cfdca35 · crush · git.iko.soy

internal/discover/discover.go+10

compare with main

unifiedsplit

@@ -17,6 +17,16 @@ import (

17// even if the caller forgets to set a context deadline.

18var httpClient = &http.Client{Timeout: 10 * time.Second}

19

20+// stripV1Suffix removes a trailing /v1 from a base URL. Enricher

21+// endpoints (e.g. Ollama's /api/show, LM Studio's /api/v1/models) are

22+// served at the server root, not under the OpenAI-compatible /v1

23+// prefix. Since provider configs typically include /v1 in the base URL

24+// for chat completions, enrichers must strip it before constructing

25+// their own request paths.

26+func stripV1Suffix(baseURL string) string {

27+ return strings.TrimSuffix(strings.TrimRight(baseURL, "/"), "/v1")

28+}

29+

20// doRequest builds and executes an authenticated HTTP request using the

30// doRequest builds and executes an authenticated HTTP request using the

21// shared client. It resolves variable references in the base URL, API

31// shared client. It resolves variable references in the base URL, API

22// key, and extra headers via the provided Resolver. The path is joined

32// key, and extra headers via the provided Resolver. The path is joined

internal/discover/discover_test.go+19

@@ -200,3 +200,22 @@ func TestDiscoverModels_NoAuthWhenNoAPIKey(t *testing.T) {

200 require.NoError(t, err)

201 require.Len(t, models, 1)

202}

203+

204+func TestStripV1Suffix(t *testing.T) {

205+ t.Parallel()

206+ tests := []struct {

207+ input string

208+ want string

209+ }{

210+ {"http://localhost:8000/v1", "http://localhost:8000"},

211+ {"http://localhost:8000/v1/", "http://localhost:8000"},

212+ {"http://localhost:8000", "http://localhost:8000"},

213+ {"http://localhost:8000/", "http://localhost:8000"},

214+ {"http://localhost:8000/api/v1", "http://localhost:8000/api"},

215+ {"", ""},

216+ }

217+ for _, tt := range tests {

218+ got := stripV1Suffix(tt.input)

219+ require.Equal(t, tt.want, got, "stripV1Suffix(%q)", tt.input)

220+ }

221+}

internal/discover/litellm.go+1−1

@@ -41,7 +41,7 @@ func init() {

41type litellmEnricher struct{}

42

43func (e *litellmEnricher) EnrichModels(ctx context.Context, cfg Config, resolver Resolver, models []catwalk.Model) ([]catwalk.Model, error) {

44− resp, err := doRequest(ctx, http.MethodGet, cfg.BaseURL, "/model/info", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

44+ resp, err := doRequest(ctx, http.MethodGet, stripV1Suffix(cfg.BaseURL), "/model/info", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

45 if err != nil {

46 return models, nil

47 }

internal/discover/litellm_test.go+4−2

@@ -42,8 +42,10 @@ func TestLitellmEnricher(t *testing.T) {

42 defer srv.Close()

43

44 cfg := Config{

45− ID: "test-litellm",

45+ ID: "test-litellm",

46+ // Base URL includes /v1 (as Crush configures it); the

47+ // enricher strips it so /model/info resolves at the root.

46− BaseURL: srv.URL,

48+ BaseURL: srv.URL + "/v1",

47 APIKey: "test-key",

49 APIKey: "test-key",

48 }

50 }

49 models := []catwalk.Model{

51 models := []catwalk.Model{

internal/discover/lmstudio.go+7−6

@@ -13,10 +13,11 @@ func init() {

13}

14

15// lmstudioModelsResponse mirrors the response from LM Studio's native

16+// GET /api/v1/models endpoint. The model array is returned under the

17+// "models" key (not "data" like the OpenAI-compatible endpoint). Only

16−// GET /api/v1/models endpoint. Only the fields we care about are

18+// the fields we care about are decoded.

17−// decoded.

18type lmstudioModelsResponse struct {

19type lmstudioModelsResponse struct {

19− Data []lmstudioModelEntry `json:"data"`

20+ Models []lmstudioModelEntry `json:"models"`

20}

21}

21

22

22// lmstudioModelEntry is a single entry from /api/v1/models.

23// lmstudioModelEntry is a single entry from /api/v1/models.

@@ -44,7 +45,7 @@ type lmstudioInstanceConfig struct {

44type lmstudioEnricher struct{}

45type lmstudioEnricher struct{}

45

46

46func (e *lmstudioEnricher) EnrichModels(ctx context.Context, cfg Config, resolver Resolver, models []catwalk.Model) ([]catwalk.Model, error) {

47func (e *lmstudioEnricher) EnrichModels(ctx context.Context, cfg Config, resolver Resolver, models []catwalk.Model) ([]catwalk.Model, error) {

47− resp, err := doRequest(ctx, http.MethodGet, cfg.BaseURL, "/api/v1/models", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

48+ resp, err := doRequest(ctx, http.MethodGet, stripV1Suffix(cfg.BaseURL), "/api/v1/models", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

48 if err != nil {

49 if err != nil {

49 return models, nil

50 return models, nil

50 }

51 }

@@ -60,8 +61,8 @@ func (e *lmstudioEnricher) EnrichModels(ctx context.Context, cfg Config, resolve

60 }

61 }

61

62

62 // Index by key for O(1) lookup.

63 // Index by key for O(1) lookup.

63− metaByKey := make(map[string]lmstudioModelEntry, len(modelsResp.Data))

64+ metaByKey := make(map[string]lmstudioModelEntry, len(modelsResp.Models))

64− for _, m := range modelsResp.Data {

65+ for _, m := range modelsResp.Models {

65 metaByKey[m.Key] = m

66 metaByKey[m.Key] = m

66 }

67 }

67

68

internal/discover/lmstudio_test.go+19−15

@@ -19,24 +19,28 @@ func TestLmstudioEnricher(t *testing.T) {

19 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {

20 require.Equal(t, "/api/v1/models", r.URL.Path)

21 w.Header().Set("Content-Type", "application/json")

22− json.NewEncoder(w).Encode(lmstudioModelsResponse{

22+ // Raw JSON matching LM Studio's real /api/v1/models wire

23− Data: []lmstudioModelEntry{

23+ // format: array under "models", not "data".

24+ _, _ = w.Write([]byte(`{

25+ "models": [

24 {

26 {

25− Key: "qwen2.5-7b-instruct",

27+ "key": "qwen2.5-7b-instruct",

26− DisplayName: "Qwen 2.5 7B Instruct",

28+ "display_name": "Qwen 2.5 7B Instruct",

27− MaxContextLength: 32768,

29+ "max_context_length": 32768

28 },

30 },

29 {

31 {

30− Key: "llama-3.1-8b",

32+ "key": "llama-3.1-8b",

31− DisplayName: "Llama 3.1 8B",

33+ "display_name": "Llama 3.1 8B",

32− MaxContextLength: 131072,

34+ "max_context_length": 131072

33− },

35+ }

34− },

36+ ]

35− })

37+ }`))

36 }))

38 }))

37 defer srv.Close()

39 defer srv.Close()

38

40

41+ // Base URL includes /v1 (as Crush configures it); the enricher

42+ // strips it so the native endpoint resolves at the server root.

39− cfg := Config{ID: "test-lmstudio", BaseURL: srv.URL}

43+ cfg := Config{ID: "test-lmstudio", BaseURL: srv.URL + "/v1"}

40 models := []catwalk.Model{

44 models := []catwalk.Model{

41 {ID: "qwen2.5-7b-instruct", Name: "qwen2.5-7b-instruct"},

45 {ID: "qwen2.5-7b-instruct", Name: "qwen2.5-7b-instruct"},

42 {ID: "llama-3.1-8b", Name: "llama-3.1-8b"},

46 {ID: "llama-3.1-8b", Name: "llama-3.1-8b"},

@@ -60,7 +64,7 @@ func TestLmstudioEnricher(t *testing.T) {

60 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

64 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

61 w.Header().Set("Content-Type", "application/json")

65 w.Header().Set("Content-Type", "application/json")

62 json.NewEncoder(w).Encode(lmstudioModelsResponse{

66 json.NewEncoder(w).Encode(lmstudioModelsResponse{

63− Data: []lmstudioModelEntry{

67+ Models: []lmstudioModelEntry{

64 {

68 {

65 Key: "m1",

69 Key: "m1",

66 MaxContextLength: 131072,

70 MaxContextLength: 131072,

@@ -87,7 +91,7 @@ func TestLmstudioEnricher(t *testing.T) {

87 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

91 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

88 w.Header().Set("Content-Type", "application/json")

92 w.Header().Set("Content-Type", "application/json")

89 json.NewEncoder(w).Encode(lmstudioModelsResponse{

93 json.NewEncoder(w).Encode(lmstudioModelsResponse{

90− Data: []lmstudioModelEntry{

94+ Models: []lmstudioModelEntry{

91 {

95 {

92 Key: "m1",

96 Key: "m1",

93 DisplayName: "Should Not Override",

97 DisplayName: "Should Not Override",

@@ -132,7 +136,7 @@ func TestLmstudioEnricher(t *testing.T) {

132 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

136 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {

133 w.Header().Set("Content-Type", "application/json")

137 w.Header().Set("Content-Type", "application/json")

134 json.NewEncoder(w).Encode(lmstudioModelsResponse{

138 json.NewEncoder(w).Encode(lmstudioModelsResponse{

135− Data: []lmstudioModelEntry{

139+ Models: []lmstudioModelEntry{

136 {Key: "m1", DisplayName: "API Name"},

140 {Key: "m1", DisplayName: "API Name"},

137 },

141 },

138 })

142 })

internal/discover/ollama.go+1−1

@@ -51,7 +51,7 @@ func (e *ollamaEnricher) EnrichModels(ctx context.Context, cfg Config, resolver

51 sem <- struct{}{}

52 defer func() { <-sem }()

53

54− resp, err := doRequest(ctx, http.MethodPost, cfg.BaseURL, "/api/show",

54+ resp, err := doRequest(ctx, http.MethodPost, stripV1Suffix(cfg.BaseURL), "/api/show",

55 cfg.APIKey, cfg.ExtraHeaders, resolver,

56 map[string]string{"model": models[idx].ID})

57 if err != nil {

internal/discover/ollama_test.go+3−1

@@ -43,7 +43,9 @@ func TestOllamaEnricher(t *testing.T) {

43 }))

44 defer srv.Close()

45

46+ // Base URL includes /v1 (as Crush configures it); the enricher

47+ // strips it so /api/show resolves at the server root.

46− cfg := Config{ID: "test-ollama", BaseURL: srv.URL}

48+ cfg := Config{ID: "test-ollama", BaseURL: srv.URL + "/v1"}

47 models := []catwalk.Model{

49 models := []catwalk.Model{

48 {ID: "llama3:latest", Name: "llama3:latest"},

50 {ID: "llama3:latest", Name: "llama3:latest"},

49 {ID: "qwen2:latest", Name: "qwen2:latest"},

51 {ID: "qwen2:latest", Name: "qwen2:latest"},

internal/discover/omlx.go+6−3

@@ -13,7 +13,7 @@ func init() {

13}

14

15// omlxModelsStatusResponse mirrors the response from oMLX's

16−// GET /v1/models/status endpoint, which returns model metadata

16+// GET /models/status endpoint, which returns model metadata

17// including max_context_window and max_tokens.

18type omlxModelsStatusResponse struct {

19 Models []omlxModelStatus `json:"models"`

@@ -26,13 +26,16 @@ type omlxModelStatus struct {

26 MaxTokens *int64 `json:"max_tokens"`

27}

28

29−// omlxEnricher fetches model metadata from oMLX's /v1/models/status

29+// omlxEnricher fetches model metadata from oMLX's /models/status

30// endpoint and populates context window and max tokens on discovered

31// models.

32type omlxEnricher struct{}

33

34func (e *omlxEnricher) EnrichModels(ctx context.Context, cfg Config, resolver Resolver, models []catwalk.Model) ([]catwalk.Model, error) {

35+ // oMLX serves /models/status under the OpenAI-compatible /v1

36+ // namespace, so the path is relative to the configured base URL

37+ // (which already includes /v1) rather than the server root.

35− resp, err := doRequest(ctx, http.MethodGet, cfg.BaseURL, "/v1/models/status", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

38+ resp, err := doRequest(ctx, http.MethodGet, cfg.BaseURL, "/models/status", cfg.APIKey, cfg.ExtraHeaders, resolver, nil)

36 if err != nil {

39 if err != nil {

37 return models, nil

40 return models, nil

38 }

41 }

internal/discover/omlx_test.go+3−1

@@ -17,6 +17,8 @@ func TestOmlxEnricher(t *testing.T) {

17 t.Run("populates context window and max tokens from /v1/models/status", func(t *testing.T) {

18 t.Parallel()

19 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {

20+ // oMLX serves the status endpoint under /v1 (base URL

21+ // includes /v1), so the resolved path keeps the prefix.

20 require.Equal(t, "/v1/models/status", r.URL.Path)

22 require.Equal(t, "/v1/models/status", r.URL.Path)

21 w.Header().Set("Content-Type", "application/json")

23 w.Header().Set("Content-Type", "application/json")

22 json.NewEncoder(w).Encode(omlxModelsStatusResponse{

24 json.NewEncoder(w).Encode(omlxModelsStatusResponse{

@@ -28,7 +30,7 @@ func TestOmlxEnricher(t *testing.T) {

28 }))

30 }))

29 defer srv.Close()

31 defer srv.Close()

30

32

31− cfg := Config{ID: "test-omlx", BaseURL: srv.URL}

33+ cfg := Config{ID: "test-omlx", BaseURL: srv.URL + "/v1"}

32 models := []catwalk.Model{

34 models := []catwalk.Model{

33 {ID: "qwen3:latest", Name: "qwen3:latest"},

35 {ID: "qwen3:latest", Name: "qwen3:latest"},

34 {ID: "llama3:latest", Name: "llama3:latest"},

36 {ID: "llama3:latest", Name: "llama3:latest"},