feat: add starcoder (#236)

Makefile

@@ -6,7 +6,7 @@ BINARY_NAME=local-ai
 GOLLAMA_VERSION?=c03e8adbc45c866e0f6d876af1887d6b01d57eb4
 GPT4ALL_REPO?=https://github.com/go-skynet/gpt4all
 GPT4ALL_VERSION?=3657f9417e17edf378c27d0a9274a1bf41caa914
-GOGPT2_VERSION?=abf038a7d8efa4eefdc7c891f05ad33d4e59e49d
+GOGPT2_VERSION?=6a10572
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp
 RWKV_VERSION?=07166da10cb2a9e8854395a4f210464dcea76e47
 WHISPER_CPP_VERSION?=bf2449dfae35a46b2cd92ab22661ce81a48d4993

README.md

@@ -92,6 +92,28 @@ It should also be compatible with StableLM and GPTNeoX ggml models (untested).
 Depending on the model you are attempting to run might need more RAM or CPU resources. Check out also [here](https://github.com/ggerganov/llama.cpp#memorydisk-requirements) for `ggml` based backends. `rwkv` is less expensive on resources.
 
 
+### Feature support matrix
+
+<details>
+
+| Backend         | Compatible models     | Completion/Chat endpoint | Audio transcription | Embeddings support                | Token stream support | Github                                     | Bindings                                  |
+|-----------------|-----------------------|--------------------------|---------------------|-----------------------------------|----------------------|--------------------------------------------|-------------------------------------------|
+| llama           | Vicuna, Alpaca, LLaMa | yes                      | no                  | yes (doesn't seem to be accurate) | yes                  | https://github.com/ggerganov/llama.cpp     | https://github.com/go-skynet/go-llama.cpp |
+| gpt4all-llama   | Vicuna, Alpaca, LLaMa | yes                      | no                  | no                                | yes                  | https://github.com/nomic-ai/gpt4all        | https://github.com/go-skynet/gpt4all      |
+| gpt4all-mpt     | MPT                   | yes                      | no                  | no                                | yes                  | https://github.com/nomic-ai/gpt4all        | https://github.com/go-skynet/gpt4all      |
+| gpt4all-j       | GPT4ALL-J             | yes                      | no                  | no                                | yes                  | https://github.com/nomic-ai/gpt4all        | https://github.com/go-skynet/gpt4all      |
+| gpt2            | GPT/NeoX, Cerebras    | yes                      | no                  | no                                | no                   | https://github.com/ggerganov/ggml          | https://github.com/go-skynet/go-gpt2.cpp  |
+| dolly           | Dolly                 | yes                      | no                  | no                                | no                   | https://github.com/ggerganov/ggml          | https://github.com/go-skynet/go-gpt2.cpp  |
+| redpajama       | RedPajama             | yes                      | no                  | no                                | no                   | https://github.com/ggerganov/ggml          | https://github.com/go-skynet/go-gpt2.cpp  |
+| stableLM        | StableLM GPT/NeoX     | yes                      | no                  | no                                | no                   | https://github.com/ggerganov/ggml          | https://github.com/go-skynet/go-gpt2.cpp  |
+| starcoder       | Starcoder             | yes                      | no                  | no                                | no                   | https://github.com/ggerganov/ggml          | https://github.com/go-skynet/go-gpt2.cpp  |
+| bloomz          | Bloom                 | yes                      | no                  | no                                | no                   | https://github.com/NouamaneTazi/bloomz.cpp | https://github.com/go-skynet/bloomz.cpp   |
+| rwkv            | RWKV                  | yes                      | no                  | no                                | yes                  | https://github.com/saharNooby/rwkv.cpp     | https://github.com/donomii/go-rwkv.cpp    |
+| bert-embeddings | bert                  | no                       | no                  | yes                               | no                   | https://github.com/skeskinen/bert.cpp      | https://github.com/go-skynet/go-bert.cpp  |
+| whisper         | whisper               | no                       | yes                 | no                                | no                   | https://github.com/ggerganov/whisper.cpp   | https://github.com/ggerganov/whisper.cpp  |
+
+</details>
+
 ## Usage
 
 > `LocalAI` comes by default as a container image. You can check out all the available images with corresponding tags [here](https://quay.io/repository/go-skynet/local-ai?tab=tags&tag=latest).

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: 9 errors occurred:"))
+			Expect(err.Error()).To(ContainSubstring("error, status code: 500, message: could not load model - all backends returned error: 10 errors occurred:"))
 		})
 
 	})

api/prediction.go

@@ -199,6 +199,30 @@ func ModelInference(s string, loader *model.ModelLoader, c Config, tokenCallback
 
 			return response, nil
 		}
+	case *gpt2.Starcoder:
+		fn = func() (string, error) {
+			// Generate the prediction using the language model
+			predictOptions := []gpt2.PredictOption{
+				gpt2.SetTemperature(c.Temperature),
+				gpt2.SetTopP(c.TopP),
+				gpt2.SetTopK(c.TopK),
+				gpt2.SetTokens(c.Maxtokens),
+				gpt2.SetThreads(c.Threads),
+			}
+
+			if c.Batch != 0 {
+				predictOptions = append(predictOptions, gpt2.SetBatch(c.Batch))
+			}
+
+			if c.Seed != 0 {
+				predictOptions = append(predictOptions, gpt2.SetSeed(c.Seed))
+			}
+
+			return model.Predict(
+				s,
+				predictOptions...,
+			)
+		}
 	case *gpt2.RedPajama:
 		fn = func() (string, error) {
 			// Generate the prediction using the language model

pkg/model/initializers.go

@@ -20,6 +20,7 @@ const tokenizerSuffix = ".tokenizer.json"
 const (
 	LlamaBackend          = "llama"
 	BloomzBackend         = "bloomz"
+	StarcoderBackend      = "starcoder"
 	StableLMBackend       = "stablelm"
 	DollyBackend          = "dolly"
 	RedPajamaBackend      = "redpajama"
@@ -45,6 +46,11 @@ var backends []string = []string{
 	DollyBackend,
 	RedPajamaBackend,
 	BertEmbeddingsBackend,
+	StarcoderBackend,
+}
+
+var starCoder = func(modelFile string) (interface{}, error) {
+	return gpt2.NewStarcoder(modelFile)
 }
 
 var redPajama = func(modelFile string) (interface{}, error) {
@@ -110,6 +116,8 @@ func (ml *ModelLoader) BackendLoader(backendString string, modelFile string, lla
 		return ml.LoadModel(modelFile, redPajama)
 	case Gpt2Backend:
 		return ml.LoadModel(modelFile, gpt2LM)
+	case StarcoderBackend:
+		return ml.LoadModel(modelFile, starCoder)
 	case Gpt4AllLlamaBackend:
 		return ml.LoadModel(modelFile, gpt4allLM(gpt4all.SetThreads(int(threads)), gpt4all.SetModelType(gpt4all.LLaMAType)))
 	case Gpt4AllMptBackend: