Merge pull request #686 from jmorganca/mxyng/starcoder

decode starcoder

llm/falcon.go

@@ -1,7 +1,5 @@
 package llm
 
-const ModelFamilyFalcon = "falcon"
-
 const (
 	falconModelType7B   = 32
 	falconModelType40B  = 60
@@ -17,6 +15,6 @@ func falconModelType(numLayer uint32) string {
 	case 80:
 		return "180B"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }

llm/ggml.go

@@ -69,7 +69,7 @@ func fileType(fileType uint32) string {
 	case fileTypeQ6_K:
 		return "Q6_K"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }
 

llm/gguf.go

@@ -109,9 +109,13 @@ func (llm *ggufModel) ModelType() string {
 		if blocks, ok := llm.kv["falcon.block_count"].(uint32); ok {
 			return falconModelType(blocks)
 		}
+	case "starcoder":
+		if blocks, ok := llm.kv["starcoder.block_count"].(uint32); ok {
+			return starCoderModelType(blocks)
+		}
 	}
 
-	return "Unknown"
+	return "unknown"
 }
 
 func (llm *ggufModel) FileType() string {
@@ -120,7 +124,7 @@ func (llm *ggufModel) FileType() string {
 		return fileType(t)
 	}
 
-	return "Unknown"
+	return "unknown"
 }
 
 func (llm *ggufModel) Decode(r io.Reader) error {

llm/llama.go

@@ -143,7 +143,7 @@ func llamaModelType(numLayer uint32) string {
 	case 80:
 		return "65B"
 	default:
-		return "Unknown"
+		return "unknown"
 	}
 }
 

llm/starcoder.go

@@ -0,0 +1,23 @@
+package llm
+
+const (
+	starCoderModelType1B  = 24
+	starCoderModelType3B  = 36
+	starCoderModelType7B  = 42
+	starCoderModelType15B = 40
+)
+
+func starCoderModelType(numLayer uint32) string {
+	switch numLayer {
+	case 24:
+		return "1B"
+	case 36:
+		return "3B"
+	case 42:
+		return "7B"
+	case 40:
+		return "15B"
+	default:
+		return "unknown"
+	}
+}