feat: add rope settings and negative prompt, drop grammar backend (#797)

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

api/api_test.go

@@ -297,7 +297,7 @@ var _ = Describe("API test", func() {
 				response := postModelApplyRequest("http://127.0.0.1:9090/models/apply", modelApplyRequest{
 					URL:       "github:go-skynet/model-gallery/openllama_3b.yaml",
 					Name:      "openllama_3b",
-					Overrides: map[string]string{"backend": "llama-grammar"},
+					Overrides: map[string]string{"backend": "llama"},
 				})
 
 				Expect(response["uuid"]).ToNot(BeEmpty(), fmt.Sprint(response))
@@ -355,7 +355,7 @@ var _ = Describe("API test", func() {
 				var res map[string]string
 				err = json.Unmarshal([]byte(resp2.Choices[0].Message.FunctionCall.Arguments), &res)
 				Expect(err).ToNot(HaveOccurred())
-				Expect(res["location"]).To(Equal("San Francisco"), fmt.Sprint(res))
+				Expect(res["location"]).To(Equal("San Francisco, California, United States"), fmt.Sprint(res))
 				Expect(res["unit"]).To(Equal("celcius"), fmt.Sprint(res))
 				Expect(string(resp2.Choices[0].FinishReason)).To(Equal("function_call"), fmt.Sprint(resp2.Choices[0].FinishReason))
 			})

api/backend/options.go

@@ -39,34 +39,37 @@ func gRPCPredictOpts(c config.Config, modelPath string) *pb.PredictOptions {
 		promptCachePath = p
 	}
 	return &pb.PredictOptions{
-		Temperature:     float32(c.Temperature),
-		TopP:            float32(c.TopP),
-		TopK:            int32(c.TopK),
-		Tokens:          int32(c.Maxtokens),
-		Threads:         int32(c.Threads),
-		PromptCacheAll:  c.PromptCacheAll,
-		PromptCacheRO:   c.PromptCacheRO,
-		PromptCachePath: promptCachePath,
-		F16KV:           c.F16,
-		DebugMode:       c.Debug,
-		Grammar:         c.Grammar,
-
-		Mirostat:          int32(c.Mirostat),
-		MirostatETA:       float32(c.MirostatETA),
-		MirostatTAU:       float32(c.MirostatTAU),
-		Debug:             c.Debug,
-		StopPrompts:       c.StopWords,
-		Repeat:            int32(c.RepeatPenalty),
-		NKeep:             int32(c.Keep),
-		Batch:             int32(c.Batch),
-		IgnoreEOS:         c.IgnoreEOS,
-		Seed:              int32(c.Seed),
-		FrequencyPenalty:  float32(c.FrequencyPenalty),
-		MLock:             c.MMlock,
-		MMap:              c.MMap,
-		MainGPU:           c.MainGPU,
-		TensorSplit:       c.TensorSplit,
-		TailFreeSamplingZ: float32(c.TFZ),
-		TypicalP:          float32(c.TypicalP),
+		Temperature:         float32(c.Temperature),
+		TopP:                float32(c.TopP),
+		TopK:                int32(c.TopK),
+		Tokens:              int32(c.Maxtokens),
+		Threads:             int32(c.Threads),
+		PromptCacheAll:      c.PromptCacheAll,
+		PromptCacheRO:       c.PromptCacheRO,
+		PromptCachePath:     promptCachePath,
+		F16KV:               c.F16,
+		DebugMode:           c.Debug,
+		Grammar:             c.Grammar,
+		NegativePromptScale: c.NegativePromptScale,
+		RopeFreqBase:        c.RopeFreqBase,
+		RopeFreqScale:       c.RopeFreqScale,
+		NegativePrompt:      c.NegativePrompt,
+		Mirostat:            int32(c.Mirostat),
+		MirostatETA:         float32(c.MirostatETA),
+		MirostatTAU:         float32(c.MirostatTAU),
+		Debug:               c.Debug,
+		StopPrompts:         c.StopWords,
+		Repeat:              int32(c.RepeatPenalty),
+		NKeep:               int32(c.Keep),
+		Batch:               int32(c.Batch),
+		IgnoreEOS:           c.IgnoreEOS,
+		Seed:                int32(c.Seed),
+		FrequencyPenalty:    float32(c.FrequencyPenalty),
+		MLock:               c.MMlock,
+		MMap:                c.MMap,
+		MainGPU:             c.MainGPU,
+		TensorSplit:         c.TensorSplit,
+		TailFreeSamplingZ:   float32(c.TFZ),
+		TypicalP:            float32(c.TypicalP),
 	}
 }

api/config/prediction.go

@@ -34,4 +34,9 @@ type PredictionOptions struct {
 
 	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"`
 }

api/openai/request.go

@@ -71,6 +71,22 @@ func updateConfig(config *config.Config, input *OpenAIRequest) {
 		config.TopP = input.TopP
 	}
 
+	if input.NegativePromptScale != 0 {
+		config.NegativePromptScale = input.NegativePromptScale
+	}
+
+	if input.NegativePrompt != "" {
+		config.NegativePrompt = input.NegativePrompt
+	}
+
+	if input.RopeFreqBase != 0 {
+		config.RopeFreqBase = input.RopeFreqBase
+	}
+
+	if input.RopeFreqScale != 0 {
+		config.RopeFreqScale = input.RopeFreqScale
+	}
+
 	if input.Grammar != "" {
 		config.Grammar = input.Grammar
 	}