LocalAI/pkg/model/loader.go
Ettore Di Giacinto 2c425e9c69
Some checks are pending
Explorer deployment / build-linux (push) Waiting to run
GPU tests / ubuntu-latest (1.21.x) (push) Waiting to run
generate and publish intel docker caches / generate_caches (intel/oneapi-basekit:2025.0.0-0-devel-ubuntu22.04, linux/amd64, ubuntu-latest) (push) Waiting to run
build container images / hipblas-jobs (-aio-gpu-hipblas, rocm/dev-ubuntu-22.04:6.1, hipblas, true, ubuntu:22.04, extras, latest-gpu-hipblas, latest-aio-gpu-hipblas, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -hipblas-ffmpeg) (push) Waiting to run
build container images / hipblas-jobs (rocm/dev-ubuntu-22.04:6.1, hipblas, false, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -hipblas-core) (push) Waiting to run
build container images / hipblas-jobs (rocm/dev-ubuntu-22.04:6.1, hipblas, false, ubuntu:22.04, extras, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -hipblas) (push) Waiting to run
build container images / hipblas-jobs (rocm/dev-ubuntu-22.04:6.1, hipblas, true, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -hipblas-ffmpeg-core) (push) Waiting to run
build container images / self-hosted-jobs (-aio-gpu-intel-f16, quay.io/go-skynet/intel-oneapi-base:latest, sycl_f16, true, ubuntu:22.04, extras, latest-gpu-intel-f16, latest-aio-gpu-intel-f16, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -sycl-f16-ffmpeg) (push) Waiting to run
build container images / self-hosted-jobs (-aio-gpu-intel-f32, quay.io/go-skynet/intel-oneapi-base:latest, sycl_f32, true, ubuntu:22.04, extras, latest-gpu-intel-f32, latest-aio-gpu-intel-f32, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -sycl-f32-ffmpeg) (push) Waiting to run
build container images / self-hosted-jobs (-aio-gpu-nvidia-cuda-11, ubuntu:22.04, cublas, 11, 7, true, extras, latest-gpu-nvidia-cuda-11, latest-aio-gpu-nvidia-cuda-11, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -cublas-cuda11-ffmpeg) (push) Waiting to run
build container images / self-hosted-jobs (-aio-gpu-nvidia-cuda-12, ubuntu:22.04, cublas, 12, 0, true, extras, latest-gpu-nvidia-cuda-12, latest-aio-gpu-nvidia-cuda-12, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -cublas-cuda12-ffmpeg) (push) Waiting to run
build container images / self-hosted-jobs (quay.io/go-skynet/intel-oneapi-base:latest, sycl_f16, false, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -sycl-f16-core) (push) Waiting to run
build container images / self-hosted-jobs (quay.io/go-skynet/intel-oneapi-base:latest, sycl_f16, true, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -sycl-f16-ffmpeg-core) (push) Waiting to run
build container images / self-hosted-jobs (quay.io/go-skynet/intel-oneapi-base:latest, sycl_f32, false, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -sycl-f32-core) (push) Waiting to run
build container images / self-hosted-jobs (quay.io/go-skynet/intel-oneapi-base:latest, sycl_f32, true, ubuntu:22.04, core, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -sycl-f32-ffmpeg-core) (push) Waiting to run
build container images / self-hosted-jobs (ubuntu:22.04, , , extras, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, ) (push) Waiting to run
build container images / self-hosted-jobs (ubuntu:22.04, , true, extras, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, auto, -ffmpeg) (push) Waiting to run
build container images / self-hosted-jobs (ubuntu:22.04, cublas, 11, 7, , extras, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -cublas-cuda11) (push) Waiting to run
build container images / self-hosted-jobs (ubuntu:22.04, cublas, 12, 0, , extras, --jobs=3 --output-sync=target, linux/amd64, arc-runner-set, false, -cublas-cuda12) (push) Waiting to run
build container images / core-image-build (-aio-cpu, ubuntu:22.04, , true, core, latest-cpu, latest-aio-cpu, --jobs=4 --output-sync=target, linux/amd64,linux/arm64, arc-runner-set, false, auto, -ffmpeg-core) (push) Waiting to run
build container images / core-image-build (ubuntu:22.04, cublas, 11, 7, , core, --jobs=4 --output-sync=target, linux/amd64, arc-runner-set, false, false, -cublas-cuda11-core) (push) Waiting to run
build container images / core-image-build (ubuntu:22.04, cublas, 11, 7, true, core, --jobs=4 --output-sync=target, linux/amd64, arc-runner-set, false, false, -cublas-cuda11-ffmpeg-core) (push) Waiting to run
build container images / core-image-build (ubuntu:22.04, cublas, 12, 0, , core, --jobs=4 --output-sync=target, linux/amd64, arc-runner-set, false, false, -cublas-cuda12-core) (push) Waiting to run
build container images / core-image-build (ubuntu:22.04, cublas, 12, 0, true, core, --jobs=4 --output-sync=target, linux/amd64, arc-runner-set, false, false, -cublas-cuda12-ffmpeg-core) (push) Waiting to run
build container images / core-image-build (ubuntu:22.04, vulkan, true, core, latest-vulkan-ffmpeg-core, --jobs=4 --output-sync=target, linux/amd64, arc-runner-set, false, false, -vulkan-ffmpeg-core) (push) Waiting to run
build container images / gh-runner (nvcr.io/nvidia/l4t-jetpack:r36.4.0, cublas, 12, 0, true, core, latest-nvidia-l4t-arm64-core, --jobs=4 --output-sync=target, linux/arm64, ubuntu-24.04-arm, true, false, -nvidia-l4t-arm64-core) (push) Waiting to run
Security Scan / tests (push) Waiting to run
Tests extras backends / tests-transformers (push) Waiting to run
Tests extras backends / tests-rerankers (push) Waiting to run
Tests extras backends / tests-diffusers (push) Waiting to run
Tests extras backends / tests-coqui (push) Waiting to run
tests / tests-linux (1.21.x) (push) Waiting to run
tests / tests-aio-container (push) Waiting to run
tests / tests-apple (1.21.x) (push) Waiting to run
feat(loader): enhance single active backend by treating as singleton (#5107)
feat(loader): enhance single active backend by treating at singleton

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
2025-04-01 20:58:11 +02:00

189 lines
4.1 KiB
Go

package model
import (
"context"
"fmt"
"os"
"path/filepath"
"strings"
"sync"
"time"
"github.com/mudler/LocalAI/pkg/utils"
"github.com/rs/zerolog/log"
)
// new idea: what if we declare a struct of these here, and use a loop to check?
// TODO: Split ModelLoader and TemplateLoader? Just to keep things more organized. Left together to share a mutex until I look into that. Would split if we seperate directories for .bin/.yaml and .tmpl
type ModelLoader struct {
ModelPath string
mu sync.Mutex
singletonLock sync.Mutex
singletonMode bool
models map[string]*Model
wd *WatchDog
}
func NewModelLoader(modelPath string, singleActiveBackend bool) *ModelLoader {
nml := &ModelLoader{
ModelPath: modelPath,
models: make(map[string]*Model),
singletonMode: singleActiveBackend,
}
return nml
}
func (ml *ModelLoader) SetWatchDog(wd *WatchDog) {
ml.wd = wd
}
func (ml *ModelLoader) ExistsInModelPath(s string) bool {
return utils.ExistsInPath(ml.ModelPath, s)
}
var knownFilesToSkip []string = []string{
"MODEL_CARD",
"README",
"README.md",
}
var knownModelsNameSuffixToSkip []string = []string{
".tmpl",
".keep",
".yaml",
".yml",
".json",
".txt",
".pt",
".onnx",
".md",
".MD",
".DS_Store",
".",
".safetensors",
".partial",
".tar.gz",
}
const retryTimeout = time.Duration(2 * time.Minute)
func (ml *ModelLoader) ListFilesInModelPath() ([]string, error) {
files, err := os.ReadDir(ml.ModelPath)
if err != nil {
return []string{}, err
}
models := []string{}
FILE:
for _, file := range files {
for _, skip := range knownFilesToSkip {
if strings.EqualFold(file.Name(), skip) {
continue FILE
}
}
// Skip templates, YAML, .keep, .json, and .DS_Store files
for _, skip := range knownModelsNameSuffixToSkip {
if strings.HasSuffix(file.Name(), skip) {
continue FILE
}
}
// Skip directories
if file.IsDir() {
continue
}
models = append(models, file.Name())
}
return models, nil
}
func (ml *ModelLoader) ListModels() []*Model {
ml.mu.Lock()
defer ml.mu.Unlock()
models := []*Model{}
for _, model := range ml.models {
models = append(models, model)
}
return models
}
func (ml *ModelLoader) LoadModel(modelID, modelName string, loader func(string, string, string) (*Model, error)) (*Model, error) {
// Check if we already have a loaded model
if model := ml.CheckIsLoaded(modelID); model != nil {
return model, nil
}
// Load the model and keep it in memory for later use
modelFile := filepath.Join(ml.ModelPath, modelName)
log.Debug().Msgf("Loading model in memory from file: %s", modelFile)
ml.mu.Lock()
defer ml.mu.Unlock()
model, err := loader(modelID, modelName, modelFile)
if err != nil {
return nil, fmt.Errorf("failed to load model with internal loader: %s", err)
}
if model == nil {
return nil, fmt.Errorf("loader didn't return a model")
}
ml.models[modelID] = model
return model, nil
}
func (ml *ModelLoader) ShutdownModel(modelName string) error {
ml.mu.Lock()
defer ml.mu.Unlock()
return ml.deleteProcess(modelName)
}
func (ml *ModelLoader) CheckIsLoaded(s string) *Model {
ml.mu.Lock()
defer ml.mu.Unlock()
m, ok := ml.models[s]
if !ok {
return nil
}
log.Debug().Msgf("Model already loaded in memory: %s", s)
client := m.GRPC(false, ml.wd)
log.Debug().Msgf("Checking model availability (%s)", s)
cTimeout, cancel := context.WithTimeout(context.Background(), 2*time.Minute)
defer cancel()
alive, err := client.HealthCheck(cTimeout)
if !alive {
log.Warn().Msgf("GRPC Model not responding: %s", err.Error())
log.Warn().Msgf("Deleting the process in order to recreate it")
process := m.Process()
if process == nil {
log.Error().Msgf("Process not found for '%s' and the model is not responding anymore !", s)
return m
}
if !process.IsAlive() {
log.Debug().Msgf("GRPC Process is not responding: %s", s)
// stop and delete the process, this forces to re-load the model and re-create again the service
err := ml.deleteProcess(s)
if err != nil {
log.Error().Err(err).Str("process", s).Msg("error stopping process")
}
return nil
}
}
return m
}