Update app.py

app.py

@@ -1,13 +1,31 @@
 import gradio as gr
-from docx import Document
-import tempfile
-from transformers import pipeline
 import time
-
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
 # ---- LOAD LLM ----
-llm = pipeline("text-generation", model="microsoft/Phi-4-mini-instruct", device=0)
+model_name = "microsoft/Phi-4-mini-instruct"
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.float16,
+    device_map="auto"  # automatically uses CPU or GPU if available
+)
 
+def llm(prompt, max_new_tokens=500, temperature=0.3):
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    output = model.generate(
+        **inputs,
+        max_new_tokens=max_new_tokens,
+        temperature=temperature,
+        do_sample=True,
+        top_p=0.95,
+        top_k=50,
+        pad_token_id=tokenizer.eos_token_id
+    )
+    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
+    return [{"generated_text": generated_text}]
 
 
 # Define all the screening questions
@@ -215,7 +233,6 @@ questions = [
     ("Have these difficulties been present since childhood and continue to affect your work or relationships?", "क्या ये कठिनाइयाँ बचपन से रही हैं और अब भी आपके काम या रिश्तों को प्रभावित कर रही हैं?"),
 ]
 
-
 # ---- STATE ----
 state = {
     "index": 0,
@@ -223,7 +240,6 @@ state = {
 }
 
 # ---- FUNCTIONS ----
-
 def render_question():
     idx = state["index"]
     total = len(questions)
@@ -264,21 +280,18 @@ def next_step(response):
         )
 
 def run_final_analysis():
-    # Step 1: Immediately show loading spinner message
     yield (
-        gr.update(value="🧠 AI is analyzing your responses... Please wait. ⏳"),   # question_display
-        gr.update(value=""),                                                      # progress_bar
-        gr.update(visible=False),                                                  # yes_btn
-        gr.update(visible=False),                                                  # no_btn
-        gr.update(visible=False),                                                  # back_btn
-        gr.update(visible=False),                                                  # result_btn
-        gr.update(visible=False)                                                   # result_box (temporarily hidden)
+        gr.update(value="🧠 AI is analyzing your responses... Please wait. ⏳"),
+        gr.update(value=""),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(visible=False)
     )
 
-    # Optional: small delay to show "loading" clearly
     time.sleep(2)
 
-    # Step 2: Prepare prompt and generate AI output
     llm_input = format_responses(questions, state["responses"])
     prompt = (
         "You are a clinical assistant AI evaluating a user's mental health screening responses.\n"
@@ -287,23 +300,19 @@ def run_final_analysis():
         "Respond in clear English and Hindi.\n\n"
         f"{llm_input}"
     )
-    
-    output = llm(prompt, max_new_tokens=500, temperature=0.3, do_sample=True)
-
+    output = llm(prompt, max_new_tokens=500, temperature=0.3)
     ai_result = output[0]["generated_text"]
 
-    # Step 3: Display final AI-generated analysis
     yield (
-        gr.update(value="✅ AI Analysis Completed."),      # question_display
-        gr.update(value=""),                               # progress_bar
-        gr.update(visible=False),                          # yes_btn
-        gr.update(visible=False),                          # no_btn
-        gr.update(visible=False),                          # back_btn
-        gr.update(visible=False),                          # result_btn
-        gr.update(value=ai_result, visible=True)            # result_box
+        gr.update(value="✅ AI Analysis Completed."),
+        gr.update(value=""),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(visible=False),
+        gr.update(value=ai_result, visible=True)
     )
 
-
 def go_back():
     if state["index"] > 0:
         state["index"] -= 1
@@ -358,16 +367,13 @@ with gr.Blocks(theme=gr.themes.Soft()) as app:
         question_display, progress_bar, yes_btn, no_btn, back_btn,
         result_btn, result_box
     ])
-    
     result_btn.click(
-    run_final_analysis,
-    outputs=[
-        question_display, progress_bar, yes_btn, no_btn, back_btn,
-        result_btn, result_box
-    ],
-    concurrency_limit=1,
-    
-)
+        run_final_analysis,
+        outputs=[
+            question_display, progress_bar, yes_btn, no_btn, back_btn,
+            result_btn, result_box
+        ],
+        concurrency_limit=1
+    )
 
 app.launch()
-