add format bytes

api/client.go

@@ -127,7 +127,7 @@ func (c *Client) do(ctx context.Context, method, path string, reqData, respData
 	return nil
 }
 
-const maxBufferSize = 512 * 1024 // 512KB
+const maxBufferSize = 512 * 1000 // 512KB
 
 func (c *Client) stream(ctx context.Context, method, path string, data any, fn func([]byte) error) error {
 	var buf *bytes.Buffer

format/bytes.go

+package format
+
+import "fmt"
+
+func HumanBytes(b int64) string {
+	switch {
+	case b > 1000*1000*1000:
+		return fmt.Sprintf("%d GB", b/1000/1000/1000)
+	case b > 1000*1000:
+		return fmt.Sprintf("%d MB", b/1000/1000)
+	case b > 1000:
+		return fmt.Sprintf("%d KB", b/1000)
+	default:
+		return fmt.Sprintf("%d B", b)
+	}
+}

llm/llama.go

@@ -454,7 +454,7 @@ type PredictRequest struct {
 	Stop             []string `json:"stop,omitempty"`
 }
 
-const maxBufferSize = 512 * 1024 // 512KB
+const maxBufferSize = 512 * 1000 // 512KB
 
 func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string, fn func(api.GenerateResponse)) error {
 	prevConvo, err := llm.Decode(ctx, prevContext)

llm/llm.go

@@ -60,33 +60,33 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
 	totalResidentMemory := memory.TotalMemory()
 	switch ggml.ModelType() {
 	case "3B", "7B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 16*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 16GB of memory")
-		} else if totalResidentMemory < 8*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 8GB of memory")
+		if ggml.FileType() == "F16" && totalResidentMemory < 16*1000*1000 {
+			return nil, fmt.Errorf("F16 model requires at least 16 GB of memory")
+		} else if totalResidentMemory < 8*1000*1000 {
+			return nil, fmt.Errorf("model requires at least 8 GB of memory")
 		}
 	case "13B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 32*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 32GB of memory")
-		} else if totalResidentMemory < 16*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 16GB of memory")
+		if ggml.FileType() == "F16" && totalResidentMemory < 32*1000*1000 {
+			return nil, fmt.Errorf("F16 model requires at least 32 GB of memory")
+		} else if totalResidentMemory < 16*1000*1000 {
+			return nil, fmt.Errorf("model requires at least 16 GB of memory")
 		}
 	case "30B", "34B", "40B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 64*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 64GB of memory")
-		} else if totalResidentMemory < 32*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 32GB of memory")
+		if ggml.FileType() == "F16" && totalResidentMemory < 64*1000*1000 {
+			return nil, fmt.Errorf("F16 model requires at least 64 GB of memory")
+		} else if totalResidentMemory < 32*1000*1000 {
+			return nil, fmt.Errorf("model requires at least 32 GB of memory")
 		}
 	case "65B", "70B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 128*1024*1024 {
-			return nil, fmt.Errorf("F16 model requires at least 128GB of memory")
-		} else if totalResidentMemory < 64*1024*1024 {
-			return nil, fmt.Errorf("model requires at least 64GB of memory")
+		if ggml.FileType() == "F16" && totalResidentMemory < 128*1000*1000 {
+			return nil, fmt.Errorf("F16 model requires at least 128 GB of memory")
+		} else if totalResidentMemory < 64*1000*1000 {
+			return nil, fmt.Errorf("model requires at least 64 GB of memory")
 		}
 	case "180B":
-		if ggml.FileType() == "F16" && totalResidentMemory < 512*1024*1024 {
+		if ggml.FileType() == "F16" && totalResidentMemory < 512*1000*1000 {
 			return nil, fmt.Errorf("F16 model requires at least 512GB of memory")
-		} else if totalResidentMemory < 128*1024*1024 {
+		} else if totalResidentMemory < 128*1000*1000 {
 			return nil, fmt.Errorf("model requires at least 128GB of memory")
 		}
 	}