feat: add rwkv support (#158)

Signed-off-by: mudler <mudler@mocaccino.org>
2024-12-24 06:46:39 +00:00 · 2023-05-03 11:45:22 +02:00 · 2023-05-03 11:45:22 +02:00 · 751b7eca62
commit 751b7eca62
parent 1ae7150810
7 changed files with 113 additions and 28 deletions
--- a/55
+++ b/55
@ -9,14 +9,18 @@ GOGPT4ALLJ_VERSION?=1f7bff57f66cb7062e40d0ac3abd2217815e5109
 # renovate: datasource=git-refs packageNameTemplate=https://github.com/go-skynet/go-gpt2.cpp currentValueTemplate=master depNameTemplate=go-gpt2.cpp
 GOGPT2_VERSION?=245a5bfe6708ab80dc5c733dcdbfbe3cfd2acdaa

+# here until https://github.com/donomii/go-rwkv.cpp/pull/1 is merged
+RWKV_REPO?=https://github.com/mudler/go-rwkv.cpp
+RWKV_VERSION?=6ba15255b03016b5ecce36529b500d21815399a7
+
 GREEN  := $(shell tput -Txterm setaf 2)
 YELLOW := $(shell tput -Txterm setaf 3)
 WHITE  := $(shell tput -Txterm setaf 7)
 CYAN   := $(shell tput -Txterm setaf 6)
 RESET  := $(shell tput -Txterm sgr0)

-C_INCLUDE_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2
-LIBRARY_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2
+C_INCLUDE_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2:$(shell pwd)/go-rwkv
+LIBRARY_PATH=$(shell pwd)/go-llama:$(shell pwd)/go-gpt4all-j:$(shell pwd)/go-gpt2:$(shell pwd)/go-rwkv

 # Use this if you want to set the default behavior
 ifndef BUILD_TYPE
@ -33,16 +37,6 @@ endif

 all: help

-## Build:
-
-build: prepare ## Build the project
-	$(info ${GREEN}I local-ai build info:${RESET})
-	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
-	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) build -o $(BINARY_NAME) ./
-
-generic-build: ## Build the project using generic
-	BUILD_TYPE="generic" $(MAKE) build
-
 ## GPT4ALL-J
 go-gpt4all-j:
 	git clone --recurse-submodules https://github.com/go-skynet/go-gpt4all-j.cpp go-gpt4all-j
@ -57,11 +51,19 @@ go-gpt4all-j:
 	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/void replace/void json_gptj_replace/g' {} +
 	@find ./go-gpt4all-j -type f -name "*.cpp" -exec sed -i'' -e 's/::replace/::json_gptj_replace/g' {} +

+## RWKV
+go-rwkv:
+	git clone --recurse-submodules $(RWKV_REPO) go-rwkv
+	cd go-rwkv && git checkout -b build $(RWKV_VERSION) && git submodule update --init --recursive --depth 1
+
+go-rwkv/librwkv.a: go-rwkv
+	cd go-rwkv && cd rwkv.cpp &&	cmake . -DRWKV_BUILD_SHARED_LIBRARY=OFF &&	cmake --build . && 	cp librwkv.a .. && cp ggml/src/libggml.a ..
+
 go-gpt4all-j/libgptj.a: go-gpt4all-j
 	$(MAKE) -C go-gpt4all-j $(GENERIC_PREFIX)libgptj.a

-# CEREBRAS GPT
-go-gpt2:
+## CEREBRAS GPT
+go-gpt2: 
 	git clone --recurse-submodules https://github.com/go-skynet/go-gpt2.cpp go-gpt2
 	cd go-gpt2 && git checkout -b build $(GOGPT2_VERSION) && git submodule update --init --recursive --depth 1
 	# This is hackish, but needed as both go-llama and go-gpt4allj have their own version of ggml..
@ -74,7 +76,6 @@ go-gpt2:

 go-gpt2/libgpt2.a: go-gpt2
 	$(MAKE) -C go-gpt2 $(GENERIC_PREFIX)libgpt2.a
-	

 go-llama:
 	git clone -b $(GOLLAMA_VERSION) --recurse-submodules https://github.com/go-skynet/go-llama.cpp go-llama
@ -86,26 +87,40 @@ replace:
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt4all-j.cpp=$(shell pwd)/go-gpt4all-j
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-gpt2.cpp=$(shell pwd)/go-gpt2
+	$(GOCMD) mod edit -replace github.com/donomii/go-rwkv.cpp=$(shell pwd)/go-rwkv

-prepare-sources: go-llama go-gpt2 go-gpt4all-j
+prepare-sources: go-llama go-gpt2 go-gpt4all-j go-rwkv
 	$(GOCMD) mod download

-rebuild:
+## GENERIC
+rebuild: ## Rebuilds the project
 	$(MAKE) -C go-llama clean
 	$(MAKE) -C go-gpt4all-j clean
 	$(MAKE) -C go-gpt2 clean
+	$(MAKE) -C go-rwkv clean
 	$(MAKE) build

-prepare: prepare-sources go-llama/libbinding.a go-gpt4all-j/libgptj.a go-gpt2/libgpt2.a replace
+prepare: prepare-sources go-llama/libbinding.a go-gpt4all-j/libgptj.a go-gpt2/libgpt2.a go-rwkv/librwkv.a replace ## Prepares for building

 clean: ## Remove build related file
 	rm -fr ./go-llama
 	rm -rf ./go-gpt4all-j
 	rm -rf ./go-gpt2
+	rm -rf ./go-rwkv
 	rm -rf $(BINARY_NAME)

-## Run:
-run: prepare
+## Build:
+
+build: prepare ## Build the project
+	$(info ${GREEN}I local-ai build info:${RESET})
+	$(info ${GREEN}I BUILD_TYPE: ${YELLOW}$(BUILD_TYPE)${RESET})
+	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) build -o $(BINARY_NAME) ./
+
+generic-build: ## Build the project using generic
+	BUILD_TYPE="generic" $(MAKE) build
+
+## Run
+run: prepare ## run local-ai
 	C_INCLUDE_PATH=${C_INCLUDE_PATH} LIBRARY_PATH=${LIBRARY_PATH} $(GOCMD) run ./main.go

 test-models/testmodel:
--- a/README.md
+++ b/README.md
@ -15,7 +15,7 @@
 - Supports multiple-models
 - Once loaded the first time, it keep models loaded in memory for faster inference
 - Support for prompt templates
- Doesn't shell-out, but uses C bindings for a faster inference and better performance. Uses [go-llama.cpp](https://github.com/go-skynet/go-llama.cpp) and [go-gpt4all-j.cpp](https://github.com/go-skynet/go-gpt4all-j.cpp).
+- Doesn't shell-out, but uses C bindings for a faster inference and better performance. 

 LocalAI is a community-driven project, focused on making the AI accessible to anyone. Any contribution, feedback and PR is welcome! It was initially created by [mudler](https://github.com/mudler/) at the [SpectroCloud OSS Office](https://github.com/spectrocloud).

@ -39,6 +39,7 @@ Tested with:
 - [GPT4ALL-J](https://gpt4all.io/models/ggml-gpt4all-j.bin)
 - Koala
 - [cerebras-GPT with ggml](https://huggingface.co/lxe/Cerebras-GPT-2.7B-Alpaca-SP-ggml)
+- [RWKV](https://github.com/BlinkDL/RWKV-LM) with [rwkv.cpp](https://github.com/saharNooby/rwkv.cpp)

 It should also be compatible with StableLM and GPTNeoX ggml models (untested)

@ -506,6 +507,13 @@ LocalAI is a community-driven project. It was initially created by [mudler](http

 MIT

+## Golang bindings used
+
+- [go-skynet/go-llama.cpp](https://github.com/go-skynet/go-llama.cpp)
+- [go-skynet/go-gpt4all-j.cpp](https://github.com/go-skynet/go-gpt4all-j.cpp)
+- [go-skynet/go-gpt2.cpp](https://github.com/go-skynet/go-gpt2.cpp)
+- [donomii/go-rwkv.cpp](https://github.com/donomii/go-rwkv.cpp)
+
 ## Acknowledgements

 - [llama.cpp](https://github.com/ggerganov/llama.cpp)
--- a/api/api_test.go
+++ b/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: 4 errors occurred:"))
+			Expect(err.Error()).To(ContainSubstring("error, status code: 500, message: could not load model - all backends returned error: 5 errors occurred:"))
 		})

 	})
--- a/api/prediction.go
+++ b/api/prediction.go
@ -6,6 +6,7 @@ import (
 	"strings"
 	"sync"

+	"github.com/donomii/go-rwkv.cpp"
 	model "github.com/go-skynet/LocalAI/pkg/model"
 	gpt2 "github.com/go-skynet/go-gpt2.cpp"
 	gptj "github.com/go-skynet/go-gpt4all-j.cpp"
@ -13,6 +14,8 @@ import (
 	"github.com/hashicorp/go-multierror"
 )

+const tokenizerSuffix = ".tokenizer.json"
+
 // mutex still needed, see: https://github.com/ggerganov/llama.cpp/discussions/784
 var mutexMap sync.Mutex
 var mutexes map[string]*sync.Mutex = make(map[string]*sync.Mutex)
@ -20,7 +23,7 @@ var mutexes map[string]*sync.Mutex = make(map[string]*sync.Mutex)
 var loadedModels map[string]interface{} = map[string]interface{}{}
 var muModels sync.Mutex

-func backendLoader(backendString string, loader *model.ModelLoader, modelFile string, llamaOpts []llama.ModelOption) (model interface{}, err error) {
+func backendLoader(backendString string, loader *model.ModelLoader, modelFile string, llamaOpts []llama.ModelOption, threads uint32) (model interface{}, err error) {
 	switch strings.ToLower(backendString) {
 	case "llama":
 		return loader.LoadLLaMAModel(modelFile, llamaOpts...)
@ -30,12 +33,14 @@ func backendLoader(backendString string, loader *model.ModelLoader, modelFile st
 		return loader.LoadGPT2Model(modelFile)
 	case "gptj":
 		return loader.LoadGPTJModel(modelFile)
+	case "rwkv":
+		return loader.LoadRWKV(modelFile, modelFile+tokenizerSuffix, threads)
 	default:
 		return nil, fmt.Errorf("backend unsupported: %s", backendString)
 	}
 }

-func greedyLoader(loader *model.ModelLoader, modelFile string, llamaOpts []llama.ModelOption) (model interface{}, err error) {
+func greedyLoader(loader *model.ModelLoader, modelFile string, llamaOpts []llama.ModelOption, threads uint32) (model interface{}, err error) {
 	updateModels := func(model interface{}) {
 		muModels.Lock()
 		defer muModels.Unlock()
@ -82,6 +87,14 @@ func greedyLoader(loader *model.ModelLoader, modelFile string, llamaOpts []llama
 		err = multierror.Append(err, modelerr)
 	}

+	model, modelerr = loader.LoadRWKV(modelFile, modelFile+tokenizerSuffix, threads)
+	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())
 }

@ -101,9 +114,9 @@ func ModelInference(s string, loader *model.ModelLoader, c Config, tokenCallback
 	var inferenceModel interface{}
 	var err error
 	if c.Backend == "" {
-		inferenceModel, err = greedyLoader(loader, modelFile, llamaOpts)
+		inferenceModel, err = greedyLoader(loader, modelFile, llamaOpts, uint32(c.Threads))
 	} else {
-		inferenceModel, err = backendLoader(c.Backend, loader, modelFile, llamaOpts)
+		inferenceModel, err = backendLoader(c.Backend, loader, modelFile, llamaOpts, uint32(c.Threads))
 	}
 	if err != nil {
 		return nil, err
@ -112,6 +125,20 @@ func ModelInference(s string, loader *model.ModelLoader, c Config, tokenCallback
 	var fn func() (string, error)

 	switch model := inferenceModel.(type) {
+	case *rwkv.RwkvState:
+		supportStreams = true
+
+		fn = func() (string, error) {
+			//model.ProcessInput("You are a chatbot that is very good at chatting.  blah blah blah")
+			stopWord := "\n"
+			if len(c.StopWords) > 0 {
+				stopWord = c.StopWords[0]
+			}
+
+			response := model.GenerateResponse(c.Maxtokens, stopWord, float32(c.Temperature), float32(c.TopP), tokenCallback)
+
+			return response, nil
+		}
 	case *gpt2.StableLM:
 		fn = func() (string, error) {
 			// Generate the prediction using the language model
--- a/go.mod
+++ b/go.mod
@ -23,6 +23,7 @@ require (
 	github.com/StackExchange/wmi v1.2.1 // indirect
 	github.com/andybalholm/brotli v1.0.5 // indirect
 	github.com/cpuguy83/go-md2man/v2 v2.0.2 // indirect
+	github.com/donomii/go-rwkv.cpp v0.0.0-20230502223004-0a3db3d72e7d // indirect
 	github.com/ghodss/yaml v1.0.0 // indirect
 	github.com/go-logr/logr v1.2.3 // indirect
 	github.com/go-ole/go-ole v1.2.6 // indirect
--- a/go.sum
+++ b/go.sum
@ -12,6 +12,8 @@ github.com/creack/pty v1.1.9/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ3
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/donomii/go-rwkv.cpp v0.0.0-20230502223004-0a3db3d72e7d h1:lSHwlYf1H4WAWYgf7rjEVTGen1qmigUq2Egpu8mnQiY=
+github.com/donomii/go-rwkv.cpp v0.0.0-20230502223004-0a3db3d72e7d/go.mod h1:H6QBF7/Tz6DAEBDXQged4H1BvsmqY/K5FG9wQRGa01g=
 github.com/ghodss/yaml v1.0.0 h1:wQHKEahhL6wmXdzwWG11gIVCkOv05bNOh+Rxn0yngAk=
 github.com/ghodss/yaml v1.0.0/go.mod h1:4dBDuWmgqj2HViK6kFavaiC9ZROes6MMH2rRYeMEF04=
 github.com/go-logr/logr v1.2.3 h1:2DntVwHkVopvECVRSlL5PSo9eG+cAkDCuckLubN+rq0=
--- a/pkg/model/loader.go
+++ b/pkg/model/loader.go
@ -12,6 +12,7 @@ import (

 	"github.com/rs/zerolog/log"

+	rwkv "github.com/donomii/go-rwkv.cpp"
 	gpt2 "github.com/go-skynet/go-gpt2.cpp"
 	gptj "github.com/go-skynet/go-gpt4all-j.cpp"
 	llama "github.com/go-skynet/go-llama.cpp"
@ -25,8 +26,8 @@ type ModelLoader struct {
 	gptmodels         map[string]*gptj.GPTJ
 	gpt2models        map[string]*gpt2.GPT2
 	gptstablelmmodels map[string]*gpt2.StableLM
-
-	promptsTemplates map[string]*template.Template
+	rwkv              map[string]*rwkv.RwkvState
+	promptsTemplates  map[string]*template.Template
 }

 func NewModelLoader(modelPath string) *ModelLoader {
@ -36,6 +37,7 @@ func NewModelLoader(modelPath string) *ModelLoader {
 		gptmodels:         make(map[string]*gptj.GPTJ),
 		gptstablelmmodels: make(map[string]*gpt2.StableLM),
 		models:            make(map[string]*llama.LLama),
+		rwkv:              make(map[string]*rwkv.RwkvState),
 		promptsTemplates:  make(map[string]*template.Template),
 	}
 }
@ -218,6 +220,36 @@ func (ml *ModelLoader) LoadGPTJModel(modelName string) (*gptj.GPTJ, error) {
 	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()