feat(speculative-sampling): allow to specify a draft model in the model config (#1052)

**Description**

This PR fixes #1013.

It adds `draft_model` and `n_draft` to the model YAML config in order to
load models with speculative sampling. This should be compatible as well
with grammars.

example:

```yaml
backend: llama                                                                                                                                                                   
context_size: 1024                                                                                                                                                                        
name: my-model-name
parameters:
  model: foo-bar
n_draft: 16                                                                                                                                                                      
draft_model: model-name
```

---------

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

api/backend/options.go

@@ -42,6 +42,7 @@ func gRPCModelOpts(c config.Config) *pb.ModelOptions {
 		Seed:          int32(c.Seed),
 		NBatch:        int32(b),
 		NoMulMatQ:     c.NoMulMatQ,
+		DraftModel:    c.DraftModel,
 		AudioPath:     c.VallE.AudioPath,
 		LoraAdapter:   c.LoraAdapter,
 		LoraBase:      c.LoraBase,
@@ -79,6 +80,7 @@ func gRPCPredictOpts(c config.Config, modelPath string) *pb.PredictOptions {
 	return &pb.PredictOptions{
 		Temperature:         float32(c.Temperature),
 		TopP:                float32(c.TopP),
+		NDraft:              c.NDraft,
 		TopK:                int32(c.TopK),
 		Tokens:              int32(c.Maxtokens),
 		Threads:             int32(c.Threads),

api/config/config.go

@@ -101,6 +101,8 @@ type LLMConfig struct {
 	LoraAdapter     string   `yaml:"lora_adapter"`
 	LoraBase        string   `yaml:"lora_base"`
 	NoMulMatQ       bool     `yaml:"no_mulmatq"`
+	DraftModel      string   `yaml:"draft_model"`
+	NDraft          int32    `yaml:"n_draft"`
 }
 
 type AutoGPTQ struct {