mirror of
https://github.com/mudler/LocalAI.git
synced 2025-02-23 18:30:24 +00:00
feat(transformers): add support to Mamba
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto
parent
715071b68d
commit
d1912bcbdd
@ -15,7 +15,7 @@ ARG TARGETARCH
|
||||
ARG TARGETVARIANT
|
||||
|
||||
ENV DEBIAN_FRONTEND=noninteractive
|
||||
ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,transformers:/build/backend/python/transformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,openvoice:/build/backend/python/openvoice/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,mamba:/build/backend/python/mamba/run.sh,exllama2:/build/backend/python/exllama2/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
|
||||
ENV EXTERNAL_GRPC_BACKENDS="coqui:/build/backend/python/coqui/run.sh,transformers:/build/backend/python/transformers/run.sh,rerankers:/build/backend/python/rerankers/run.sh,autogptq:/build/backend/python/autogptq/run.sh,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,openvoice:/build/backend/python/openvoice/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/run.sh,parler-tts:/build/backend/python/parler-tts/run.sh"
|
||||
|
||||
|
||||
RUN apt-get update && \
|
||||
@ -442,9 +442,6 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vllm" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE
|
||||
; fi && \
|
||||
if [[ ( "${EXTRA_BACKENDS}" =~ "rerankers" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
|
||||
make -C backend/python/rerankers \
|
||||
; fi && \
|
||||
if [[ ( "${EXTRA_BACKENDS}" =~ "mamba" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
|
||||
make -C backend/python/mamba \
|
||||
; fi
|
||||
|
||||
# Make sure the models directory exists
|
||||
|
||||
13
Makefile
13
Makefile
@ -533,10 +533,10 @@ protogen-go-clean:
|
||||
$(RM) bin/*
|
||||
|
||||
.PHONY: protogen-python
|
||||
protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen mamba-protogen rerankers-protogen transformers-protogen parler-tts-protogen kokoro-protogen vllm-protogen openvoice-protogen
|
||||
protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen rerankers-protogen transformers-protogen parler-tts-protogen kokoro-protogen vllm-protogen openvoice-protogen
|
||||
|
||||
.PHONY: protogen-python-clean
|
||||
protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama2-protogen-clean mamba-protogen-clean rerankers-protogen-clean transformers-protogen-clean parler-tts-protogen-clean kokoro-protogen-clean vllm-protogen-clean openvoice-protogen-clean
|
||||
protogen-python-clean: autogptq-protogen-clean bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama2-protogen-clean rerankers-protogen-clean transformers-protogen-clean parler-tts-protogen-clean kokoro-protogen-clean vllm-protogen-clean openvoice-protogen-clean
|
||||
|
||||
.PHONY: autogptq-protogen
|
||||
autogptq-protogen:
|
||||
@ -578,14 +578,6 @@ exllama2-protogen:
|
||||
exllama2-protogen-clean:
|
||||
$(MAKE) -C backend/python/exllama2 protogen-clean
|
||||
|
||||
.PHONY: mamba-protogen
|
||||
mamba-protogen:
|
||||
$(MAKE) -C backend/python/mamba protogen
|
||||
|
||||
.PHONY: mamba-protogen-clean
|
||||
mamba-protogen-clean:
|
||||
$(MAKE) -C backend/python/mamba protogen-clean
|
||||
|
||||
.PHONY: rerankers-protogen
|
||||
rerankers-protogen:
|
||||
$(MAKE) -C backend/python/rerankers protogen
|
||||
@ -642,7 +634,6 @@ prepare-extra-conda-environments: protogen-python
|
||||
$(MAKE) -C backend/python/coqui
|
||||
$(MAKE) -C backend/python/diffusers
|
||||
$(MAKE) -C backend/python/vllm
|
||||
$(MAKE) -C backend/python/mamba
|
||||
$(MAKE) -C backend/python/rerankers
|
||||
$(MAKE) -C backend/python/transformers
|
||||
$(MAKE) -C backend/python/parler-tts
|
||||
|
||||
@ -1,29 +0,0 @@
|
||||
.PHONY: mamba
|
||||
mamba: protogen
|
||||
bash install.sh
|
||||
|
||||
.PHONY: run
|
||||
run: protogen
|
||||
@echo "Running mamba..."
|
||||
bash run.sh
|
||||
@echo "mamba run."
|
||||
|
||||
.PHONY: test
|
||||
test: protogen
|
||||
@echo "Testing mamba..."
|
||||
bash test.sh
|
||||
@echo "mamba 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) -r venv __pycache__
|
||||
@ -1,5 +0,0 @@
|
||||
# Creating a separate environment for the mamba project
|
||||
|
||||
```
|
||||
make mamba
|
||||
```
|
||||
@ -1,179 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
from concurrent import futures
|
||||
import time
|
||||
import argparse
|
||||
import signal
|
||||
import sys
|
||||
import os
|
||||
|
||||
import backend_pb2
|
||||
import backend_pb2_grpc
|
||||
|
||||
import grpc
|
||||
|
||||
import torch
|
||||
from transformers import AutoTokenizer, AutoModelForCausalLM
|
||||
from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel
|
||||
|
||||
_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'))
|
||||
MAMBA_CHAT= os.environ.get('MAMBA_CHAT', '1') == '1'
|
||||
|
||||
# Implement the BackendServicer class with the service methods
|
||||
class BackendServicer(backend_pb2_grpc.BackendServicer):
|
||||
"""
|
||||
A gRPC servicer that implements the Backend service defined in backend.proto.
|
||||
"""
|
||||
def generate(self,prompt, max_new_tokens):
|
||||
"""
|
||||
Generates text based on the given prompt and maximum number of new tokens.
|
||||
|
||||
Args:
|
||||
prompt (str): The prompt to generate text from.
|
||||
max_new_tokens (int): The maximum number of new tokens to generate.
|
||||
|
||||
Returns:
|
||||
str: The generated text.
|
||||
"""
|
||||
self.generator.end_beam_search()
|
||||
|
||||
# Tokenizing the input
|
||||
ids = self.generator.tokenizer.encode(prompt)
|
||||
|
||||
self.generator.gen_begin_reuse(ids)
|
||||
initial_len = self.generator.sequence[0].shape[0]
|
||||
has_leading_space = False
|
||||
decoded_text = ''
|
||||
for i in range(max_new_tokens):
|
||||
token = self.generator.gen_single_token()
|
||||
if i == 0 and self.generator.tokenizer.tokenizer.IdToPiece(int(token)).startswith('▁'):
|
||||
has_leading_space = True
|
||||
|
||||
decoded_text = self.generator.tokenizer.decode(self.generator.sequence[0][initial_len:])
|
||||
if has_leading_space:
|
||||
decoded_text = ' ' + decoded_text
|
||||
|
||||
if token.item() == self.generator.tokenizer.eos_token_id:
|
||||
break
|
||||
return decoded_text
|
||||
|
||||
def Health(self, request, context):
|
||||
"""
|
||||
Returns a health check message.
|
||||
|
||||
Args:
|
||||
request: The health check request.
|
||||
context: The gRPC context.
|
||||
|
||||
Returns:
|
||||
backend_pb2.Reply: The health check reply.
|
||||
"""
|
||||
return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
|
||||
|
||||
def LoadModel(self, request, context):
|
||||
"""
|
||||
Loads a language model.
|
||||
|
||||
Args:
|
||||
request: The load model request.
|
||||
context: The gRPC context.
|
||||
|
||||
Returns:
|
||||
backend_pb2.Result: The load model result.
|
||||
"""
|
||||
try:
|
||||
tokenizerModel = request.Tokenizer
|
||||
if tokenizerModel == "":
|
||||
tokenizerModel = request.Model
|
||||
|
||||
tokenizer = AutoTokenizer.from_pretrained(tokenizerModel)
|
||||
if MAMBA_CHAT:
|
||||
tokenizer.eos_token = "<|endoftext|>"
|
||||
tokenizer.pad_token = tokenizer.eos_token
|
||||
self.tokenizer = tokenizer
|
||||
self.model = MambaLMHeadModel.from_pretrained(request.Model, device="cuda", dtype=torch.float16)
|
||||
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 Predict(self, request, context):
|
||||
"""
|
||||
Generates text based on the given prompt and sampling parameters.
|
||||
|
||||
Args:
|
||||
request: The predict request.
|
||||
context: The gRPC context.
|
||||
|
||||
Returns:
|
||||
backend_pb2.Result: The predict result.
|
||||
"""
|
||||
if request.TopP == 0:
|
||||
request.TopP = 0.9
|
||||
|
||||
max_tokens = request.Tokens
|
||||
|
||||
if request.Tokens == 0:
|
||||
max_tokens = 2000
|
||||
|
||||
# encoded_input = self.tokenizer(request.Prompt)
|
||||
tokens = self.tokenizer(request.Prompt, return_tensors="pt")
|
||||
input_ids = tokens.input_ids.to(device="cuda")
|
||||
out = self.model.generate(input_ids=input_ids, max_length=max_tokens, temperature=request.Temperature,
|
||||
top_p=request.TopP, eos_token_id=self.tokenizer.eos_token_id)
|
||||
|
||||
decoded = self.tokenizer.batch_decode(out)
|
||||
|
||||
generated_text = decoded[0]
|
||||
|
||||
# Remove prompt from response if present
|
||||
if request.Prompt in generated_text:
|
||||
generated_text = generated_text.replace(request.Prompt, "")
|
||||
|
||||
return backend_pb2.Reply(message=bytes(generated_text, encoding='utf-8'))
|
||||
|
||||
def PredictStream(self, request, context):
|
||||
"""
|
||||
Generates text based on the given prompt and sampling parameters, and streams the results.
|
||||
|
||||
Args:
|
||||
request: The predict stream request.
|
||||
context: The gRPC context.
|
||||
|
||||
Returns:
|
||||
backend_pb2.Result: The predict stream result.
|
||||
"""
|
||||
yield self.Predict(request, context)
|
||||
|
||||
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)
|
||||
@ -1,9 +0,0 @@
|
||||
#!/bin/bash
|
||||
set -e
|
||||
|
||||
LIMIT_TARGETS="cublas"
|
||||
EXTRA_PIP_INSTALL_FLAGS="--no-build-isolation"
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
installRequirements
|
||||
@ -1,2 +0,0 @@
|
||||
causal-conv1d==1.4.0
|
||||
mamba-ssm==2.2.2
|
||||
@ -1,2 +0,0 @@
|
||||
torch==2.4.1
|
||||
transformers
|
||||
@ -1,3 +0,0 @@
|
||||
--extra-index-url https://download.pytorch.org/whl/cu118
|
||||
torch==2.4.1+cu118
|
||||
transformers
|
||||
@ -1,2 +0,0 @@
|
||||
torch==2.4.1
|
||||
transformers
|
||||
@ -1,6 +0,0 @@
|
||||
# mabma does not specify it's build dependencies per PEP517, so we need to disable build isolation
|
||||
# this also means that we need to install the basic build dependencies into the venv ourselves
|
||||
# https://github.com/Dao-AILab/causal-conv1d/issues/24
|
||||
packaging
|
||||
setuptools
|
||||
wheel
|
||||
@ -1,3 +0,0 @@
|
||||
grpcio==1.69.0
|
||||
protobuf
|
||||
certifi
|
||||
@ -1,6 +0,0 @@
|
||||
#!/bin/bash
|
||||
LIMIT_TARGETS="cublas"
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
startBackend $@
|
||||
@ -1,76 +0,0 @@
|
||||
import unittest
|
||||
import subprocess
|
||||
import time
|
||||
import backend_pb2
|
||||
import backend_pb2_grpc
|
||||
|
||||
import grpc
|
||||
|
||||
import unittest
|
||||
import subprocess
|
||||
import time
|
||||
import grpc
|
||||
import backend_pb2_grpc
|
||||
import backend_pb2
|
||||
|
||||
class TestBackendServicer(unittest.TestCase):
|
||||
"""
|
||||
TestBackendServicer is the class that tests the gRPC service.
|
||||
|
||||
This class contains methods to test the startup and shutdown of the gRPC service.
|
||||
"""
|
||||
def setUp(self):
|
||||
self.service = subprocess.Popen(["python", "backend.py", "--addr", "localhost:50051"])
|
||||
time.sleep(10)
|
||||
|
||||
def tearDown(self) -> None:
|
||||
self.service.terminate()
|
||||
self.service.wait()
|
||||
|
||||
def test_server_startup(self):
|
||||
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="facebook/opt-125m"))
|
||||
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_text(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="facebook/opt-125m"))
|
||||
self.assertTrue(response.success)
|
||||
req = backend_pb2.PredictOptions(Prompt="The capital of France is")
|
||||
resp = stub.Predict(req)
|
||||
self.assertIsNotNone(resp.message)
|
||||
except Exception as err:
|
||||
print(err)
|
||||
self.fail("text service failed")
|
||||
finally:
|
||||
self.tearDown()
|
||||
@ -1,6 +0,0 @@
|
||||
#!/bin/bash
|
||||
set -e
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
runUnittests
|
||||
@ -21,7 +21,7 @@ import torch.cuda
|
||||
|
||||
|
||||
XPU=os.environ.get("XPU", "0") == "1"
|
||||
from transformers import AutoTokenizer, AutoModel, set_seed, TextIteratorStreamer, StoppingCriteriaList, StopStringCriteria
|
||||
from transformers import AutoTokenizer, AutoModel, set_seed, TextIteratorStreamer, StoppingCriteriaList, StopStringCriteria, MambaConfig, MambaForCausalLM
|
||||
from transformers import AutoProcessor, MusicgenForConditionalGeneration
|
||||
from scipy.io import wavfile
|
||||
import outetts
|
||||
@ -245,6 +245,10 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
|
||||
autoTokenizer = False
|
||||
self.model = SentenceTransformer(model_name, trust_remote_code=request.TrustRemoteCode)
|
||||
self.SentenceTransformer = True
|
||||
elif request.Type == "Mamba":
|
||||
autoTokenizer = False
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
|
||||
self.model = MambaForCausalLM.from_pretrained(model_name)
|
||||
else:
|
||||
print("Automodel", file=sys.stderr)
|
||||
self.model = AutoModel.from_pretrained(model_name,
|
||||
|
||||
@ -29,12 +29,14 @@ var Aliases map[string]string = map[string]string{
|
||||
"langchain-huggingface": LCHuggingFaceBackend,
|
||||
"transformers-musicgen": TransformersBackend,
|
||||
"sentencetransformers": TransformersBackend,
|
||||
"mamba": TransformersBackend,
|
||||
"stablediffusion": StableDiffusionGGMLBackend,
|
||||
}
|
||||
|
||||
var TypeAlias map[string]string = map[string]string{
|
||||
"sentencetransformers": "SentenceTransformer",
|
||||
"huggingface-embeddings": "SentenceTransformer",
|
||||
"mamba": "Mamba",
|
||||
"transformers-musicgen": "MusicgenForConditionalGeneration",
|
||||
}
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user