Fix inconsistent unit formatting in docker model ls output

Remove spaces between numbers and units in the PARAMETERS and SIZE columns to enable reliable column-based parsing with tools like awk.

Changes:

- Safetensors: Remove spaces from parameter units (K, M, B, T)

- Safetensors: Use decimal units (MB, GB) matching 'docker images' style

- GGUF: Add regex normalization to remove spaces from parser output

- Update all tests to reflect new no-space formatting

The formatting now produces consistent output like "361.82M" and "256MB" instead of "361.82 M" and "256.35 MiB", making the output parseable by standard column-based tools.

Fixes #349

Author: hwclass

pkg/distribution/internal/gguf/create.go

@@ -2,6 +2,7 @@ package gguf
 
 import (
 	"fmt"
+	"regexp"
 	"strings"
 	"time"
 
@@ -55,10 +56,29 @@ func configFromFile(path string) types.Config {
 	}
 	return types.Config{
 		Format:       types.FormatGGUF,
-		Parameters:   strings.TrimSpace(gguf.Metadata().Parameters.String()),
+		Parameters:   normalizeUnitString(gguf.Metadata().Parameters.String()),
 		Architecture: strings.TrimSpace(gguf.Metadata().Architecture),
 		Quantization: strings.TrimSpace(gguf.Metadata().FileType.String()),
-		Size:         strings.TrimSpace(gguf.Metadata().Size.String()),
+		Size:         normalizeUnitString(gguf.Metadata().Size.String()),
 		GGUF:         extractGGUFMetadata(&gguf.Header),
 	}
 }
+
+var (
+	// spaceBeforeUnitRegex matches one or more spaces between a number/decimal and a letter (unit)
+	// Used to remove spaces between numbers and units (e.g., "16.78 M" -> "16.78M")
+	// Pattern: digits/decimals, then whitespace, then letters (unit)
+	spaceBeforeUnitRegex = regexp.MustCompile(`([0-9.]+)\s+([A-Za-z]+)`)
+)
+
+// normalizeUnitString removes spaces between numbers and units for consistent formatting
+// Examples: "16.78 M" -> "16.78M", "256.35 MiB" -> "256.35MiB", "409M" -> "409M"
+func normalizeUnitString(s string) string {
+	s = strings.TrimSpace(s)
+	if len(s) == 0 {
+		return s
+	}
+	// Remove space(s) between numbers/decimals and unit letters using regex
+	// Pattern matches: number(s) or decimal, then whitespace, then letters (unit)
+	return spaceBeforeUnitRegex.ReplaceAllString(s, "$1$2")
+}

pkg/distribution/internal/gguf/model_test.go

@@ -32,8 +32,8 @@ func TestGGUF(t *testing.T) {
 			if cfg.Quantization != "Unknown" { // todo: testdata with a real value
 				t.Fatalf("Unexpected quantization: got %s expected %s", cfg.Quantization, "Unknown")
 			}
-			if cfg.Size != "864 B" {
-				t.Fatalf("Unexpected quantization: got %s expected %s", cfg.Quantization, "Unknown")
+			if cfg.Size != "864B" {
+				t.Fatalf("Unexpected size: got %s expected %s", cfg.Size, "864B")
 			}
 
 			// Test GGUF metadata
@@ -118,8 +118,8 @@ func TestGGUFShards(t *testing.T) {
 			if cfg.Quantization != "Unknown" { // todo: testdata with a real value
 				t.Fatalf("Unexpected quantization: got %s expected %s", cfg.Quantization, "Unknown")
 			}
-			if cfg.Size != "864 B" {
-				t.Fatalf("Unexpected quantization: got %s expected %s", cfg.Quantization, "Unknown")
+			if cfg.Size != "864B" {
+				t.Fatalf("Unexpected size: got %s expected %s", cfg.Size, "864B")
 			}
 
 			// Test GGUF metadata

pkg/distribution/internal/safetensors/metadata.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"strings"
 
 	"github.com/docker/go-units"
 )
@@ -183,13 +184,16 @@ func (h *Header) ExtractMetadata() map[string]string {
 	return metadata
 }
 
-// formatParameters converts parameter count to human-readable format matching GGUF style
-// Returns format like "361.82 M" or "1.5 B" (space before unit, base 1000, where B = Billion)
+// formatParameters converts parameter count to human-readable format
+// Returns format like "361.82M" or "1.5B" (no space before unit, base 1000, where B = Billion)
 func formatParameters(params int64) string {
-	return units.CustomSize("%.2f%s", float64(params), 1000.0, []string{"", " K", " M", " B", " T"})
+	return units.CustomSize("%.2f%s", float64(params), 1000.0, []string{"", "K", "M", "B", "T"})
 }
 
-// formatSize converts bytes to human-readable format
+// formatSize converts bytes to human-readable format matching Docker's style
+// Returns format like "256MB" (decimal units, no space, matching `docker images`)
 func formatSize(bytes int64) string {
-	return units.HumanSizeWithPrecision(float64(bytes), 2)
+	formatted := units.HumanSize(float64(bytes))
+	// Remove space between number and unit to match Docker format (e.g., "256 MB" -> "256MB")
+	return strings.ReplaceAll(formatted, " ", "")
 }

pkg/distribution/internal/safetensors/model_test.go

@@ -92,7 +92,7 @@ func TestNewModel_WithMetadata(t *testing.T) {
 	}
 
 	// Verify parameters (4096*4096 + 4096 = 16781312)
-	expectedParams := "16.78 M"
+	expectedParams := "16.78M"
 	if config.Parameters != expectedParams {
 		t.Errorf("Config.Parameters = %v, want %v", config.Parameters, expectedParams)
 	}
@@ -220,7 +220,7 @@ func TestNewModel_NoMetadata(t *testing.T) {
 	}
 
 	// Verify parameters (100*200 = 20000)
-	expectedParams := "20.00 K"
+	expectedParams := "20.00K"
 	if config.Parameters != expectedParams {
 		t.Errorf("Config.Parameters = %v, want %v", config.Parameters, expectedParams)
 	}