Update app.py

app.py

@@ -336,8 +336,8 @@ def load_llm_model(model_path: str) -> Tuple[str, str]:
             state.llm_model = LLM(
                 model=model_path,
                 tensor_parallel_size=tp_size,
-                gpu_memory_utilization=0.20,
-                max_model_len=10000
+                gpu_memory_utilization=0.15,
+                max_model_len=5000
             )
             state.llm_tokenizer = state.llm_model.get_tokenizer()
             return f"✓ LLM loaded from {model_path} (vLLM, tp={tp_size})", ""
@@ -916,10 +916,11 @@ def match_trials(patient_summary: str, patient_boilerplate: str, top_k: int = 20
         
         result_df = top_trials[display_cols].reset_index(drop=True)
         
-        # Round probability columns to 2 decimal places
-        result_df['eligibility_probability'] = result_df['eligibility_probability'].round(2)
-        result_df['exclusion_probability'] = result_df['exclusion_probability'].round(2)
-        result_df['similarity_score'] = result_df['similarity_score'].round(3)
+        # Convert probability columns to strings with fixed decimal places
+        # This ensures Gradio displays them correctly without extra decimals
+        result_df['eligibility_probability'] = result_df['eligibility_probability'].apply(lambda x: f"{x:.2f}")
+        result_df['exclusion_probability'] = result_df['exclusion_probability'].apply(lambda x: f"{x:.2f}")
+        result_df['similarity_score'] = result_df['similarity_score'].apply(lambda x: f"{x:.3f}")
         
         return result_df