Upload app.py with huggingface_hub

app.py

@@ -142,8 +142,8 @@ ANSWER: [your detailed answer]"""
 async def root():
     return HealthResponse(
         status="healthy",
-        model=HFInferenceModel.QWEN_72B.value,
-        provider=HFInferenceProvider.CEREBRAS.value,
+        model=HFInferenceModel.QWEN_7B.value,
+        provider=HFInferenceProvider.TOGETHER.value,
         judge_model=HFJudgeModel.QWEN_7B.value,
         judge_provider=HFInferenceProvider.TOGETHER.value,
         seeds_loaded=len(loaded_seeds) > 0
@@ -154,8 +154,8 @@ async def root():
 async def health():
     return HealthResponse(
         status="healthy",
-        model=HFInferenceModel.QWEN_72B.value,
-        provider=HFInferenceProvider.CEREBRAS.value,
+        model=HFInferenceModel.QWEN_7B.value,
+        provider=HFInferenceProvider.TOGETHER.value,
         judge_model=HFJudgeModel.QWEN_7B.value,
         judge_provider=HFInferenceProvider.TOGETHER.value,
         seeds_loaded=len(loaded_seeds) > 0