|
| 1 | +// Package main provides a command-line tool to fetch models from Hugging Face Router |
| 2 | +// and generate a configuration file for the provider. |
| 3 | +package main |
| 4 | + |
| 5 | +import ( |
| 6 | + "context" |
| 7 | + "encoding/json" |
| 8 | + "fmt" |
| 9 | + "io" |
| 10 | + "log" |
| 11 | + "net/http" |
| 12 | + "os" |
| 13 | + "slices" |
| 14 | + "time" |
| 15 | + |
| 16 | + "github.com/charmbracelet/catwalk/pkg/catwalk" |
| 17 | +) |
| 18 | + |
| 19 | +// SupportedProviders defines which providers we want to support. |
| 20 | +// Add or remove providers from this slice to control which ones are included. |
| 21 | +var SupportedProviders = []string{ |
| 22 | + // "together", // Multiple issues |
| 23 | + "fireworks-ai", |
| 24 | + // "nebius", // Tool Call ID not unique |
| 25 | + // "novita", // Usage report is wrong |
| 26 | + "groq", |
| 27 | + "cerebras", |
| 28 | + // "hyperbolic", |
| 29 | + // "nscale", |
| 30 | + // "sambanova", |
| 31 | + // "cohere", |
| 32 | + "hf-inference", |
| 33 | +} |
| 34 | + |
| 35 | +// HFModel represents a model from the Hugging Face Router API. |
| 36 | +type HFModel struct { |
| 37 | + ID string `json:"id"` |
| 38 | + Object string `json:"object"` |
| 39 | + Created int64 `json:"created"` |
| 40 | + OwnedBy string `json:"owned_by"` |
| 41 | + Providers []HFProvider `json:"providers"` |
| 42 | +} |
| 43 | + |
| 44 | +// HFProvider represents a provider configuration for a model. |
| 45 | +type HFProvider struct { |
| 46 | + Provider string `json:"provider"` |
| 47 | + Status string `json:"status"` |
| 48 | + ContextLength int64 `json:"context_length,omitempty"` |
| 49 | + Pricing *HFPricing `json:"pricing,omitempty"` |
| 50 | + SupportsTools bool `json:"supports_tools"` |
| 51 | + SupportsStructuredOutput bool `json:"supports_structured_output"` |
| 52 | +} |
| 53 | + |
| 54 | +// HFPricing contains the pricing information for a provider. |
| 55 | +type HFPricing struct { |
| 56 | + Input float64 `json:"input"` |
| 57 | + Output float64 `json:"output"` |
| 58 | +} |
| 59 | + |
| 60 | +// HFModelsResponse is the response structure for the Hugging Face Router models API. |
| 61 | +type HFModelsResponse struct { |
| 62 | + Object string `json:"object"` |
| 63 | + Data []HFModel `json:"data"` |
| 64 | +} |
| 65 | + |
| 66 | +func fetchHuggingFaceModels() (*HFModelsResponse, error) { |
| 67 | + client := &http.Client{Timeout: 30 * time.Second} |
| 68 | + req, _ := http.NewRequestWithContext( |
| 69 | + context.Background(), |
| 70 | + "GET", |
| 71 | + "https://router.huggingface.co/v1/models", |
| 72 | + nil, |
| 73 | + ) |
| 74 | + req.Header.Set("User-Agent", "Crush-Client/1.0") |
| 75 | + resp, err := client.Do(req) |
| 76 | + if err != nil { |
| 77 | + return nil, err //nolint:wrapcheck |
| 78 | + } |
| 79 | + defer resp.Body.Close() //nolint:errcheck |
| 80 | + if resp.StatusCode != 200 { |
| 81 | + body, _ := io.ReadAll(resp.Body) |
| 82 | + return nil, fmt.Errorf("status %d: %s", resp.StatusCode, body) |
| 83 | + } |
| 84 | + var mr HFModelsResponse |
| 85 | + if err := json.NewDecoder(resp.Body).Decode(&mr); err != nil { |
| 86 | + return nil, err //nolint:wrapcheck |
| 87 | + } |
| 88 | + return &mr, nil |
| 89 | +} |
| 90 | + |
| 91 | +// findContextWindow looks for a context window from any provider for the given model. |
| 92 | +func findContextWindow(model HFModel) int64 { |
| 93 | + for _, provider := range model.Providers { |
| 94 | + if provider.ContextLength > 0 { |
| 95 | + return provider.ContextLength |
| 96 | + } |
| 97 | + } |
| 98 | + return 0 |
| 99 | +} |
| 100 | + |
| 101 | +func main() { |
| 102 | + modelsResp, err := fetchHuggingFaceModels() |
| 103 | + if err != nil { |
| 104 | + log.Fatal("Error fetching Hugging Face models:", err) |
| 105 | + } |
| 106 | + |
| 107 | + hfProvider := catwalk.Provider{ |
| 108 | + Name: "Hugging Face", |
| 109 | + ID: catwalk.InferenceProviderHuggingFace, |
| 110 | + APIKey: "$HUGGINGFACE_API_KEY", |
| 111 | + APIEndpoint: "https://router.huggingface.co/v1", |
| 112 | + Type: catwalk.TypeOpenAI, |
| 113 | + DefaultLargeModelID: "moonshotai/Kimi-K2-Instruct-0905:groq", |
| 114 | + DefaultSmallModelID: "openai/gpt-oss-20b", |
| 115 | + Models: []catwalk.Model{}, |
| 116 | + DefaultHeaders: map[string]string{ |
| 117 | + "HTTP-Referer": "https://charm.land", |
| 118 | + "X-Title": "Crush", |
| 119 | + }, |
| 120 | + } |
| 121 | + |
| 122 | + for _, model := range modelsResp.Data { |
| 123 | + // Find context window from any provider for this model |
| 124 | + fallbackContextLength := findContextWindow(model) |
| 125 | + if fallbackContextLength == 0 { |
| 126 | + fmt.Printf("Skipping model %s - no context window found in any provider\n", model.ID) |
| 127 | + continue |
| 128 | + } |
| 129 | + |
| 130 | + for _, provider := range model.Providers { |
| 131 | + // Skip unsupported providers |
| 132 | + if !slices.Contains(SupportedProviders, provider.Provider) { |
| 133 | + continue |
| 134 | + } |
| 135 | + |
| 136 | + // Skip providers that don't support tools |
| 137 | + if !provider.SupportsTools { |
| 138 | + continue |
| 139 | + } |
| 140 | + |
| 141 | + // Skip non-live providers |
| 142 | + if provider.Status != "live" { |
| 143 | + continue |
| 144 | + } |
| 145 | + |
| 146 | + // Create model with provider-specific ID and name |
| 147 | + modelID := fmt.Sprintf("%s:%s", model.ID, provider.Provider) |
| 148 | + modelName := fmt.Sprintf("%s (%s)", model.ID, provider.Provider) |
| 149 | + |
| 150 | + // Use provider's context length, or fallback if not available |
| 151 | + contextLength := provider.ContextLength |
| 152 | + if contextLength == 0 { |
| 153 | + contextLength = fallbackContextLength |
| 154 | + } |
| 155 | + |
| 156 | + // Calculate pricing (convert from per-token to per-1M tokens) |
| 157 | + var costPer1MIn, costPer1MOut float64 |
| 158 | + if provider.Pricing != nil { |
| 159 | + costPer1MIn = provider.Pricing.Input |
| 160 | + costPer1MOut = provider.Pricing.Output |
| 161 | + } |
| 162 | + |
| 163 | + // Set default max tokens (conservative estimate) |
| 164 | + defaultMaxTokens := min(contextLength/4, 10000) |
| 165 | + |
| 166 | + m := catwalk.Model{ |
| 167 | + ID: modelID, |
| 168 | + Name: modelName, |
| 169 | + CostPer1MIn: costPer1MIn, |
| 170 | + CostPer1MOut: costPer1MOut, |
| 171 | + CostPer1MInCached: 0, // Not provided by HF Router |
| 172 | + CostPer1MOutCached: 0, // Not provided by HF Router |
| 173 | + ContextWindow: contextLength, |
| 174 | + DefaultMaxTokens: defaultMaxTokens, |
| 175 | + CanReason: false, // Not provided by HF Router |
| 176 | + SupportsImages: false, // Not provided by HF Router |
| 177 | + } |
| 178 | + |
| 179 | + hfProvider.Models = append(hfProvider.Models, m) |
| 180 | + fmt.Printf("Added model %s with context window %d from provider %s\n", |
| 181 | + modelID, contextLength, provider.Provider) |
| 182 | + } |
| 183 | + } |
| 184 | + |
| 185 | + // Save the JSON in internal/providers/configs/huggingface.json |
| 186 | + data, err := json.MarshalIndent(hfProvider, "", " ") |
| 187 | + if err != nil { |
| 188 | + log.Fatal("Error marshaling Hugging Face provider:", err) |
| 189 | + } |
| 190 | + |
| 191 | + if err := os.WriteFile("internal/providers/configs/huggingface.json", data, 0o600); err != nil { |
| 192 | + log.Fatal("Error writing Hugging Face provider config:", err) |
| 193 | + } |
| 194 | + |
| 195 | + fmt.Printf("Generated huggingface.json with %d models\n", len(hfProvider.Models)) |
| 196 | +} |
0 commit comments