Revert "[Refactor]: Core/API Split" (#1550)

Revert "[Refactor]: Core/API Split (#1506)"

This reverts commit ab7b4d5ee9.
This commit is contained in:
Ettore Di Giacinto 2024-01-05 12:04:46 -05:00 committed by GitHub
parent ab7b4d5ee9
commit db926896bd
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
77 changed files with 3132 additions and 3456 deletions

359
api/config/config.go Normal file
View file

@ -0,0 +1,359 @@
package api_config
import (
"errors"
"fmt"
"io/fs"
"os"
"path/filepath"
"strings"
"sync"
"github.com/go-skynet/LocalAI/pkg/utils"
"github.com/rs/zerolog/log"
"gopkg.in/yaml.v3"
)
type Config struct {
PredictionOptions `yaml:"parameters"`
Name string `yaml:"name"`
F16 bool `yaml:"f16"`
Threads int `yaml:"threads"`
Debug bool `yaml:"debug"`
Roles map[string]string `yaml:"roles"`
Embeddings bool `yaml:"embeddings"`
Backend string `yaml:"backend"`
TemplateConfig TemplateConfig `yaml:"template"`
PromptStrings, InputStrings []string `yaml:"-"`
InputToken [][]int `yaml:"-"`
functionCallString, functionCallNameString string `yaml:"-"`
FunctionsConfig Functions `yaml:"function"`
FeatureFlag FeatureFlag `yaml:"feature_flags"` // Feature Flag registry. We move fast, and features may break on a per model/backend basis. Registry for (usually temporary) flags that indicate aborting something early.
// LLM configs (GPT4ALL, Llama.cpp, ...)
LLMConfig `yaml:",inline"`
// AutoGPTQ specifics
AutoGPTQ AutoGPTQ `yaml:"autogptq"`
// Diffusers
Diffusers Diffusers `yaml:"diffusers"`
Step int `yaml:"step"`
// GRPC Options
GRPC GRPC `yaml:"grpc"`
// Vall-e-x
VallE VallE `yaml:"vall-e"`
// CUDA
// Explicitly enable CUDA or not (some backends might need it)
CUDA bool `yaml:"cuda"`
DownloadFiles []File `yaml:"download_files"`
}
type File struct {
Filename string `yaml:"filename" json:"filename"`
SHA256 string `yaml:"sha256" json:"sha256"`
URI string `yaml:"uri" json:"uri"`
}
type VallE struct {
AudioPath string `yaml:"audio_path"`
}
type FeatureFlag map[string]*bool
func (ff FeatureFlag) Enabled(s string) bool {
v, exist := ff[s]
return exist && v != nil && *v
}
type GRPC struct {
Attempts int `yaml:"attempts"`
AttemptsSleepTime int `yaml:"attempts_sleep_time"`
}
type Diffusers struct {
CUDA bool `yaml:"cuda"`
PipelineType string `yaml:"pipeline_type"`
SchedulerType string `yaml:"scheduler_type"`
EnableParameters string `yaml:"enable_parameters"` // A list of comma separated parameters to specify
CFGScale float32 `yaml:"cfg_scale"` // Classifier-Free Guidance Scale
IMG2IMG bool `yaml:"img2img"` // Image to Image Diffuser
ClipSkip int `yaml:"clip_skip"` // Skip every N frames
ClipModel string `yaml:"clip_model"` // Clip model to use
ClipSubFolder string `yaml:"clip_subfolder"` // Subfolder to use for clip model
ControlNet string `yaml:"control_net"`
}
type LLMConfig struct {
SystemPrompt string `yaml:"system_prompt"`
TensorSplit string `yaml:"tensor_split"`
MainGPU string `yaml:"main_gpu"`
RMSNormEps float32 `yaml:"rms_norm_eps"`
NGQA int32 `yaml:"ngqa"`
PromptCachePath string `yaml:"prompt_cache_path"`
PromptCacheAll bool `yaml:"prompt_cache_all"`
PromptCacheRO bool `yaml:"prompt_cache_ro"`
MirostatETA float64 `yaml:"mirostat_eta"`
MirostatTAU float64 `yaml:"mirostat_tau"`
Mirostat int `yaml:"mirostat"`
NGPULayers int `yaml:"gpu_layers"`
MMap bool `yaml:"mmap"`
MMlock bool `yaml:"mmlock"`
LowVRAM bool `yaml:"low_vram"`
Grammar string `yaml:"grammar"`
StopWords []string `yaml:"stopwords"`
Cutstrings []string `yaml:"cutstrings"`
TrimSpace []string `yaml:"trimspace"`
TrimSuffix []string `yaml:"trimsuffix"`
ContextSize int `yaml:"context_size"`
NUMA bool `yaml:"numa"`
LoraAdapter string `yaml:"lora_adapter"`
LoraBase string `yaml:"lora_base"`
LoraScale float32 `yaml:"lora_scale"`
NoMulMatQ bool `yaml:"no_mulmatq"`
DraftModel string `yaml:"draft_model"`
NDraft int32 `yaml:"n_draft"`
Quantization string `yaml:"quantization"`
MMProj string `yaml:"mmproj"`
RopeScaling string `yaml:"rope_scaling"`
YarnExtFactor float32 `yaml:"yarn_ext_factor"`
YarnAttnFactor float32 `yaml:"yarn_attn_factor"`
YarnBetaFast float32 `yaml:"yarn_beta_fast"`
YarnBetaSlow float32 `yaml:"yarn_beta_slow"`
}
type AutoGPTQ struct {
ModelBaseName string `yaml:"model_base_name"`
Device string `yaml:"device"`
Triton bool `yaml:"triton"`
UseFastTokenizer bool `yaml:"use_fast_tokenizer"`
}
type Functions struct {
DisableNoAction bool `yaml:"disable_no_action"`
NoActionFunctionName string `yaml:"no_action_function_name"`
NoActionDescriptionName string `yaml:"no_action_description_name"`
}
type TemplateConfig struct {
Chat string `yaml:"chat"`
ChatMessage string `yaml:"chat_message"`
Completion string `yaml:"completion"`
Edit string `yaml:"edit"`
Functions string `yaml:"function"`
}
type ConfigLoader struct {
configs map[string]Config
sync.Mutex
}
func (c *Config) SetFunctionCallString(s string) {
c.functionCallString = s
}
func (c *Config) SetFunctionCallNameString(s string) {
c.functionCallNameString = s
}
func (c *Config) ShouldUseFunctions() bool {
return ((c.functionCallString != "none" || c.functionCallString == "") || c.ShouldCallSpecificFunction())
}
func (c *Config) ShouldCallSpecificFunction() bool {
return len(c.functionCallNameString) > 0
}
func (c *Config) FunctionToCall() string {
return c.functionCallNameString
}
func defaultPredictOptions(modelFile string) PredictionOptions {
return PredictionOptions{
TopP: 0.7,
TopK: 80,
Maxtokens: 512,
Temperature: 0.9,
Model: modelFile,
}
}
func DefaultConfig(modelFile string) *Config {
return &Config{
PredictionOptions: defaultPredictOptions(modelFile),
}
}
func NewConfigLoader() *ConfigLoader {
return &ConfigLoader{
configs: make(map[string]Config),
}
}
func ReadConfigFile(file string) ([]*Config, error) {
c := &[]*Config{}
f, err := os.ReadFile(file)
if err != nil {
return nil, fmt.Errorf("cannot read config file: %w", err)
}
if err := yaml.Unmarshal(f, c); err != nil {
return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
}
return *c, nil
}
func ReadConfig(file string) (*Config, error) {
c := &Config{}
f, err := os.ReadFile(file)
if err != nil {
return nil, fmt.Errorf("cannot read config file: %w", err)
}
if err := yaml.Unmarshal(f, c); err != nil {
return nil, fmt.Errorf("cannot unmarshal config file: %w", err)
}
return c, nil
}
func (cm *ConfigLoader) LoadConfigFile(file string) error {
cm.Lock()
defer cm.Unlock()
c, err := ReadConfigFile(file)
if err != nil {
return fmt.Errorf("cannot load config file: %w", err)
}
for _, cc := range c {
cm.configs[cc.Name] = *cc
}
return nil
}
func (cm *ConfigLoader) LoadConfig(file string) error {
cm.Lock()
defer cm.Unlock()
c, err := ReadConfig(file)
if err != nil {
return fmt.Errorf("cannot read config file: %w", err)
}
cm.configs[c.Name] = *c
return nil
}
func (cm *ConfigLoader) GetConfig(m string) (Config, bool) {
cm.Lock()
defer cm.Unlock()
v, exists := cm.configs[m]
return v, exists
}
func (cm *ConfigLoader) GetAllConfigs() []Config {
cm.Lock()
defer cm.Unlock()
var res []Config
for _, v := range cm.configs {
res = append(res, v)
}
return res
}
func (cm *ConfigLoader) ListConfigs() []string {
cm.Lock()
defer cm.Unlock()
var res []string
for k := range cm.configs {
res = append(res, k)
}
return res
}
// Preload prepare models if they are not local but url or huggingface repositories
func (cm *ConfigLoader) Preload(modelPath string) error {
cm.Lock()
defer cm.Unlock()
status := func(fileName, current, total string, percent float64) {
utils.DisplayDownloadFunction(fileName, current, total, percent)
}
log.Info().Msgf("Preloading models from %s", modelPath)
for i, config := range cm.configs {
// Download files and verify their SHA
for _, file := range config.DownloadFiles {
log.Debug().Msgf("Checking %q exists and matches SHA", file.Filename)
if err := utils.VerifyPath(file.Filename, modelPath); err != nil {
return err
}
// Create file path
filePath := filepath.Join(modelPath, file.Filename)
if err := utils.DownloadFile(file.URI, filePath, file.SHA256, status); err != nil {
return err
}
}
modelURL := config.PredictionOptions.Model
modelURL = utils.ConvertURL(modelURL)
if utils.LooksLikeURL(modelURL) {
// md5 of model name
md5Name := utils.MD5(modelURL)
// check if file exists
if _, err := os.Stat(filepath.Join(modelPath, md5Name)); errors.Is(err, os.ErrNotExist) {
err := utils.DownloadFile(modelURL, filepath.Join(modelPath, md5Name), "", status)
if err != nil {
return err
}
}
cc := cm.configs[i]
c := &cc
c.PredictionOptions.Model = md5Name
cm.configs[i] = *c
}
}
return nil
}
func (cm *ConfigLoader) LoadConfigs(path string) error {
cm.Lock()
defer cm.Unlock()
entries, err := os.ReadDir(path)
if err != nil {
return err
}
files := make([]fs.FileInfo, 0, len(entries))
for _, entry := range entries {
info, err := entry.Info()
if err != nil {
return err
}
files = append(files, info)
}
for _, file := range files {
// Skip templates, YAML and .keep files
if !strings.Contains(file.Name(), ".yaml") && !strings.Contains(file.Name(), ".yml") {
continue
}
c, err := ReadConfig(filepath.Join(path, file.Name()))
if err == nil {
cm.configs[c.Name] = *c
}
}
return nil
}

56
api/config/config_test.go Normal file
View file

@ -0,0 +1,56 @@
package api_config_test
import (
"os"
. "github.com/go-skynet/LocalAI/api/config"
"github.com/go-skynet/LocalAI/api/options"
"github.com/go-skynet/LocalAI/pkg/model"
. "github.com/onsi/ginkgo/v2"
. "github.com/onsi/gomega"
)
var _ = Describe("Test cases for config related functions", func() {
var (
configFile string
)
Context("Test Read configuration functions", func() {
configFile = os.Getenv("CONFIG_FILE")
It("Test ReadConfigFile", func() {
config, err := ReadConfigFile(configFile)
Expect(err).To(BeNil())
Expect(config).ToNot(BeNil())
// two configs in config.yaml
Expect(config[0].Name).To(Equal("list1"))
Expect(config[1].Name).To(Equal("list2"))
})
It("Test LoadConfigs", func() {
cm := NewConfigLoader()
opts := options.NewOptions()
modelLoader := model.NewModelLoader(os.Getenv("MODELS_PATH"))
options.WithModelLoader(modelLoader)(opts)
err := cm.LoadConfigs(opts.Loader.ModelPath)
Expect(err).To(BeNil())
Expect(cm.ListConfigs()).ToNot(BeNil())
// config should includes gpt4all models's api.config
Expect(cm.ListConfigs()).To(ContainElements("gpt4all"))
// config should includes gpt2 models's api.config
Expect(cm.ListConfigs()).To(ContainElements("gpt4all-2"))
// config should includes text-embedding-ada-002 models's api.config
Expect(cm.ListConfigs()).To(ContainElements("text-embedding-ada-002"))
// config should includes rwkv_test models's api.config
Expect(cm.ListConfigs()).To(ContainElements("rwkv_test"))
// config should includes whisper-1 models's api.config
Expect(cm.ListConfigs()).To(ContainElements("whisper-1"))
})
})
})

50
api/config/prediction.go Normal file
View file

@ -0,0 +1,50 @@
package api_config
type PredictionOptions struct {
// Also part of the OpenAI official spec
Model string `json:"model" yaml:"model"`
// Also part of the OpenAI official spec
Language string `json:"language"`
// Also part of the OpenAI official spec. use it for returning multiple results
N int `json:"n"`
// Common options between all the API calls, part of the OpenAI spec
TopP float64 `json:"top_p" yaml:"top_p"`
TopK int `json:"top_k" yaml:"top_k"`
Temperature float64 `json:"temperature" yaml:"temperature"`
Maxtokens int `json:"max_tokens" yaml:"max_tokens"`
Echo bool `json:"echo"`
// Custom parameters - not present in the OpenAI API
Batch int `json:"batch" yaml:"batch"`
F16 bool `json:"f16" yaml:"f16"`
IgnoreEOS bool `json:"ignore_eos" yaml:"ignore_eos"`
RepeatPenalty float64 `json:"repeat_penalty" yaml:"repeat_penalty"`
Keep int `json:"n_keep" yaml:"n_keep"`
MirostatETA float64 `json:"mirostat_eta" yaml:"mirostat_eta"`
MirostatTAU float64 `json:"mirostat_tau" yaml:"mirostat_tau"`
Mirostat int `json:"mirostat" yaml:"mirostat"`
FrequencyPenalty float64 `json:"frequency_penalty" yaml:"frequency_penalty"`
TFZ float64 `json:"tfz" yaml:"tfz"`
TypicalP float64 `json:"typical_p" yaml:"typical_p"`
Seed int `json:"seed" yaml:"seed"`
NegativePrompt string `json:"negative_prompt" yaml:"negative_prompt"`
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"`
// RWKV (?)
Tokenizer string `json:"tokenizer" yaml:"tokenizer"`
}