Trial and error.

app.py

@@ -34,40 +34,11 @@ class BasicAgent:
                 {"role": "user", "content": question}
             ]
 
-            print(f"Messages prepared for model: {messages}")
-
             # Generate response
-            if self.model and self.tokenizer:
-                printf("Using local model for generation...")
-
-                # Local model generation
-                text = self.tokenizer.apply_chat_template(
-                    messages,
-                    tokenize=False,
-                    add_generation_prompt=True
-                )
-                model_inputs = self.tokenizer([text], return_tensors="pt").to(self.model.device)
-
-                print(f"Model inputs prepared: {model_inputs}")
-
-                generated_ids = self.model.generate(
-                    **model_inputs,
-                    max_new_tokens=512
-                )
-                generated_ids = [
-                    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
-                ]
-
-                print(f"Generated IDs: {generated_ids}")
-
-                answer = self.tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-            else:
-                print("Using Inference API for generation...")
-
-                # Fallback to Inference API
-                response = self.client.chat(messages=messages)
-                answer = response.content
-                
+            print("Using Inference API for generation...")
+            response = self.client.text_generation(messages=messages)
+            print(f"Inference API response received. {len(response.choices)} choices.")
+            answer = response.choices[0].message.content
             print(f"Agent generated response (first 50 chars): {answer[:50]}...")
             return answer
         except Exception as e:
@@ -202,7 +173,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
 # --- Build Gradio Interface using Blocks ---
 with gr.Blocks() as demo:
-    gr.Markdown("# Basic Agent Evaluation Runner #2")
+    gr.Markdown("# Basic Agent Evaluation Runner #3")
     gr.Markdown(
         """
         **Instructions:**