
|
||||
# Replace this with your desired response
|
||||
print("Calculated embeddings for: " + request.Embeddings, file=sys.stderr)
|
||||
sentence_embeddings = self.model.encode(request.Embeddings)
|
||||
return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings)
|
||||
|
||||
|
||||
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("Server started. Listening on: " + address, file=sys.stderr)
|
||||
|
||||
# Define the signal handler function
|
||||
def signal_handler(sig, frame):
|
||||
print("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()
|
||||
|
||||
serve(args.addr)
|
||||
|
|
@ -1,14 +0,0 @@
|
|||
#!/bin/bash
|
||||
set -e
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
# This is here because the Intel pip index is broken and returns 200 status codes for every package name, it just doesn't return any package links.
|
||||
# This makes uv think that the package exists in the Intel pip index, and by default it stops looking at other pip indexes once it finds a match.
|
||||
# We need uv to continue falling through to the pypi default index to find optimum[openvino] in the pypi index
|
||||
# the --upgrade actually allows us to *downgrade* torch to the version provided in the Intel pip index
|
||||
if [ "x${BUILD_PROFILE}" == "xintel" ]; then
|
||||
EXTRA_PIP_INSTALL_FLAGS+=" --upgrade --index-strategy=unsafe-first-match"
|
||||
fi
|
||||
|
||||
installRequirements
|
||||
|
|
@ -1,6 +0,0 @@
|
|||
torch==2.4.1
|
||||
accelerate
|
||||
transformers
|
||||
bitsandbytes
|
||||
sentence-transformers==3.3.1
|
||||
transformers
|
||||
|
|
@ -1,5 +0,0 @@
|
|||
--extra-index-url https://download.pytorch.org/whl/cu118
|
||||
torch==2.4.1+cu118
|
||||
accelerate
|
||||
sentence-transformers==3.3.1
|
||||
transformers
|
||||
|
|
@ -1,4 +0,0 @@
|
|||
torch==2.4.1
|
||||
accelerate
|
||||
sentence-transformers==3.3.1
|
||||
transformers
|
||||
|
|
@ -1,5 +0,0 @@
|
|||
--extra-index-url https://download.pytorch.org/whl/rocm6.0
|
||||
torch==2.4.1+rocm6.0
|
||||
accelerate
|
||||
sentence-transformers==3.3.1
|
||||
transformers
|
||||
|
|
@ -1,9 +0,0 @@
|
|||
--extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/
|
||||
intel-extension-for-pytorch==2.3.110+xpu
|
||||
torch==2.3.1+cxx11.abi
|
||||
oneccl_bind_pt==2.3.100+xpu
|
||||
optimum[openvino]
|
||||
setuptools
|
||||
accelerate
|
||||
sentence-transformers==3.3.1
|
||||
transformers
|
||||
|
|
@ -1,5 +0,0 @@
|
|||
grpcio==1.69.0
|
||||
protobuf
|
||||
certifi
|
||||
datasets
|
||||
einops
|
||||
|
|
@ -1,4 +0,0 @@
|
|||
#!/bin/bash
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
startBackend $@
|
||||
|
|
@ -1,81 +0,0 @@
|
|||
"""
|
||||
A test script to test the gRPC service
|
||||
"""
|
||||
import unittest
|
||||
import subprocess
|
||||
import time
|
||||
import backend_pb2
|
||||
import backend_pb2_grpc
|
||||
|
||||
import grpc
|
||||
|
||||
|
||||
class TestBackendServicer(unittest.TestCase):
|
||||
"""
|
||||
TestBackendServicer is the class that tests the gRPC service
|
||||
"""
|
||||
def setUp(self):
|
||||
"""
|
||||
This method sets up the gRPC service by starting the server
|
||||
"""
|
||||
self.service = subprocess.Popen(["python3", "backend.py", "--addr", "localhost:50051"])
|
||||
time.sleep(10)
|
||||
|
||||
def tearDown(self) -> None:
|
||||
"""
|
||||
This method tears down the gRPC service by terminating the server
|
||||
"""
|
||||
self.service.kill()
|
||||
self.service.wait()
|
||||
|
||||
def test_server_startup(self):
|
||||
"""
|
||||
This method tests if the server starts up successfully
|
||||
"""
|
||||
try:
|
||||
self.setUp()
|
||||
with grpc.insecure_channel("localhost:50051") as channel:
|
||||
stub = backend_pb2_grpc.BackendStub(channel)
|
||||
response = stub.Health(backend_pb2.HealthMessage())
|
||||
self.assertEqual(response.message, b'OK')
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("Server failed to start")
|
||||
finally:
|
||||
self.tearDown()
|
||||
|
||||
def test_load_model(self):
|
||||
"""
|
||||
This method tests if the model is loaded successfully
|
||||
"""
|
||||
try:
|
||||
self.setUp()
|
||||
with grpc.insecure_channel("localhost:50051") as channel:
|
||||
stub = backend_pb2_grpc.BackendStub(channel)
|
||||
response = stub.LoadModel(backend_pb2.ModelOptions(Model="bert-base-nli-mean-tokens"))
|
||||
self.assertTrue(response.success)
|
||||
self.assertEqual(response.message, "Model loaded successfully")
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("LoadModel service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
|
||||
def test_embedding(self):
|
||||
"""
|
||||
This method tests if the embeddings are generated successfully
|
||||
"""
|
||||
try:
|
||||
self.setUp()
|
||||
with grpc.insecure_channel("localhost:50051") as channel:
|
||||
stub = backend_pb2_grpc.BackendStub(channel)
|
||||
response = stub.LoadModel(backend_pb2.ModelOptions(Model="bert-base-nli-mean-tokens"))
|
||||
self.assertTrue(response.success)
|
||||
embedding_request = backend_pb2.PredictOptions(Embeddings="This is a test sentence.")
|
||||
embedding_response = stub.Embedding(embedding_request)
|
||||
self.assertIsNotNone(embedding_response.embeddings)
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("Embedding service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
|
|
@ -1,6 +0,0 @@
|
|||
#!/bin/bash
|
||||
set -e
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
runUnittests
|
||||
|
|
@ -25,6 +25,8 @@ from transformers import AutoTokenizer, AutoModel, set_seed, TextIteratorStreame
|
|||
from transformers import AutoProcessor, MusicgenForConditionalGeneration
|
||||
from scipy.io import wavfile
|
||||
import outetts
|
||||
from sentence_transformers import SentenceTransformer
|
||||
|
||||
|
||||
_ONE_DAY_IN_SECONDS = 60 * 60 * 24
|
||||
|
||||
|
|
@ -88,10 +90,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
|||
self.CUDA = torch.cuda.is_available()
|
||||
self.OV=False
|
||||
self.OuteTTS=False
|
||||
self.SentenceTransformer = False
|
||||
|
||||
device_map="cpu"
|
||||
|
||||
quantization = None
|
||||
autoTokenizer = True
|
||||
|
||||
if self.CUDA:
|
||||
from transformers import BitsAndBytesConfig, AutoModelForCausalLM
|
||||
|
|
@ -195,9 +199,11 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
|||
device=device_map)
|
||||
self.OV = True
|
||||
elif request.Type == "MusicgenForConditionalGeneration":
|
||||
autoTokenizer = False
|
||||
self.processor = AutoProcessor.from_pretrained(model_name)
|
||||
self.model = MusicgenForConditionalGeneration.from_pretrained(model_name)
|
||||
elif request.Type == "OuteTTS":
|
||||
autoTokenizer = False
|
||||
options = request.Options
|
||||
MODELNAME = "OuteAI/OuteTTS-0.3-1B"
|
||||
TOKENIZER = "OuteAI/OuteTTS-0.3-1B"
|
||||
|
|
@ -235,6 +241,10 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
|||
self.speaker = self.interface.create_speaker(audio_path=self.AudioPath)
|
||||
else:
|
||||
self.speaker = self.interface.load_default_speaker(name=SPEAKER)
|
||||
elif request.Type == "SentenceTransformer":
|
||||
autoTokenizer = False
|
||||
self.model = SentenceTransformer(model_name, trust_remote_code=request.TrustRemoteCode)
|
||||
self.SentenceTransformer = True
|
||||
else:
|
||||
print("Automodel", file=sys.stderr)
|
||||
self.model = AutoModel.from_pretrained(model_name,
|
||||
|
|
@ -250,7 +260,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
|||
else:
|
||||
self.max_tokens = 512
|
||||
|
||||
if request.Type != "MusicgenForConditionalGeneration":
|
||||
if autoTokenizer:
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(model_name, use_safetensors=True)
|
||||
self.XPU = False
|
||||
|
||||
|
|
@ -286,18 +296,26 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
|||
max_length = 512
|
||||
if request.Tokens != 0:
|
||||
max_length = request.Tokens
|
||||
encoded_input = self.tokenizer(request.Embeddings, padding=True, truncation=True, max_length=max_length, return_tensors="pt")
|
||||
|
||||
# Create word embeddings
|
||||
if self.CUDA:
|
||||
encoded_input = encoded_input.to("cuda")
|
||||
embeds = None
|
||||
|
||||
with torch.no_grad():
|
||||
model_output = self.model(**encoded_input)
|
||||
if self.SentenceTransformer:
|
||||
print("Calculated embeddings for: " + request.Embeddings, file=sys.stderr)
|
||||
embeds = self.model.encode(request.Embeddings)
|
||||
else:
|
||||
encoded_input = self.tokenizer(request.Embeddings, padding=True, truncation=True, max_length=max_length, return_tensors="pt")
|
||||
|
||||
# Pool to get sentence embeddings; i.e. generate one 1024 vector for the entire sentence
|
||||
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
|
||||
return backend_pb2.EmbeddingResult(embeddings=sentence_embeddings[0])
|
||||
# Create word embeddings
|
||||
if self.CUDA:
|
||||
encoded_input = encoded_input.to("cuda")
|
||||
|
||||
with torch.no_grad():
|
||||
model_output = self.model(**encoded_input)
|
||||
|
||||
# Pool to get sentence embeddings; i.e. generate one 1024 vector for the entire sentence
|
||||
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
|
||||
embeds = sentence_embeddings[0]
|
||||
return backend_pb2.EmbeddingResult(embeddings=embeds)
|
||||
|
||||
async def _predict(self, request, context, streaming=False):
|
||||
set_seed(request.Seed)
|
||||
|
|
|
|||
|
|
@ -3,4 +3,5 @@ llvmlite==0.43.0
|
|||
accelerate
|
||||
transformers
|
||||
bitsandbytes
|
||||
outetts
|
||||
outetts
|
||||
sentence-transformers==3.3.1
|
||||
|
|
|
|||
|
|
@ -4,4 +4,5 @@ llvmlite==0.43.0
|
|||
accelerate
|
||||
transformers
|
||||
bitsandbytes
|
||||
outetts
|
||||
outetts
|
||||
sentence-transformers==3.3.1
|
||||
|
|
|
|||
|
|
@ -3,4 +3,5 @@ accelerate
|
|||
llvmlite==0.43.0
|
||||
transformers
|
||||
bitsandbytes
|
||||
outetts
|
||||
outetts
|
||||
sentence-transformers==3.3.1
|
||||
|
|
|
|||
|
|
@ -4,4 +4,6 @@ accelerate
|
|||
transformers
|
||||
llvmlite==0.43.0
|
||||
bitsandbytes
|
||||
outetts
|
||||
outetts
|
||||
bitsandbytes
|
||||
sentence-transformers==3.3.1
|
||||
|
|
|
|||
|
|
@ -6,4 +6,5 @@ optimum[openvino]
|
|||
llvmlite==0.43.0
|
||||
intel-extension-for-transformers
|
||||
bitsandbytes
|
||||
outetts
|
||||
outetts
|
||||
sentence-transformers==3.3.1
|
||||
|
|
|
|||
|
|
@ -133,5 +133,41 @@ class TestBackendServicer(unittest.TestCase):
|
|||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("SoundGeneration service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
|
||||
def test_embed_load_model(self):
|
||||
"""
|
||||
This method tests if the model is loaded successfully
|
||||
"""
|
||||
try:
|
||||
self.setUp()
|
||||
with grpc.insecure_channel("localhost:50051") as channel:
|
||||
stub = backend_pb2_grpc.BackendStub(channel)
|
||||
response = stub.LoadModel(backend_pb2.ModelOptions(Model="bert-base-nli-mean-tokens",Type="SentenceTransformer"))
|
||||
self.assertTrue(response.success)
|
||||
self.assertEqual(response.message, "Model loaded successfully")
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("LoadModel service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
|
||||
def test_sentencetransformers_embedding(self):
|
||||
"""
|
||||
This method tests if the embeddings are generated successfully
|
||||
"""
|
||||
try:
|
||||
self.setUp()
|
||||
with grpc.insecure_channel("localhost:50051") as channel:
|
||||
stub = backend_pb2_grpc.BackendStub(channel)
|
||||
response = stub.LoadModel(backend_pb2.ModelOptions(Model="bert-base-nli-mean-tokens",Type="SentenceTransformer"))
|
||||
self.assertTrue(response.success)
|
||||
embedding_request = backend_pb2.PredictOptions(Embeddings="This is a test sentence.")
|
||||
embedding_response = stub.Embedding(embedding_request)
|
||||
self.assertIsNotNone(embedding_response.embeddings)
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("Embedding service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
Loading…
Add table
Add a link
Reference in a new issue