Update app.py

app.py

@@ -1,41 +1,74 @@
 import gradio as gr
-from transformers import pipeline
 import pandas as pd
 import io
 
-# Load once
-classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
+from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification, AutoModelForCausalLM
+import torch
 
+# Model lists
+zero_shot_models = {
+    "EN: deberta-v3-large-zeroshot": "MoritzLaurer/deberta-v3-large-zeroshot-v2.0",
+    "MULTI: mDeBERTa-v3-base-xnli": "MoritzLaurer/mDeBERTa-v3-base-xnli-multilingual-nli-2mil7",
+    "MULTI: xlm-roberta-large-xnli": "joeddav/xlm-roberta-large-xnli"
+}
+
+text_gen_models = {
+    "Mixtral-8x7B-Instruct": "mistralai/Mixtral-8x7B-Instruct-v0.1",
+    "DeepSeek-Qwen3-8B": "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B",
+    "DeepSeek-52B": "deepseek-ai/DeepSeek-R1-0528",
+    "LLaMA-3.1-8B-Instruct": "meta-llama/Llama-3.1-8B-Instruct"
+}
+
+# Shared storage for results
 response_table = []
 
-def classify_items(questions_text, labels_text):
-    questions = [q.strip() for q in questions_text.strip().split("\n") if q.strip()]
-    labels = [l.strip() for l in labels_text.strip().split(",") if l.strip()]
-    
+def run_classification(questions_text, labels_text, model_name):
+    labels = [label.strip() for label in labels_text.split(",") if label.strip()]
+    questions = [q.strip() for q in questions_text.split("\n") if q.strip()]
     if not labels or not questions:
-        return "Please provide both items and at least two response options.", ""
+        return "Please enter both items and response options.", None
 
+    classifier = pipeline("zero-shot-classification", model=model_name)
     global response_table
     response_table = []
     output_lines = []
 
     for i, question in enumerate(questions, 1):
         result = classifier(question, labels, multi_label=False)
-        probs = dict(zip(result['labels'], result['scores']))
-
+        row = {"Item #": i, "Item": question}
         output_lines.append(f"{i}. {question}")
-        for label in labels:
-            output_lines.append(f"→ {label}: {round(probs.get(label, 0.0), 3)}")
+        for label, score in zip(result["labels"], result["scores"]):
+            row[label] = round(score, 3)
+            output_lines.append(f"→ {label}: {round(score, 3)}")
         output_lines.append("")
-
-        row = {"Item #": i, "Item": question}
-        row.update({label: round(probs.get(label, 0.0), 3) for label in labels})
         response_table.append(row)
 
     return "\n".join(output_lines), None
 
-def download_csv():
+def run_generation(questions_text, model_name):
+    questions = [q.strip() for q in questions_text.split("\n") if q.strip()]
+    if not questions:
+        return "Please enter at least one item.", None
+
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32)
+    model.eval()
+
     global response_table
+    response_table = []
+    output_lines = []
+
+    for i, question in enumerate(questions, 1):
+        prompt = f"Please respond to the following item as if you were a survey participant:\n\"{question}\""
+        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+        outputs = model.generate(**inputs, max_new_tokens=60)
+        response = tokenizer.decode(outputs[0], skip_special_tokens=True).split("\n")[-1]
+        output_lines.append(f"{i}. {question}\n→ {response.strip()}\n")
+        response_table.append({"Item #": i, "Item": question, "Response": response.strip()})
+
+    return "\n".join(output_lines), None
+
+def download_csv():
     if not response_table:
         return None
     df = pd.DataFrame(response_table)
@@ -43,22 +76,37 @@ def download_csv():
     df.to_csv(csv_buffer, index=False)
     return csv_buffer.getvalue()
 
-# Gradio UI
 with gr.Blocks() as demo:
-    gr.Markdown("# 🧠 Zero-Shot Classification for Questionnaire Responses")
-    gr.Markdown("Paste questionnaire items (one per line), and provide your own response labels (comma-separated).")
+    gr.Markdown("# 🧠 LLM Questionnaire Response Tool")
+    gr.Markdown("Choose between **zero-shot classification** and **text generation**. Enter your questionnaire items, select a model, and view responses. You can download all results as a CSV.")
+
+    task_type = gr.Radio(["Zero-shot classification", "Text generation"], label="Task Type", value="Zero-shot classification")
 
     with gr.Row():
-        with gr.Column():
-            questions_input = gr.Textbox(label="Questionnaire Items", lines=10, placeholder="e.g.\nI feel in control of my life.\nI enjoy being around others.")
-            labels_input = gr.Textbox(label="Response Options (comma-separated)", placeholder="Strongly disagree, Disagree, Neutral, Agree, Strongly agree")
-            submit_btn = gr.Button("Classify Items")
-            csv_btn = gr.Button("📥 Download CSV")
-        with gr.Column():
-            output_box = gr.Textbox(label="Classification Output", lines=20)
-            file_output = gr.File(label="Download CSV", visible=False)
-
-    submit_btn.click(fn=classify_items, inputs=[questions_input, labels_input], outputs=[output_box, file_output])
-    csv_btn.click(fn=download_csv, inputs=[], outputs=file_output)
+        model_selector = gr.Dropdown(label="Choose Model")
+        labels_input = gr.Textbox(label="Response Options (comma-separated)", visible=True)
+
+    questions_input = gr.Textbox(label="Questionnaire Items (one per line)", lines=10)
+    output_box = gr.Textbox(label="Model Output", lines=20)
+    submit_btn = gr.Button("Run")
+    download_btn = gr.Button("📥 Download CSV")
+    file_output = gr.File(label="CSV", visible=False)
+
+    def update_model_and_labels(task):
+        if task == "Zero-shot classification":
+            return gr.Dropdown.update(choices=list(zero_shot_models.keys()), value=list(zero_shot_models.keys())[0]), gr.Textbox.update(visible=True)
+        else:
+            return gr.Dropdown.update(choices=list(text_gen_models.keys()), value=list(text_gen_models.keys())[0]), gr.Textbox.update(visible=False)
+
+    task_type.change(fn=update_model_and_labels, inputs=task_type, outputs=[model_selector, labels_input])
+
+    def route_task(questions, labels, model_ui_name, task):
+        if task == "Zero-shot classification":
+            return run_classification(questions, labels, zero_shot_models[model_ui_name])
+        else:
+            return run_generation(questions, text_gen_models[model_ui_name])
+
+    submit_btn.click(fn=route_task, inputs=[questions_input, labels_input, model_selector, task_type], outputs=[output_box, file_output])
+    download_btn.click(fn=download_csv, inputs=[], outputs=file_output)
 
 demo.launch()