Update app.py

app.py

@@ -276,7 +276,7 @@ def load_tagger_model(model_path: str) -> Tuple[str, str]:
             device=0 if state.device == "cuda" else -1,
             truncation=True,
             padding="max_length",
-            max_length=128
+            max_length=512
         )
         return f"✓ Tagger model loaded from {model_path}", ""
     except Exception as e:
@@ -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=0.50,
-                max_model_len=15000
+                gpu_memory_utilization=0.20,
+                max_model_len=10000
             )
             state.llm_tokenizer = state.llm_model.get_tokenizer()
             return f"✓ LLM loaded from {model_path} (vLLM, tp={tp_size})", ""
@@ -679,7 +679,7 @@ Now, write your summary. Do not add preceding text before the abstraction, and d
                 SamplingParams(
                     temperature=0.0,
                     top_k=1,
-                    max_tokens=4000,
+                    max_tokens=7500,
                     repetition_penalty=1.2
                 )
             )
@@ -696,7 +696,7 @@ Now, write your summary. Do not add preceding text before the abstraction, and d
             with torch.no_grad():
                 outputs = state.llm_model.generate(
                     input_ids,
-                    max_new_tokens=4000,
+                    max_new_tokens=7500,
                     temperature=0.00,
                     do_sample=True,
                     repetition_penalty=1.2
@@ -774,7 +774,7 @@ def extract_trial_spaces(trial_text: str) -> str:
                 SamplingParams(
                     temperature=0.0,
                     top_k=1,
-                    max_tokens=4000,
+                    max_tokens=7500,
                     repetition_penalty=1.3
                 )
             )
@@ -791,7 +791,7 @@ def extract_trial_spaces(trial_text: str) -> str:
             with torch.no_grad():
                 outputs = state.llm_model.generate(
                     input_ids,
-                    max_new_tokens=1500,
+                    max_new_tokens=7500,
                     temperature=0.0,
                     do_sample=False,
                     repetition_penalty=1.3
@@ -936,9 +936,19 @@ def get_trial_details(df: pd.DataFrame, evt: gr.SelectData) -> str:
         
         row_idx = evt.index[0]
         nct_id = df.iloc[row_idx]['nct_id']
+        this_space = df.iloc[row_idx]['this_space']
         
-        # Find in original dataframe
-        trial_row = state.trial_spaces_df[state.trial_spaces_df['nct_id'] == nct_id].iloc[0]
+        # Find the specific trial space in original dataframe
+        # Match both NCT ID and the exact trial space text
+        matching_rows = state.trial_spaces_df[
+            (state.trial_spaces_df['nct_id'] == nct_id) & 
+            (state.trial_spaces_df['this_space'] == this_space)
+        ]
+        
+        if len(matching_rows) == 0:
+            return f"Error: Could not find matching trial space for {nct_id}"
+        
+        trial_row = matching_rows.iloc[0]
         
         # Create clinicaltrials.gov link
         ct_gov_link = f"https://clinicaltrials.gov/study/{nct_id}"
@@ -950,7 +960,7 @@ def get_trial_details(df: pd.DataFrame, evt: gr.SelectData) -> str:
 
 ---
 
-## Eligibility Criteria Summary
+## Eligibility Criteria Summary (Selected Space)
 {trial_row['this_space']}
 
 ## Full Trial Text