Update app.py

app.py

@@ -1,5 +1,5 @@
 import gradio as gr
-from gradio.components import Textbox
+from gradio.components import Textbox, Checkbox
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, T5ForConditionalGeneration
 from peft import PeftModel, PeftConfig
 import torch
@@ -9,8 +9,8 @@ import datasets
 model_name = "google/flan-t5-large"
 peft_name = "legacy107/flan-t5-large-ia3-cpgQA"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
-model = PeftModel.from_pretrained(model, peft_name)
+pretrained_model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
+model = PeftModel.from_pretrained(pretrained_model, peft_name)
 
 peft_name = "legacy107/flan-t5-large-ia3-bioasq-paraphrase"
 peft_config = PeftConfig.from_pretrained(peft_name)
@@ -23,7 +23,7 @@ max_target_length = 200
 # Load your dataset
 dataset = datasets.load_dataset("minh21/cpgQA-v1.0-unique-context-test-10-percent-validation-10-percent", split="test")
 dataset = dataset.shuffle()
-dataset = dataset.select(range(5))
+dataset = dataset.select(range(10))
 
 
 def paraphrase_answer(question, answer):
@@ -50,7 +50,7 @@ def paraphrase_answer(question, answer):
 
 
 # Define your function to generate answers
-def generate_answer(question, context, ground_truth):
+def generate_answer(question, context, ground_truth, do_pretrained, do_natural):
     # Combine question and context
     input_text = f"question: {question} context: {context}"
 
@@ -71,9 +71,18 @@ def generate_answer(question, context, ground_truth):
     generated_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
 
     # Paraphrase answer
-    paraphrased_answer = paraphrase_answer(question, generated_answer)
+    paraphrase_answer = ""
+    if do_natural:
+        paraphrased_answer = paraphrase_answer(question, generated_answer)
 
-    return generated_answer, paraphrased_answer
+    # Get pretrained model's answer
+    pretrained_answer = ""
+    if do_pretrained:
+        with torch.no_grad():
+            generated_ids = pretrained_model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
+            pretrained_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+
+    return generated_answer, paraphrased_answer, pretrained_answer
 
 
 # Define a function to list examples from the dataset
@@ -83,7 +92,7 @@ def list_examples():
         context = example["context"]
         question = example["question"]
         answer = example["answer_text"]
-        examples.append([question, context, answer])
+        examples.append([question, context, answer, True, True])
     return examples
 
 
@@ -93,11 +102,14 @@ iface = gr.Interface(
     inputs=[
         Textbox(label="Question"),
         Textbox(label="Context"),
-        Textbox(label="Ground truth")
+        Textbox(label="Ground truth"),
+        Checkbox(label="Include pretrained model's result"),
+        Checkbox(label="Include natural answer")
     ],
     outputs=[
         Textbox(label="Generated Answer"),
-        Textbox(label="Natural Answer")
+        Textbox(label="Natural Answer"),
+        Textbox(label="Pretrained Model's Answer"),
     ],
     examples=list_examples()
 )