mirror of
https://github.com/mudler/LocalAI.git
synced 2025-04-20 00:51:13 +00:00
chore(autogptq): drop archived backend
Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto
parent
8abecb4a18
commit
a7be2d2313
4
.github/dependabot.yml
vendored
4
.github/dependabot.yml
vendored
@ -29,10 +29,6 @@ updates:
|
||||
schedule:
|
||||
# Check for updates to GitHub Actions every weekday
|
||||
interval: "weekly"
|
||||
- package-ecosystem: "pip"
|
||||
directory: "/backend/python/autogptq"
|
||||
schedule:
|
||||
interval: "weekly"
|
||||
- package-ecosystem: "pip"
|
||||
directory: "/backend/python/bark"
|
||||
schedule:
|
||||
|
||||
@ -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,faster-whisper:/build/backend/python/faster-whisper/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/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,bark:/build/backend/python/bark/run.sh,diffusers:/build/backend/python/diffusers/run.sh,faster-whisper:/build/backend/python/faster-whisper/run.sh,kokoro:/build/backend/python/kokoro/run.sh,vllm:/build/backend/python/vllm/run.sh,exllama2:/build/backend/python/exllama2/run.sh"
|
||||
|
||||
RUN apt-get update && \
|
||||
apt-get install -y --no-install-recommends \
|
||||
@ -431,9 +431,6 @@ RUN if [[ ( "${EXTRA_BACKENDS}" =~ "kokoro" || -z "${EXTRA_BACKENDS}" ) && "$IMA
|
||||
RUN if [[ ( "${EXTRA_BACKENDS}" =~ "vllm" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
|
||||
make -C backend/python/vllm \
|
||||
; fi && \
|
||||
if [[ ( "${EXTRA_BACKENDS}" =~ "autogptq" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
|
||||
make -C backend/python/autogptq \
|
||||
; fi && \
|
||||
if [[ ( "${EXTRA_BACKENDS}" =~ "bark" || -z "${EXTRA_BACKENDS}" ) && "$IMAGE_TYPE" == "extras" ]]; then \
|
||||
make -C backend/python/bark \
|
||||
; fi && \
|
||||
|
||||
13
Makefile
13
Makefile
@ -505,18 +505,10 @@ protogen-go-clean:
|
||||
$(RM) bin/*
|
||||
|
||||
.PHONY: protogen-python
|
||||
protogen-python: autogptq-protogen bark-protogen coqui-protogen diffusers-protogen exllama2-protogen rerankers-protogen transformers-protogen kokoro-protogen vllm-protogen faster-whisper-protogen
|
||||
protogen-python: bark-protogen coqui-protogen diffusers-protogen exllama2-protogen rerankers-protogen transformers-protogen kokoro-protogen vllm-protogen faster-whisper-protogen
|
||||
|
||||
.PHONY: protogen-python-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 kokoro-protogen-clean vllm-protogen-clean faster-whisper-protogen-clean
|
||||
|
||||
.PHONY: autogptq-protogen
|
||||
autogptq-protogen:
|
||||
$(MAKE) -C backend/python/autogptq protogen
|
||||
|
||||
.PHONY: autogptq-protogen-clean
|
||||
autogptq-protogen-clean:
|
||||
$(MAKE) -C backend/python/autogptq protogen-clean
|
||||
protogen-python-clean: bark-protogen-clean coqui-protogen-clean diffusers-protogen-clean exllama2-protogen-clean rerankers-protogen-clean transformers-protogen-clean kokoro-protogen-clean vllm-protogen-clean faster-whisper-protogen-clean
|
||||
|
||||
.PHONY: bark-protogen
|
||||
bark-protogen:
|
||||
@ -593,7 +585,6 @@ vllm-protogen-clean:
|
||||
## GRPC
|
||||
# Note: it is duplicated in the Dockerfile
|
||||
prepare-extra-conda-environments: protogen-python
|
||||
$(MAKE) -C backend/python/autogptq
|
||||
$(MAKE) -C backend/python/bark
|
||||
$(MAKE) -C backend/python/coqui
|
||||
$(MAKE) -C backend/python/diffusers
|
||||
|
||||
@ -190,11 +190,7 @@ message ModelOptions {
|
||||
int32 NGQA = 20;
|
||||
string ModelFile = 21;
|
||||
|
||||
// AutoGPTQ
|
||||
string Device = 22;
|
||||
bool UseTriton = 23;
|
||||
string ModelBaseName = 24;
|
||||
bool UseFastTokenizer = 25;
|
||||
|
||||
|
||||
// Diffusers
|
||||
string PipelineType = 26;
|
||||
|
||||
@ -1,17 +0,0 @@
|
||||
.PHONY: autogptq
|
||||
autogptq: protogen
|
||||
bash install.sh
|
||||
|
||||
.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 autogptq project
|
||||
|
||||
```
|
||||
make autogptq
|
||||
```
|
||||
@ -1,158 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
from concurrent import futures
|
||||
import argparse
|
||||
import signal
|
||||
import sys
|
||||
import os
|
||||
import time
|
||||
import base64
|
||||
|
||||
import grpc
|
||||
import backend_pb2
|
||||
import backend_pb2_grpc
|
||||
|
||||
from auto_gptq import AutoGPTQForCausalLM
|
||||
from transformers import AutoTokenizer, AutoModelForCausalLM
|
||||
from transformers import TextGenerationPipeline
|
||||
|
||||
_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):
|
||||
def Health(self, request, context):
|
||||
return backend_pb2.Reply(message=bytes("OK", 'utf-8'))
|
||||
def LoadModel(self, request, context):
|
||||
try:
|
||||
device = "cuda:0"
|
||||
if request.Device != "":
|
||||
device = request.Device
|
||||
|
||||
# support loading local model files
|
||||
model_path = os.path.join(os.environ.get('MODELS_PATH', './'), request.Model)
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True, trust_remote_code=request.TrustRemoteCode)
|
||||
|
||||
# support model `Qwen/Qwen-VL-Chat-Int4`
|
||||
if "qwen-vl" in request.Model.lower():
|
||||
self.model_name = "Qwen-VL-Chat"
|
||||
model = AutoModelForCausalLM.from_pretrained(model_path,
|
||||
trust_remote_code=request.TrustRemoteCode,
|
||||
device_map="auto").eval()
|
||||
else:
|
||||
model = AutoGPTQForCausalLM.from_quantized(model_path,
|
||||
model_basename=request.ModelBaseName,
|
||||
use_safetensors=True,
|
||||
trust_remote_code=request.TrustRemoteCode,
|
||||
device=device,
|
||||
use_triton=request.UseTriton,
|
||||
quantize_config=None)
|
||||
|
||||
self.model = model
|
||||
self.tokenizer = tokenizer
|
||||
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):
|
||||
penalty = 1.0
|
||||
if request.Penalty != 0.0:
|
||||
penalty = request.Penalty
|
||||
tokens = 512
|
||||
if request.Tokens != 0:
|
||||
tokens = request.Tokens
|
||||
top_p = 0.95
|
||||
if request.TopP != 0.0:
|
||||
top_p = request.TopP
|
||||
|
||||
|
||||
prompt_images = self.recompile_vl_prompt(request)
|
||||
compiled_prompt = prompt_images[0]
|
||||
print(f"Prompt: {compiled_prompt}", file=sys.stderr)
|
||||
|
||||
# Implement Predict RPC
|
||||
pipeline = TextGenerationPipeline(
|
||||
model=self.model,
|
||||
tokenizer=self.tokenizer,
|
||||
max_new_tokens=tokens,
|
||||
temperature=request.Temperature,
|
||||
top_p=top_p,
|
||||
repetition_penalty=penalty,
|
||||
)
|
||||
t = pipeline(compiled_prompt)[0]["generated_text"]
|
||||
print(f"generated_text: {t}", file=sys.stderr)
|
||||
|
||||
if compiled_prompt in t:
|
||||
t = t.replace(compiled_prompt, "")
|
||||
# house keeping. Remove the image files from /tmp folder
|
||||
for img_path in prompt_images[1]:
|
||||
try:
|
||||
os.remove(img_path)
|
||||
except Exception as e:
|
||||
print(f"Error removing image file: {img_path}, {e}", file=sys.stderr)
|
||||
|
||||
return backend_pb2.Result(message=bytes(t, encoding='utf-8'))
|
||||
|
||||
def PredictStream(self, request, context):
|
||||
# Implement PredictStream RPC
|
||||
#for reply in some_data_generator():
|
||||
# yield reply
|
||||
# Not implemented yet
|
||||
return self.Predict(request, context)
|
||||
|
||||
def recompile_vl_prompt(self, request):
|
||||
prompt = request.Prompt
|
||||
image_paths = []
|
||||
|
||||
if "qwen-vl" in self.model_name.lower():
|
||||
# request.Images is an array which contains base64 encoded images. Iterate the request.Images array, decode and save each image to /tmp folder with a random filename.
|
||||
# Then, save the image file paths to an array "image_paths".
|
||||
# read "request.Prompt", replace "[img-%d]" with the image file paths in the order they appear in "image_paths". Save the new prompt to "prompt".
|
||||
for i, img in enumerate(request.Images):
|
||||
timestamp = str(int(time.time() * 1000)) # Generate timestamp
|
||||
img_path = f"/tmp/vl-{timestamp}.jpg" # Use timestamp in filename
|
||||
with open(img_path, "wb") as f:
|
||||
f.write(base64.b64decode(img))
|
||||
image_paths.append(img_path)
|
||||
prompt = prompt.replace(f"[img-{i}]", "<img>" + img_path + "</img>,")
|
||||
else:
|
||||
prompt = request.Prompt
|
||||
return (prompt, image_paths)
|
||||
|
||||
def serve(address):
|
||||
server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS),
|
||||
options=[
|
||||
('grpc.max_message_length', 50 * 1024 * 1024), # 50MB
|
||||
('grpc.max_send_message_length', 50 * 1024 * 1024), # 50MB
|
||||
('grpc.max_receive_message_length', 50 * 1024 * 1024), # 50MB
|
||||
])
|
||||
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
|
||||
@ -1,2 +0,0 @@
|
||||
--extra-index-url https://download.pytorch.org/whl/cu118
|
||||
torch==2.4.1+cu118
|
||||
@ -1 +0,0 @@
|
||||
torch==2.4.1
|
||||
@ -1,2 +0,0 @@
|
||||
--extra-index-url https://download.pytorch.org/whl/rocm6.0
|
||||
torch==2.4.1+rocm6.0
|
||||
@ -1,6 +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
|
||||
@ -1,6 +0,0 @@
|
||||
accelerate
|
||||
auto-gptq==0.7.1
|
||||
grpcio==1.71.0
|
||||
protobuf
|
||||
certifi
|
||||
transformers
|
||||
@ -1,4 +0,0 @@
|
||||
#!/bin/bash
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
startBackend $@
|
||||
@ -1,6 +0,0 @@
|
||||
#!/bin/bash
|
||||
set -e
|
||||
|
||||
source $(dirname $0)/../common/libbackend.sh
|
||||
|
||||
runUnittests
|
||||
@ -184,11 +184,6 @@ func grpcModelOpts(c config.BackendConfig) *pb.ModelOptions {
|
||||
MainGPU: c.MainGPU,
|
||||
Threads: int32(*c.Threads),
|
||||
TensorSplit: c.TensorSplit,
|
||||
// AutoGPTQ
|
||||
ModelBaseName: c.AutoGPTQ.ModelBaseName,
|
||||
Device: c.AutoGPTQ.Device,
|
||||
UseTriton: c.AutoGPTQ.Triton,
|
||||
UseFastTokenizer: c.AutoGPTQ.UseFastTokenizer,
|
||||
// RWKV
|
||||
Tokenizer: c.Tokenizer,
|
||||
}
|
||||
|
||||
@ -50,9 +50,6 @@ type BackendConfig struct {
|
||||
// LLM configs (GPT4ALL, Llama.cpp, ...)
|
||||
LLMConfig `yaml:",inline"`
|
||||
|
||||
// AutoGPTQ specifics
|
||||
AutoGPTQ AutoGPTQ `yaml:"autogptq"`
|
||||
|
||||
// Diffusers
|
||||
Diffusers Diffusers `yaml:"diffusers"`
|
||||
Step int `yaml:"step"`
|
||||
@ -176,14 +173,6 @@ type LimitMMPerPrompt struct {
|
||||
LimitAudioPerPrompt int `yaml:"audio"`
|
||||
}
|
||||
|
||||
// AutoGPTQ is a struct that holds the configuration specific to the AutoGPTQ backend
|
||||
type AutoGPTQ struct {
|
||||
ModelBaseName string `yaml:"model_base_name"`
|
||||
Device string `yaml:"device"`
|
||||
Triton bool `yaml:"triton"`
|
||||
UseFastTokenizer bool `yaml:"use_fast_tokenizer"`
|
||||
}
|
||||
|
||||
// TemplateConfig is a struct that holds the configuration of the templating system
|
||||
type TemplateConfig struct {
|
||||
// Chat is the template used in the chat completion endpoint
|
||||
|
||||
@ -203,18 +203,10 @@ func mergeOpenAIRequestAndBackendConfig(config *config.BackendConfig, input *sch
|
||||
config.Diffusers.ClipSkip = input.ClipSkip
|
||||
}
|
||||
|
||||
if input.ModelBaseName != "" {
|
||||
config.AutoGPTQ.ModelBaseName = input.ModelBaseName
|
||||
}
|
||||
|
||||
if input.NegativePromptScale != 0 {
|
||||
config.NegativePromptScale = input.NegativePromptScale
|
||||
}
|
||||
|
||||
if input.UseFastTokenizer {
|
||||
config.UseFastTokenizer = input.UseFastTokenizer
|
||||
}
|
||||
|
||||
if input.NegativePrompt != "" {
|
||||
config.NegativePrompt = input.NegativePrompt
|
||||
}
|
||||
|
||||
@ -202,7 +202,6 @@ type OpenAIRequest struct {
|
||||
|
||||
Backend string `json:"backend" yaml:"backend"`
|
||||
|
||||
// AutoGPTQ
|
||||
ModelBaseName string `json:"model_base_name" yaml:"model_base_name"`
|
||||
}
|
||||
|
||||
|
||||
@ -41,8 +41,6 @@ type PredictionOptions struct {
|
||||
RopeFreqBase float32 `json:"rope_freq_base" yaml:"rope_freq_base"`
|
||||
RopeFreqScale float32 `json:"rope_freq_scale" yaml:"rope_freq_scale"`
|
||||
NegativePromptScale float32 `json:"negative_prompt_scale" yaml:"negative_prompt_scale"`
|
||||
// AutoGPTQ
|
||||
UseFastTokenizer bool `json:"use_fast_tokenizer" yaml:"use_fast_tokenizer"`
|
||||
|
||||
// Diffusers
|
||||
ClipSkip int `json:"clip_skip" yaml:"clip_skip"`
|
||||
|
||||
@ -268,14 +268,6 @@ yarn_ext_factor: 0
|
||||
yarn_attn_factor: 0
|
||||
yarn_beta_fast: 0
|
||||
yarn_beta_slow: 0
|
||||
|
||||
# AutoGPT-Q settings, for configurations specific to GPT models.
|
||||
autogptq:
|
||||
model_base_name: "" # Base name of the model.
|
||||
device: "" # Device to run the model on.
|
||||
triton: false # Whether to use Triton Inference Server.
|
||||
use_fast_tokenizer: false # Whether to use a fast tokenizer for quicker processing.
|
||||
|
||||
# configuration for diffusers model
|
||||
diffusers:
|
||||
cuda: false # Whether to use CUDA
|
||||
|
||||
@ -147,7 +147,6 @@ The devices in the following list have been tested with `hipblas` images running
|
||||
| diffusers | yes | Radeon VII (gfx906) |
|
||||
| piper | yes | Radeon VII (gfx906) |
|
||||
| whisper | no | none |
|
||||
| autogptq | no | none |
|
||||
| bark | no | none |
|
||||
| coqui | no | none |
|
||||
| transformers | no | none |
|
||||
|
||||
@ -74,49 +74,9 @@ curl http://localhost:8080/v1/models
|
||||
|
||||
## Backends
|
||||
|
||||
### AutoGPTQ
|
||||
|
||||
[AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ) is an easy-to-use LLMs quantization package with user-friendly apis, based on GPTQ algorithm.
|
||||
|
||||
#### Prerequisites
|
||||
|
||||
This is an extra backend - in the container images is already available and there is nothing to do for the setup.
|
||||
|
||||
If you are building LocalAI locally, you need to install [AutoGPTQ manually](https://github.com/PanQiWei/AutoGPTQ#quick-installation).
|
||||
|
||||
|
||||
#### Model setup
|
||||
|
||||
The models are automatically downloaded from `huggingface` if not present the first time. It is possible to define models via `YAML` config file, or just by querying the endpoint with the `huggingface` repository model name. For example, create a `YAML` config file in `models/`:
|
||||
|
||||
```
|
||||
name: orca
|
||||
backend: autogptq
|
||||
model_base_name: "orca_mini_v2_13b-GPTQ-4bit-128g.no-act.order"
|
||||
parameters:
|
||||
model: "TheBloke/orca_mini_v2_13b-GPTQ"
|
||||
# ...
|
||||
```
|
||||
|
||||
Test with:
|
||||
|
||||
```bash
|
||||
curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
|
||||
"model": "orca",
|
||||
"messages": [{"role": "user", "content": "How are you?"}],
|
||||
"temperature": 0.1
|
||||
}'
|
||||
```
|
||||
### RWKV
|
||||
|
||||
A full example on how to run a rwkv model is in the [examples](https://github.com/go-skynet/LocalAI/tree/master/examples/rwkv).
|
||||
|
||||
Note: rwkv models needs to specify the backend `rwkv` in the YAML config files and have an associated tokenizer along that needs to be provided with it:
|
||||
|
||||
```
|
||||
36464540 -rw-r--r-- 1 mudler mudler 1.2G May 3 10:51 rwkv_small
|
||||
36464543 -rw-r--r-- 1 mudler mudler 2.4M May 3 10:51 rwkv_small.tokenizer.json
|
||||
```
|
||||
RWKV support is available through llama.cpp (see below)
|
||||
|
||||
### llama.cpp
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user