Create app.py

app.py

@@ -0,0 +1,104 @@
+import gradio as gr
+from gradio.components import Textbox
+from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, T5ForConditionalGeneration
+from peft import PeftModel, PeftConfig
+import torch
+import datasets
+
+# Load your fine-tuned model and tokenizer
+model_name = "google/flan-t5-large"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
+model.load_adapter("legacy107/adapter-flan-t5-large-bottleneck-adapter-cpgQA", source="hf")
+model.set_active_adapters("question_answering")
+
+peft_name = "legacy107/flan-t5-large-ia3-bioasq-paraphrase"
+peft_config = PeftConfig.from_pretrained(peft_name)
+paraphrase_model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+paraphrase_model = PeftModel.from_pretrained(paraphrase_model, peft_name)
+
+max_length = 512
+max_target_length = 128
+
+# 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))
+
+
+def paraphrase_answer(question, answer):
+    # Combine question and context
+    input_text = f"question: {question}. Paraphrase the answer to make it more natural answer: {answer}"
+
+    # Tokenize the input text
+    input_ids = tokenizer(
+        input_text,
+        return_tensors="pt",
+        padding="max_length",
+        truncation=True,
+        max_length=max_length,
+    ).input_ids
+
+    # Generate the answer
+    with torch.no_grad():
+        generated_ids = paraphrase_model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
+
+    # Decode and return the generated answer
+    paraphrased_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+
+    return paraphrased_answer
+
+
+# Define your function to generate answers
+def generate_answer(question, context):
+    # Combine question and context
+    input_text = f"question: {question} context: {context}"
+
+    # Tokenize the input text
+    input_ids = tokenizer(
+        input_text,
+        return_tensors="pt",
+        padding="max_length",
+        truncation=True,
+        max_length=max_length,
+    ).input_ids
+
+    # Generate the answer
+    with torch.no_grad():
+        generated_ids = model.generate(input_ids, max_new_tokens=max_target_length)
+
+    # Decode and return the generated answer
+    generated_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
+
+    # Paraphrase answer
+    paraphrased_answer = paraphrase_answer(question, generated_answer)
+
+    return generated_answer, paraphrased_answer
+
+
+# Define a function to list examples from the dataset
+def list_examples():
+    examples = []
+    for example in dataset:
+        context = example["context"]
+        question = example["question"]
+        examples.append([question, context])
+    return examples
+
+
+# Create a Gradio interface
+iface = gr.Interface(
+    fn=generate_answer,
+    inputs=[
+        Textbox(label="Question"),
+        Textbox(label="Context")
+    ],
+    outputs=[
+        Textbox(label="Generated Answer"),
+        Textbox(label="Natural Answer")
+    ],
+    examples=list_examples()
+)
+
+# Launch the Gradio interface
+iface.launch()