Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

add verbose mode to the show command (#9640)

Patrick Devine4bed739259d8
diff --git a/api/types.go b/api/types.go
index fef836bd6..a38b335b7 100644
--- a/api/types.go
+++ b/api/types.go
@@ -349,6 +349,7 @@ type ShowResponse struct {
Messages []Message `json:"messages,omitempty"`
ModelInfo map[string]any `json:"model_info,omitempty"`
ProjectorInfo map[string]any `json:"projector_info,omitempty"`
+ Tensors []Tensor `json:"tensors,omitempty"`
ModifiedAt time.Time `json:"modified_at,omitempty"`
}
@@ -467,6 +468,13 @@ type ModelDetails struct {
QuantizationLevel string `json:"quantization_level"`
}
+// Tensor describes the metadata for a given tensor.
+type Tensor struct {
+ Name string `json:"name"`
+ Type string `json:"type"`
+ Shape []uint64 `json:"shape"`
+}
+
func (m *Metrics) Summary() {
if m.TotalDuration > 0 {
fmt.Fprintf(os.Stderr, "total duration: %v\n", m.TotalDuration)
diff --git a/cmd/cmd.go b/cmd/cmd.go
index c22a08f43..710f49a72 100644
--- a/cmd/cmd.go
+++ b/cmd/cmd.go
@@ -18,6 +18,7 @@ import (
"os/signal"
"path/filepath"
"runtime"
+ "sort"
"strconv"
"strings"
"sync/atomic"
@@ -568,8 +569,9 @@ func ShowHandler(cmd *cobra.Command, args []string) error {
parameters, errParams := cmd.Flags().GetBool("parameters")
system, errSystem := cmd.Flags().GetBool("system")
template, errTemplate := cmd.Flags().GetBool("template")
+ verbose, errVerbose := cmd.Flags().GetBool("verbose")
- for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate} {
+ for _, boolErr := range []error{errLicense, errModelfile, errParams, errSystem, errTemplate, errVerbose} {
if boolErr != nil {
return errors.New("error retrieving flags")
}
@@ -607,7 +609,7 @@ func ShowHandler(cmd *cobra.Command, args []string) error {
return errors.New("only one of '--license', '--modelfile', '--parameters', '--system', or '--template' can be specified")
}
- req := api.ShowRequest{Name: args[0]}
+ req := api.ShowRequest{Name: args[0], Verbose: verbose}
resp, err := client.Show(cmd.Context(), &req)
if err != nil {
return err
@@ -630,10 +632,10 @@ func ShowHandler(cmd *cobra.Command, args []string) error {
return nil
}
- return showInfo(resp, os.Stdout)
+ return showInfo(resp, verbose, os.Stdout)
}
-func showInfo(resp *api.ShowResponse, w io.Writer) error {
+func showInfo(resp *api.ShowResponse, verbose bool, w io.Writer) error {
tableRender := func(header string, rows func() [][]string) {
fmt.Fprintln(w, " ", header)
table := tablewriter.NewWriter(w)
@@ -690,6 +692,45 @@ func showInfo(resp *api.ShowResponse, w io.Writer) error {
})
}
+ if resp.ModelInfo != nil && verbose {
+ tableRender("Metadata", func() (rows [][]string) {
+ keys := make([]string, 0, len(resp.ModelInfo))
+ for k := range resp.ModelInfo {
+ keys = append(keys, k)
+ }
+ sort.Strings(keys)
+
+ for _, k := range keys {
+ var v string
+ switch vData := resp.ModelInfo[k].(type) {
+ case string:
+ v = vData
+ case float64:
+ v = fmt.Sprintf("%g", vData)
+ case []any:
+ n := 3
+ if len(vData) < n {
+ n = len(vData)
+ }
+ v = fmt.Sprintf("%v", vData[:n])
+ default:
+ v = fmt.Sprintf("%T", vData)
+ }
+ rows = append(rows, []string{"", k, v})
+ }
+ return
+ })
+ }
+
+ if len(resp.Tensors) > 0 && verbose {
+ tableRender("Tensors", func() (rows [][]string) {
+ for _, t := range resp.Tensors {
+ rows = append(rows, []string{"", t.Name, t.Type, fmt.Sprint(t.Shape)})
+ }
+ return
+ })
+ }
+
head := func(s string, n int) (rows [][]string) {
scanner := bufio.NewScanner(strings.NewReader(s))
for scanner.Scan() && (len(rows) < n || n < 0) {
@@ -1196,6 +1237,7 @@ func NewCLI() *cobra.Command {
showCmd.Flags().Bool("parameters", false, "Show parameters of a model")
showCmd.Flags().Bool("template", false, "Show template of a model")
showCmd.Flags().Bool("system", false, "Show system message of a model")
+ showCmd.Flags().BoolP("verbose", "v", false, "Show detailed model information")
runCmd := &cobra.Command{
Use: "run MODEL [PROMPT]",
diff --git a/cmd/cmd_test.go b/cmd/cmd_test.go
index e70ffbeab..f21a8f50b 100644
--- a/cmd/cmd_test.go
+++ b/cmd/cmd_test.go
@@ -27,7 +27,7 @@ func TestShowInfo(t *testing.T) {
ParameterSize: "7B",
QuantizationLevel: "FP16",
},
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
@@ -57,7 +57,7 @@ func TestShowInfo(t *testing.T) {
ParameterSize: "7B",
QuantizationLevel: "FP16",
},
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
@@ -68,6 +68,56 @@ func TestShowInfo(t *testing.T) {
embedding length 0
quantization FP16
+`
+ if diff := cmp.Diff(expect, b.String()); diff != "" {
+ t.Errorf("unexpected output (-want +got):\n%s", diff)
+ }
+ })
+
+ t.Run("verbose model", func(t *testing.T) {
+ var b bytes.Buffer
+ if err := showInfo(&api.ShowResponse{
+ Details: api.ModelDetails{
+ Family: "test",
+ ParameterSize: "8B",
+ QuantizationLevel: "FP16",
+ },
+ Parameters: `
+ stop up`,
+ ModelInfo: map[string]any{
+ "general.architecture": "test",
+ "general.parameter_count": float64(8_000_000_000),
+ "test.context_length": float64(1000),
+ "test.embedding_length": float64(11434),
+ },
+ Tensors: []api.Tensor{
+ {Name: "blk.0.attn_k.weight", Type: "BF16", Shape: []uint64{42, 3117}},
+ {Name: "blk.0.attn_q.weight", Type: "FP16", Shape: []uint64{3117, 42}},
+ },
+ }, true, &b); err != nil {
+ t.Fatal(err)
+ }
+
+ expect := ` Model
+ architecture test
+ parameters 8B
+ context length 1000
+ embedding length 11434
+ quantization FP16
+
+ Parameters
+ stop up
+
+ Metadata
+ general.architecture test
+ general.parameter_count 8e+09
+ test.context_length 1000
+ test.embedding_length 11434
+
+ Tensors
+ blk.0.attn_k.weight BF16 [42 3117]
+ blk.0.attn_q.weight FP16 [3117 42]
+
`
if diff := cmp.Diff(expect, b.String()); diff != "" {
t.Errorf("unexpected output (-want +got):\n%s", diff)
@@ -89,7 +139,7 @@ func TestShowInfo(t *testing.T) {
stop you
stop up
temperature 99`,
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
@@ -126,7 +176,7 @@ func TestShowInfo(t *testing.T) {
"clip.vision.embedding_length": float64(0),
"clip.vision.projection_dim": float64(0),
},
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
@@ -159,7 +209,7 @@ func TestShowInfo(t *testing.T) {
Ahoy, matey!
Weigh anchor!
`,
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
@@ -188,7 +238,7 @@ Weigh anchor!
QuantizationLevel: "FP16",
},
License: license,
- }, &b); err != nil {
+ }, false, &b); err != nil {
t.Fatal(err)
}
diff --git a/cmd/interactive.go b/cmd/interactive.go
index 7c11ab83e..f3489b652 100644
--- a/cmd/interactive.go
+++ b/cmd/interactive.go
@@ -347,7 +347,7 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
switch args[1] {
case "info":
- _ = showInfo(resp, os.Stderr)
+ _ = showInfo(resp, false, os.Stderr)
case "license":
if resp.License == "" {
fmt.Println("No license was specified for this model.")
diff --git a/fs/ggml/ggml.go b/fs/ggml/ggml.go
index d32296d9c..230182720 100644
--- a/fs/ggml/ggml.go
+++ b/fs/ggml/ggml.go
@@ -327,6 +327,10 @@ func (t Tensor) Size() uint64 {
return t.parameters() * t.typeSize() / t.blockSize()
}
+func (t Tensor) Type() string {
+ return fileType(t.Kind).String()
+}
+
type container interface {
Name() string
Decode(io.ReadSeeker) (model, error)
diff --git a/server/routes.go b/server/routes.go
index bc3fe3fb5..059936249 100644
--- a/server/routes.go
+++ b/server/routes.go
@@ -435,7 +435,7 @@ func (s *Server) EmbedHandler(c *gin.Context) {
return
}
- kvData, err := getKVData(m.ModelPath, false)
+ kvData, _, err := getModelData(m.ModelPath, false)
if err != nil {
c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
return
@@ -848,16 +848,23 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
fmt.Fprint(&sb, m.String())
resp.Modelfile = sb.String()
- kvData, err := getKVData(m.ModelPath, req.Verbose)
+ kvData, tensors, err := getModelData(m.ModelPath, req.Verbose)
if err != nil {
return nil, err
}
+
delete(kvData, "general.name")
delete(kvData, "tokenizer.chat_template")
resp.ModelInfo = kvData
+ tensorData := make([]api.Tensor, len(tensors.Items()))
+ for cnt, t := range tensors.Items() {
+ tensorData[cnt] = api.Tensor{Name: t.Name, Type: t.Type(), Shape: t.Shape}
+ }
+ resp.Tensors = tensorData
+
if len(m.ProjectorPaths) > 0 {
- projectorData, err := getKVData(m.ProjectorPaths[0], req.Verbose)
+ projectorData, _, err := getModelData(m.ProjectorPaths[0], req.Verbose)
if err != nil {
return nil, err
}
@@ -867,17 +874,17 @@ func GetModelInfo(req api.ShowRequest) (*api.ShowResponse, error) {
return resp, nil
}
-func getKVData(digest string, verbose bool) (ggml.KV, error) {
+func getModelData(digest string, verbose bool) (ggml.KV, ggml.Tensors, error) {
maxArraySize := 0
if verbose {
maxArraySize = -1
}
- kvData, err := llm.LoadModel(digest, maxArraySize)
+ data, err := llm.LoadModel(digest, maxArraySize)
if err != nil {
- return nil, err
+ return nil, ggml.Tensors{}, err
}
- kv := kvData.KV()
+ kv := data.KV()
if !verbose {
for k := range kv {
@@ -887,7 +894,7 @@ func getKVData(digest string, verbose bool) (ggml.KV, error) {
}
}
- return kv, nil
+ return kv, data.Tensors(), nil
}
func (s *Server) ListHandler(c *gin.Context) {