feat(huggingface): add HuggingFace provider

deploy/example.config.ts

@@ -2,7 +2,7 @@ import { env } from 'cloudflare:workers'
 import type { Config } from '@deploy/types'
 
 // can be whatever you want, just used to make linking apiKeys to providers typesafe.
-type ProviderKeys = 'a' | 'b' | 'c' | 'd' | 'e'
+type ProviderKeys = 'a' | 'b' | 'c' | 'd' | 'e' | 'huggingface'
 
 // projects, users and keys must have numeric keys, using constants here to make it easier to understand
 // of course, keys must be unique within a type (e.g. project ids must be unique) but users and projects can have the same id
@@ -67,6 +67,12 @@ export const config: Config<ProviderKeys> = {
       injectCost: true,
       credentials: env.AWS_BEARER_TOKEN_BEDROCK,
     },
+    huggingface: {
+      providerId: 'huggingface',
+      baseUrl: 'https://api-inference.huggingface.co',
+      injectCost: true,
+      credentials: env.HF_TOKEN,
+    },
   },
   // routing groups for load balancing and fallback
   routingGroups: {

deploy/example.env.local

@@ -20,5 +20,8 @@ AWS_BEARER_TOKEN_BEDROCK=...
 # python -c "import json;print(json.dumps(json.loads(open(input('Service account JSON file path: ')).read())))"
 GOOGLE_SERVICE_ACCOUNT_KEY=full service google service account key...
 
+# same for Hugging Face, generate a token (you would use env.HF_TOKEN in config.ts)
+HF_TOKEN=...
+
 # password for viewing /status/
 STATUS_AUTH_API_KEY="change-me!"

gateway/src/providers/huggingface.ts

@@ -0,0 +1,11 @@
+import type { ModelAPI } from '../api'
+import { ChatCompletionAPI } from '../api/chat'
+import { DefaultProviderProxy } from './default'
+
+export class HuggingFaceProvider extends DefaultProviderProxy {
+  defaultBaseUrl = 'https://api-inference.huggingface.co'
+
+  protected modelAPI(): ModelAPI {
+    return new ChatCompletionAPI('huggingface')
+  }
+}

gateway/src/providers/index.ts

@@ -22,6 +22,7 @@ import { BedrockProvider } from './bedrock'
 import { DefaultProviderProxy, type ProviderOptions } from './default'
 import { GoogleVertexProvider } from './google'
 import { GroqProvider } from './groq'
+import { HuggingFaceProvider } from './huggingface'
 import { OpenAIProvider } from './openai'
 import { TestProvider } from './test'
 
@@ -39,6 +40,8 @@ export function getProvider(providerId: ProviderID): ProviderSig {
       return AnthropicProvider
     case 'bedrock':
       return BedrockProvider
+    case 'huggingface':
+      return HuggingFaceProvider
     case 'test':
       return TestProvider
     default:

gateway/src/types.ts

@@ -38,7 +38,7 @@ export interface ApiKeyInfo<ProviderKey extends string = string> {
   otelSettings?: OtelSettings
 }
 
-export type ProviderID = 'groq' | 'openai' | 'google-vertex' | 'anthropic' | 'test' | 'bedrock'
+export type ProviderID = 'groq' | 'openai' | 'google-vertex' | 'anthropic' | 'test' | 'bedrock' | 'huggingface'
 // TODO | 'azure' | 'fireworks' | 'mistral' | 'cohere'
 
 const providerIds: Record<ProviderID, boolean> = {
@@ -48,6 +48,7 @@ const providerIds: Record<ProviderID, boolean> = {
   anthropic: true,
   test: true,
   bedrock: true,
+  huggingface: true,
 }
 
 export const providerIdsArray = Object.keys(providerIds) as ProviderID[]