refactor: drop code dups (#234)

README.md

@@ -73,7 +73,7 @@ Note: You might need to convert older models to the new format, see [here](https
 
 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 have an associated tokenizer along that needs to be provided with it:
+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
@@ -545,6 +545,7 @@ name: text-embedding-ada-002
 parameters:
   model: bert
 embeddings: true
+backend: "bert-embeddings"
 ```
 
 There is an example available [here](https://github.com/go-skynet/LocalAI/tree/master/examples/query_data/).
@@ -563,6 +564,7 @@ Download one of the models from https://huggingface.co/ggerganov/whisper.cpp/tre
 
 ```yaml
 name: whisper-1
+backend: whisper
 parameters:
   model: whisper-en
 ```

api/api_test.go

@@ -79,7 +79,7 @@ var _ = Describe("API test", func() {
 		It("returns errors", func() {
 			_, err := client.CreateCompletion(context.TODO(), openai.CompletionRequest{Model: "foomodel", Prompt: "abcdedfghikl"})
 			Expect(err).To(HaveOccurred())
-			Expect(err.Error()).To(ContainSubstring("error, status code: 500, message: could not load model - all backends returned error: 10 errors occurred:"))
+			Expect(err.Error()).To(ContainSubstring("error, status code: 500, message: could not load model - all backends returned error: 9 errors occurred:"))
 		})
 
 	})

api/config.go

@@ -285,5 +285,10 @@ func readConfig(cm ConfigMerger, c *fiber.Ctx, loader *model.ModelLoader, debug
 		}
 	}
 
+	// Enforce debug flag if passed from CLI
+	if debug {
+		config.Debug = true
+	}
+
 	return config, input, nil
 }

api/openai.go

@@ -12,8 +12,10 @@ import (
 	"path/filepath"
 	"strings"
 
+	"github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
 	model "github.com/go-skynet/LocalAI/pkg/model"
-	"github.com/go-skynet/LocalAI/pkg/whisper"
+	whisperutil "github.com/go-skynet/LocalAI/pkg/whisper"
+	llama "github.com/go-skynet/go-llama.cpp"
 	"github.com/gofiber/fiber/v2"
 	"github.com/rs/zerolog/log"
 	"github.com/valyala/fasthttp"
@@ -436,12 +438,14 @@ func transcriptEndpoint(cm ConfigMerger, debug bool, loader *model.ModelLoader,
 
 		log.Debug().Msgf("Audio file copied to: %+v", dst)
 
-		whisperModel, err := loader.WhisperLoader("whisper", config.Model)
+		whisperModel, err := loader.BackendLoader("whisper", config.Model, []llama.ModelOption{}, uint32(config.Threads))
 		if err != nil {
 			return c.Status(http.StatusBadRequest).JSON(fiber.Map{"error": err.Error()})
 		}
 
-		tr, err := whisper.Transcript(whisperModel, dst, input.Language)
+		w := whisperModel.(whisper.Model)
+
+		tr, err := whisperutil.Transcript(w, dst, input.Language)
 		if err != nil {
 			return c.Status(http.StatusBadRequest).JSON(fiber.Map{"error": err.Error()})
 		}

pkg/model/initializers.go

@@ -0,0 +1,158 @@
+package model
+
+import (
+	"fmt"
+	"strings"
+
+	rwkv "github.com/donomii/go-rwkv.cpp"
+	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
+	bloomz "github.com/go-skynet/bloomz.cpp"
+	bert "github.com/go-skynet/go-bert.cpp"
+	gpt2 "github.com/go-skynet/go-gpt2.cpp"
+	llama "github.com/go-skynet/go-llama.cpp"
+	"github.com/hashicorp/go-multierror"
+	gpt4all "github.com/nomic/gpt4all/gpt4all-bindings/golang"
+	"github.com/rs/zerolog/log"
+)
+
+const tokenizerSuffix = ".tokenizer.json"
+
+const (
+	LlamaBackend          = "llama"
+	BloomzBackend         = "bloomz"
+	StableLMBackend       = "stablelm"
+	DollyBackend          = "dolly"
+	RedPajamaBackend      = "redpajama"
+	Gpt2Backend           = "gpt2"
+	Gpt4AllLlamaBackend   = "gpt4all-llama"
+	Gpt4AllMptBackend     = "gpt4all-mpt"
+	Gpt4AllJBackend       = "gpt4all-j"
+	BertEmbeddingsBackend = "bert-embeddings"
+	RwkvBackend           = "rwkv"
+	WhisperBackend        = "whisper"
+)
+
+var backends []string = []string{
+	LlamaBackend,
+	Gpt4AllLlamaBackend,
+	Gpt4AllMptBackend,
+	Gpt4AllJBackend,
+	Gpt2Backend,
+	WhisperBackend,
+	RwkvBackend,
+	BloomzBackend,
+	StableLMBackend,
+	DollyBackend,
+	RedPajamaBackend,
+	BertEmbeddingsBackend,
+}
+
+var redPajama = func(modelFile string) (interface{}, error) {
+	return gpt2.NewRedPajama(modelFile)
+}
+
+var dolly = func(modelFile string) (interface{}, error) {
+	return gpt2.NewDolly(modelFile)
+}
+
+var stableLM = func(modelFile string) (interface{}, error) {
+	return gpt2.NewStableLM(modelFile)
+}
+
+var bertEmbeddings = func(modelFile string) (interface{}, error) {
+	return bert.New(modelFile)
+}
+
+var bloomzLM = func(modelFile string) (interface{}, error) {
+	return bloomz.New(modelFile)
+}
+var gpt2LM = func(modelFile string) (interface{}, error) {
+	return gpt2.New(modelFile)
+}
+
+var whisperModel = func(modelFile string) (interface{}, error) {
+	return whisper.New(modelFile)
+}
+
+func llamaLM(opts ...llama.ModelOption) func(string) (interface{}, error) {
+	return func(s string) (interface{}, error) {
+		return llama.New(s, opts...)
+	}
+}
+
+func gpt4allLM(opts ...gpt4all.ModelOption) func(string) (interface{}, error) {
+	return func(s string) (interface{}, error) {
+		return gpt4all.New(s, opts...)
+	}
+}
+
+func rwkvLM(tokenFile string, threads uint32) func(string) (interface{}, error) {
+	return func(s string) (interface{}, error) {
+		model := rwkv.LoadFiles(s, tokenFile, threads)
+		if model == nil {
+			return nil, fmt.Errorf("could not load model")
+		}
+		return model, nil
+	}
+}
+
+func (ml *ModelLoader) BackendLoader(backendString string, modelFile string, llamaOpts []llama.ModelOption, threads uint32) (model interface{}, err error) {
+	switch strings.ToLower(backendString) {
+	case LlamaBackend:
+		return ml.LoadModel(modelFile, llamaLM(llamaOpts...))
+	case BloomzBackend:
+		return ml.LoadModel(modelFile, bloomzLM)
+	case StableLMBackend:
+		return ml.LoadModel(modelFile, stableLM)
+	case DollyBackend:
+		return ml.LoadModel(modelFile, dolly)
+	case RedPajamaBackend:
+		return ml.LoadModel(modelFile, redPajama)
+	case Gpt2Backend:
+		return ml.LoadModel(modelFile, gpt2LM)
+	case Gpt4AllLlamaBackend:
+		return ml.LoadModel(modelFile, gpt4allLM(gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.LLaMAType)))
+	case Gpt4AllMptBackend:
+		return ml.LoadModel(modelFile, gpt4allLM(gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.MPTType)))
+	case Gpt4AllJBackend:
+		return ml.LoadModel(modelFile, gpt4allLM(gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.GPTJType)))
+	case BertEmbeddingsBackend:
+		return ml.LoadModel(modelFile, bertEmbeddings)
+	case RwkvBackend:
+		return ml.LoadModel(modelFile, rwkvLM(modelFile+tokenizerSuffix, threads))
+	case WhisperBackend:
+		return ml.LoadModel(modelFile, whisperModel)
+	default:
+		return nil, fmt.Errorf("backend unsupported: %s", backendString)
+	}
+}
+
+func (ml *ModelLoader) GreedyLoader(modelFile string, llamaOpts []llama.ModelOption, threads uint32) (interface{}, error) {
+	log.Debug().Msgf("Loading models greedly")
+
+	ml.mu.Lock()
+	m, exists := ml.models[modelFile]
+	if exists {
+		ml.mu.Unlock()
+		return m, nil
+	}
+	ml.mu.Unlock()
+	var err error
+
+	for _, b := range backends {
+		if b == BloomzBackend || b == WhisperBackend || b == RwkvBackend { // do not autoload bloomz/whisper/rwkv
+			continue
+		}
+		log.Debug().Msgf("[%s] Attempting to load", b)
+		model, modelerr := ml.BackendLoader(b, modelFile, llamaOpts, threads)
+		if modelerr == nil && model != nil {
+			log.Debug().Msgf("[%s] Loads OK", b)
+			return model, nil
+		} else if modelerr != nil {
+			err = multierror.Append(err, modelerr)
+			log.Debug().Msgf("[%s] Fails: %s", b, modelerr.Error())
+		}
+	}
+
+	return nil, fmt.Errorf("could not load model - all backends returned error: %s", err.Error())
+}

pkg/model/loader.go

@@ -10,14 +10,6 @@ import (
 	"sync"
 	"text/template"
 
-	rwkv "github.com/donomii/go-rwkv.cpp"
-	whisper "github.com/ggerganov/whisper.cpp/bindings/go/pkg/whisper"
-	bloomz "github.com/go-skynet/bloomz.cpp"
-	bert "github.com/go-skynet/go-bert.cpp"
-	gpt2 "github.com/go-skynet/go-gpt2.cpp"
-	llama "github.com/go-skynet/go-llama.cpp"
-	"github.com/hashicorp/go-multierror"
-	gpt4all "github.com/nomic/gpt4all/gpt4all-bindings/golang"
 	"github.com/rs/zerolog/log"
 )
 
@@ -25,33 +17,15 @@ type ModelLoader struct {
 	ModelPath string
 	mu        sync.Mutex
 	// TODO: this needs generics
-	models            map[string]*llama.LLama
+	models           map[string]interface{}
-	gptmodels         map[string]*gpt4all.Model
+	promptsTemplates map[string]*template.Template
-	gpt2models        map[string]*gpt2.GPT2
-	gptstablelmmodels map[string]*gpt2.StableLM
-	dollymodels       map[string]*gpt2.Dolly
-	redpajama         map[string]*gpt2.RedPajama
-	rwkv              map[string]*rwkv.RwkvState
-	bloomz            map[string]*bloomz.Bloomz
-	bert              map[string]*bert.Bert
-	promptsTemplates  map[string]*template.Template
-	whisperModels     map[string]whisper.Model
 }
 
 func NewModelLoader(modelPath string) *ModelLoader {
 	return &ModelLoader{
-		ModelPath:         modelPath,
+		ModelPath:        modelPath,
-		gpt2models:        make(map[string]*gpt2.GPT2),
+		models:           make(map[string]interface{}),
-		gptmodels:         make(map[string]*gpt4all.Model),
+		promptsTemplates: make(map[string]*template.Template),
-		gptstablelmmodels: make(map[string]*gpt2.StableLM),
-		dollymodels:       make(map[string]*gpt2.Dolly),
-		redpajama:         make(map[string]*gpt2.RedPajama),
-		models:            make(map[string]*llama.LLama),
-		rwkv:              make(map[string]*rwkv.RwkvState),
-		bloomz:            make(map[string]*bloomz.Bloomz),
-		bert:              make(map[string]*bert.Bert),
-		promptsTemplates:  make(map[string]*template.Template),
-		whisperModels:     make(map[string]whisper.Model),
 	}
 }
 
@@ -136,271 +110,11 @@ func (ml *ModelLoader) loadTemplateIfExists(modelName, modelFile string) error {
 	return nil
 }
 
-func (ml *ModelLoader) LoadRedPajama(modelName string) (*gpt2.RedPajama, error) {
+func (ml *ModelLoader) LoadModel(modelName string, loader func(string) (interface{}, error)) (interface{}, error) {
 	ml.mu.Lock()
 	defer ml.mu.Unlock()
 
 	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.redpajama[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := gpt2.NewRedPajama(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.redpajama[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadDollyModel(modelName string) (*gpt2.Dolly, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.dollymodels[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := gpt2.NewDolly(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.dollymodels[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadStableLMModel(modelName string) (*gpt2.StableLM, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.gptstablelmmodels[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := gpt2.NewStableLM(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.gptstablelmmodels[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadBERT(modelName string) (*bert.Bert, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.bert[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := bert.New(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.bert[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadBloomz(modelName string) (*bloomz.Bloomz, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.bloomz[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := bloomz.New(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.bloomz[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadGPT2Model(modelName string) (*gpt2.GPT2, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.gpt2models[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := gpt2.New(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.gpt2models[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadGPT4AllModel(modelName string, opts ...gpt4all.ModelOption) (*gpt4all.Model, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.gptmodels[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := gpt4all.New(modelFile, opts...)
-	if err != nil {
-		return nil, err
-	}
-
-	// If there is a prompt template, load it
-	if err := ml.loadTemplateIfExists(modelName, modelFile); err != nil {
-		return nil, err
-	}
-
-	ml.gptmodels[modelName] = model
-	return model, err
-}
-
-func (ml *ModelLoader) LoadRWKV(modelName, tokenFile string, threads uint32) (*rwkv.RwkvState, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	log.Debug().Msgf("Loading model name: %s", modelName)
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
-	if m, ok := ml.rwkv[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, nil
-	}
-
-	// Load the model and keep it in memory for later use
-	modelFile := filepath.Join(ml.ModelPath, modelName)
-	tokenPath := filepath.Join(ml.ModelPath, tokenFile)
-	log.Debug().Msgf("Loading model in memory from file: %s", modelFile)
-
-	model := rwkv.LoadFiles(modelFile, tokenPath, threads)
-	if model == nil {
-		return nil, fmt.Errorf("could not load model")
-	}
-
-	ml.rwkv[modelName] = model
-	return model, nil
-}
-
-func (ml *ModelLoader) LoadLLaMAModel(modelName string, opts ...llama.ModelOption) (*llama.LLama, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	log.Debug().Msgf("Loading model name: %s", modelName)
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist")
-	}
-
 	if m, ok := ml.models[modelName]; ok {
 		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
 		return m, nil
@@ -410,7 +124,7 @@ func (ml *ModelLoader) LoadLLaMAModel(modelName string, opts ...llama.ModelOptio
 	modelFile := filepath.Join(ml.ModelPath, modelName)
 	log.Debug().Msgf("Loading model in memory from file: %s", modelFile)
 
-	model, err := llama.New(modelFile, opts...)
+	model, err := loader(modelFile)
 	if err != nil {
 		return nil, err
 	}
@@ -421,182 +135,5 @@ func (ml *ModelLoader) LoadLLaMAModel(modelName string, opts ...llama.ModelOptio
 	}
 
 	ml.models[modelName] = model
-	return model, err
+	return model, nil
-}
-
-func (ml *ModelLoader) LoadWhisperModel(modelName string) (whisper.Model, error) {
-	ml.mu.Lock()
-	defer ml.mu.Unlock()
-
-	// Check if we already have a loaded model
-	if !ml.ExistsInModelPath(modelName) {
-		return nil, fmt.Errorf("model does not exist -- %s", modelName)
-	}
-
-	if m, ok := ml.whisperModels[modelName]; ok {
-		log.Debug().Msgf("Model already loaded in memory: %s", modelName)
-		return m, 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)
-
-	model, err := whisper.New(modelFile)
-	if err != nil {
-		return nil, err
-	}
-
-	ml.whisperModels[modelName] = model
-	return model, err
-}
-
-const tokenizerSuffix = ".tokenizer.json"
-
-var loadedModels map[string]interface{} = map[string]interface{}{}
-var muModels sync.Mutex
-
-func (ml *ModelLoader) BackendLoader(backendString string, modelFile string, llamaOpts []llama.ModelOption, threads uint32) (model interface{}, err error) {
-	switch strings.ToLower(backendString) {
-	case "llama":
-		return ml.LoadLLaMAModel(modelFile, llamaOpts...)
-	case "bloomz":
-		return ml.LoadBloomz(modelFile)
-	case "stablelm":
-		return ml.LoadStableLMModel(modelFile)
-	case "dolly":
-		return ml.LoadDollyModel(modelFile)
-	case "redpajama":
-		return ml.LoadRedPajama(modelFile)
-	case "gpt2":
-		return ml.LoadGPT2Model(modelFile)
-	case "gpt4all-llama":
-		return ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.LLaMAType))
-	case "gpt4all-mpt":
-		return ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.MPTType))
-	case "gpt4all-j":
-		return ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.GPTJType))
-	case "bert-embeddings":
-		return ml.LoadBERT(modelFile)
-	case "rwkv":
-		return ml.LoadRWKV(modelFile, modelFile+tokenizerSuffix, threads)
-	default:
-		return nil, fmt.Errorf("backend unsupported: %s", backendString)
-	}
-}
-
-func (ml *ModelLoader) WhisperLoader(backendString string, modelFile string) (model whisper.Model, err error) {
-	//TODO expose more whisper options in next PR
-	switch strings.ToLower(backendString) {
-	case "whisper":
-		return ml.LoadWhisperModel(modelFile)
-	default:
-		return nil, fmt.Errorf("whisper backend unsupported: %s", backendString)
-	}
-}
-
-func (ml *ModelLoader) GreedyLoader(modelFile string, llamaOpts []llama.ModelOption, threads uint32) (model interface{}, err error) {
-	updateModels := func(model interface{}) {
-		muModels.Lock()
-		defer muModels.Unlock()
-		loadedModels[modelFile] = model
-	}
-
-	muModels.Lock()
-	m, exists := loadedModels[modelFile]
-	if exists {
-		muModels.Unlock()
-		return m, nil
-	}
-	muModels.Unlock()
-
-	model, modelerr := ml.LoadLLaMAModel(modelFile, llamaOpts...)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.GPTJType))
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.LLaMAType))
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadGPT4AllModel(modelFile, gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.MPTType))
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadGPT2Model(modelFile)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadStableLMModel(modelFile)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadDollyModel(modelFile)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadRedPajama(modelFile)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-	// Do not autoload bloomz
-	//model, modelerr = ml.LoadBloomz(modelFile)
-	//if modelerr == nil {
-	//	updateModels(model)
-	//	return model, nil
-	//} else {
-	//	err = multierror.Append(err, modelerr)
-	//}
-
-	model, modelerr = ml.LoadRWKV(modelFile, modelFile+tokenizerSuffix, threads)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	model, modelerr = ml.LoadBERT(modelFile)
-	if modelerr == nil {
-		updateModels(model)
-		return model, nil
-	} else {
-		err = multierror.Append(err, modelerr)
-	}
-
-	return nil, fmt.Errorf("could not load model - all backends returned error: %s", err.Error())
 }