janpfeifer · Sep 24, 2026 · Sep 23, 2026 · Sep 23, 2026 · Sep 24, 2026
diff --git a/cmd/hubinfo/main.go b/cmd/hubinfo/main.go
@@ -120,8 +120,8 @@ func main() {
 			}
 		}
 		if info.CardData != nil {
-			if info.CardData.License != "" {
-				metaTable.Row("Card Data: License", info.CardData.License)
+			if len(info.CardData.License) > 0 {
+				metaTable.Row("Card Data: License", strings.Join(info.CardData.License, ", "))
 			}
 			if info.CardData.LicenseLink != "" {
 				metaTable.Row("Card Data: License Link", info.CardData.LicenseLink)

diff --git a/hub/info.go b/hub/info.go
@@ -69,13 +69,72 @@ type RepoConfig struct {
 	TokenizerConfig map[string]any `json:"tokenizer_config"`
 }
 
-// CardData contains metadata about the model card.
+// CardData contains metadata about the model/dataset card.
 type CardData struct {
 	LibraryName string `json:"library_name"`
-	License     string `json:"license"`
-	LicenseLink string `json:"license_link"`
-	PipelineTag string `json:"pipeline_tag"`
-	BaseModel   any    `json:"base_model"` // Can be a string, or slice of strings, or nil.
+	// License holds one or more licenses from the Hub card. Hugging Face may send
+	// cardData.license as a string or a list of strings (common for datasets); a
+	// single string is stored as a one-element slice.
+	License     []string `json:"license"`
+	LicenseLink string   `json:"license_link"`
+	PipelineTag string   `json:"pipeline_tag"`
+	BaseModel   any      `json:"base_model"` // Can be a string, or slice of strings, or nil.
+}
+
+// UnmarshalJSON accepts cardData.license as either a string or an array of strings.
+// Other fields decode via a type alias so new CardData fields are not omitted.
+func (c *CardData) UnmarshalJSON(data []byte) error {
+	if string(data) == "null" {
+		*c = CardData{}
+		return nil
+	}
+	type Alias CardData
+	aux := &struct {
+		License json.RawMessage `json:"license"`
+		*Alias
+	}{
+		Alias: (*Alias)(c),
+	}
+	// Pass aux (not &aux): unmarshaling JSON null into &aux would nil the pointer
+	// and panic on aux.License below.
+	if err := json.Unmarshal(data, aux); err != nil {
+		return err
+	}
+	license, err := parseCardLicense(aux.License)
+	if err != nil {
+		return err
+	}
+	c.License = license
+	return nil
+}
+
+// parseCardLicense normalizes Hub license JSON (string or []string) to []string.
+func parseCardLicense(raw json.RawMessage) ([]string, error) {
+	if len(raw) == 0 || string(raw) == "null" {
+		return nil, nil
+	}
+	var asString string
+	if err := json.Unmarshal(raw, &asString); err == nil {
+		if asString == "" {
+			return nil, nil
+		}
+		return []string{asString}, nil
+	}
+	var asList []string
+	if err := json.Unmarshal(raw, &asList); err == nil {
+		parts := make([]string, 0, len(asList))
+		for _, p := range asList {
+			if p == "" {
+				continue
+			}
+			parts = append(parts, p)
+		}
+		if len(parts) == 0 {
+			return nil, nil
+		}
+		return parts, nil
+	}
+	return nil, fmt.Errorf("cardData.license: unsupported JSON value %s", string(raw))
 }
 
 // TransformersInfo contains information related to the Hugging Face Transformers integration.

diff --git a/hub/info_test.go b/hub/info_test.go
@@ -48,7 +48,7 @@ func TestRepoInfoUnmarshal(t *testing.T) {
 	// Assert cardData
 	require.NotNil(t, info.CardData)
 	assert.Equal(t, "transformers", info.CardData.LibraryName)
-	assert.Equal(t, "apache-2.0", info.CardData.License)
+	assert.Equal(t, []string{"apache-2.0"}, info.CardData.License)
 	assert.Equal(t, "https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE", info.CardData.LicenseLink)
 	assert.Equal(t, "text-generation", info.CardData.PipelineTag)
 	assert.Equal(t, []any{"Qwen/Qwen3-0.6B-Base"}, info.CardData.BaseModel)
@@ -102,3 +102,117 @@ func TestRepoInfoLegacyModelID(t *testing.T) {
 	require.NoError(t, err)
 	assert.Equal(t, "legacy_model", info.ModelID)
 }
+
+// Hugging Face dataset cards often set license as a YAML list, which the Hub API
+// returns as a JSON array (e.g. bigcode/the-stack-v2 → ["other"], openai/gsm8k → ["mit"]).
+func TestRepoInfoUnmarshal_LicenseAsArray(t *testing.T) {
+	jsonData := `{
+		"id": "bigcode/the-stack-v2",
+		"sha": "e565caa3a78c2423bd374333a472b049eb090e47",
+		"siblings": [{"rfilename": "README.md"}],
+		"cardData": {
+			"license": ["other"],
+			"library_name": "datasets",
+			"pipeline_tag": "text-generation",
+			"pretty_name": "The Stack v2"
+		}
+	}`
+	var info RepoInfo
+	err := json.Unmarshal([]byte(jsonData), &info)
+	require.NoError(t, err, "cardData.license as JSON array must unmarshal (Hub datasets often use list licenses)")
+	require.NotNil(t, info.CardData)
+	assert.Equal(t, []string{"other"}, info.CardData.License)
+	assert.Equal(t, "datasets", info.CardData.LibraryName)
+	assert.Equal(t, "text-generation", info.CardData.PipelineTag)
+	assert.Equal(t, "e565caa3a78c2423bd374333a472b049eb090e47", info.CommitHash)
+}
+
+func TestRepoInfoUnmarshal_LicenseAsString(t *testing.T) {
+	jsonData := `{"id":"demo","cardData":{"license":"apache-2.0"}}`
+	var info RepoInfo
+	require.NoError(t, json.Unmarshal([]byte(jsonData), &info))
+	require.NotNil(t, info.CardData)
+	assert.Equal(t, []string{"apache-2.0"}, info.CardData.License)
+}
+
+func TestRepoInfoUnmarshal_LicenseAsMultiValueArray(t *testing.T) {
+	jsonData := `{"id":"demo","cardData":{"license":["cc-by-sa-3.0","gfdl"]}}`
+	var info RepoInfo
+	require.NoError(t, json.Unmarshal([]byte(jsonData), &info))
+	require.NotNil(t, info.CardData)
+	assert.Equal(t, []string{"cc-by-sa-3.0", "gfdl"}, info.CardData.License)
+}
+
+func TestRepoInfoUnmarshal_LicenseMissingOrNullOrEmpty(t *testing.T) {
+	t.Parallel()
+	cases := []struct {
+		name string
+		json string
+	}{
+		{name: "missing", json: `{"id":"demo","cardData":{"library_name":"datasets"}}`},
+		{name: "null", json: `{"id":"demo","cardData":{"license":null}}`},
+		{name: "empty array", json: `{"id":"demo","cardData":{"license":[]}}`},
+		{name: "empty string", json: `{"id":"demo","cardData":{"license":""}}`},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+			var info RepoInfo
+			require.NoError(t, json.Unmarshal([]byte(tc.json), &info))
+			require.NotNil(t, info.CardData)
+			assert.Empty(t, info.CardData.License)
+		})
+	}
+}
+
+func TestRepoInfoUnmarshal_LicenseInvalidType(t *testing.T) {
+	t.Parallel()
+	cases := []string{
+		`{"id":"demo","cardData":{"license":123}}`,
+		`{"id":"demo","cardData":{"license":true}}`,
+		`{"id":"demo","cardData":{"license":{"name":"mit"}}}`,
+	}
+	for _, jsonData := range cases {
+		var info RepoInfo
+		err := json.Unmarshal([]byte(jsonData), &info)
+		require.Error(t, err, "expected error for %s", jsonData)
+		assert.Contains(t, err.Error(), "cardData.license")
+	}
+}
+
+func TestCardDataUnmarshal_Null(t *testing.T) {
+	t.Parallel()
+	var c CardData
+	require.NoError(t, json.Unmarshal([]byte("null"), &c))
+	assert.Equal(t, CardData{}, c)
+}
+
+func TestRepoInfoUnmarshal_CardDataNull(t *testing.T) {
+	t.Parallel()
+	var info RepoInfo
+	require.NoError(t, json.Unmarshal([]byte(`{"id":"demo","cardData":null}`), &info))
+	assert.Equal(t, "demo", info.ID)
+	assert.Nil(t, info.CardData)
+}
+
+func TestDownloadInfo_DatasetLicenseArray(t *testing.T) {
+	// Live Hub call: openai/gsm8k has cardData.license = ["mit"] and is public/ungated.
+	if testing.Short() {
+		t.Skip("skipping live Hub call in short mode")
+	}
+	t.Parallel()
+	dir := t.TempDir()
+	repo := New("openai/gsm8k").
+		WithType(RepoTypeDataset).
+		WithCacheDir(dir).
+		WithProgressBar(false)
+	repo.Verbosity = 0
+
+	err := repo.DownloadInfo(false)
+	require.NoError(t, err, "DownloadInfo must succeed for datasets whose card license is a JSON array")
+	info := repo.Info()
+	require.NotNil(t, info)
+	require.NotEmpty(t, info.CommitHash)
+	require.NotNil(t, info.CardData)
+	assert.Equal(t, []string{"mit"}, info.CardData.License)
+}