Fix: Explicitly use Featherless AI provider for abliterated models via HF Router

server/_core/llm.ts

@@ -216,7 +216,7 @@ const resolveApiUrl = (model: string) => {
     return `${ENV.forgeApiUrl.replace(/\/$/, "")}/v1/chat/completions`;
   }
   
-  // Default to the new Hugging Face Router endpoint (api-inference.huggingface.co is deprecated)
+  // Use the Hugging Face Router endpoint to leverage external providers
   return "https://router.huggingface.co/v1/chat/completions";
 };
 
@@ -327,6 +327,9 @@ export async function invokeLLM(params: InvokeParams): Promise<InvokeResult> {
     headers: {
       "content-type": "application/json",
       authorization: `Bearer ${apiKey}`,
+      "x-use-cache": "false",
+      "x-wait-for-model": "true",
+      "x-inference-provider": "featherless",
     },
     body: JSON.stringify(payload),
   });