Spaces:
Runtime error
Runtime error
espiusedwards
commited on
Commit
•
db15209
1
Parent(s):
d9e4d0d
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,129 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from gradio.components import Textbox, Checkbox
|
3 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, T5ForConditionalGeneration
|
4 |
+
from peft import PeftModel, PeftConfig
|
5 |
+
import torch
|
6 |
+
import datasets
|
7 |
+
|
8 |
+
# Load your fine-tuned model and tokenizer
|
9 |
+
model_name = "google/flan-t5-large"
|
10 |
+
peft_name = "legacy107/flan-t5-large-ia3-cpgQA"
|
11 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
12 |
+
pretrained_model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
|
13 |
+
model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large")
|
14 |
+
model = PeftModel.from_pretrained(model, peft_name)
|
15 |
+
|
16 |
+
peft_name = "legacy107/flan-t5-large-ia3-bioasq-paraphrase"
|
17 |
+
peft_config = PeftConfig.from_pretrained(peft_name)
|
18 |
+
paraphrase_model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
|
19 |
+
paraphrase_model = PeftModel.from_pretrained(paraphrase_model, peft_name)
|
20 |
+
|
21 |
+
max_length = 512
|
22 |
+
max_target_length = 200
|
23 |
+
|
24 |
+
# Load your dataset
|
25 |
+
dataset = datasets.load_dataset("minh21/cpgQA-v1.0-unique-context-test-10-percent-validation-10-percent", split="test")
|
26 |
+
dataset = dataset.shuffle()
|
27 |
+
dataset = dataset.select(range(10))
|
28 |
+
|
29 |
+
|
30 |
+
def paraphrase_answer(question, answer, use_pretrained=False):
|
31 |
+
# Combine question and context
|
32 |
+
input_text = f"question: {question}. Paraphrase the answer to make it more natural answer: {answer}"
|
33 |
+
|
34 |
+
# Tokenize the input text
|
35 |
+
input_ids = tokenizer(
|
36 |
+
input_text,
|
37 |
+
return_tensors="pt",
|
38 |
+
padding="max_length",
|
39 |
+
truncation=True,
|
40 |
+
max_length=max_length,
|
41 |
+
).input_ids
|
42 |
+
|
43 |
+
# Generate the answer
|
44 |
+
with torch.no_grad():
|
45 |
+
if use_pretrained:
|
46 |
+
generated_ids = pretrained_model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
|
47 |
+
else:
|
48 |
+
generated_ids = paraphrase_model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
|
49 |
+
|
50 |
+
# Decode and return the generated answer
|
51 |
+
paraphrased_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
|
52 |
+
|
53 |
+
return paraphrased_answer
|
54 |
+
|
55 |
+
|
56 |
+
# Define your function to generate answers
|
57 |
+
def generate_answer(question, context, ground_truth, do_pretrained, do_natural, do_pretrained_natural):
|
58 |
+
# Combine question and context
|
59 |
+
input_text = f"question: {question} context: {context}"
|
60 |
+
|
61 |
+
# Tokenize the input text
|
62 |
+
input_ids = tokenizer(
|
63 |
+
input_text,
|
64 |
+
return_tensors="pt",
|
65 |
+
padding="max_length",
|
66 |
+
truncation=True,
|
67 |
+
max_length=max_length,
|
68 |
+
).input_ids
|
69 |
+
|
70 |
+
# Generate the answer
|
71 |
+
with torch.no_grad():
|
72 |
+
generated_ids = model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
|
73 |
+
|
74 |
+
# Decode and return the generated answer
|
75 |
+
generated_answer = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
|
76 |
+
|
77 |
+
# Paraphrase answer
|
78 |
+
paraphrased_answer = ""
|
79 |
+
if do_natural:
|
80 |
+
paraphrased_answer = paraphrase_answer(question, generated_answer)
|
81 |
+
|
82 |
+
# Get pretrained model's answer
|
83 |
+
pretrained_answer = ""
|
84 |
+
if do_pretrained:
|
85 |
+
with torch.no_grad():
|
86 |
+
pretrained_generated_ids = pretrained_model.generate(input_ids=input_ids, max_new_tokens=max_target_length)
|
87 |
+
pretrained_answer = tokenizer.decode(pretrained_generated_ids[0], skip_special_tokens=True)
|
88 |
+
|
89 |
+
# Get pretrained model's natural answer
|
90 |
+
pretrained_paraphrased_answer = ""
|
91 |
+
if do_pretrained_natural:
|
92 |
+
pretrained_paraphrased_answer = paraphrase_answer(question, generated_answer, True)
|
93 |
+
|
94 |
+
return generated_answer, paraphrased_answer, pretrained_answer, pretrained_paraphrased_answer
|
95 |
+
|
96 |
+
|
97 |
+
# Define a function to list examples from the dataset
|
98 |
+
def list_examples():
|
99 |
+
examples = []
|
100 |
+
for example in dataset:
|
101 |
+
context = example["context"]
|
102 |
+
question = example["question"]
|
103 |
+
answer = example["answer_text"]
|
104 |
+
examples.append([question, context, answer, True, True, True])
|
105 |
+
return examples
|
106 |
+
|
107 |
+
|
108 |
+
# Create a Gradio interface
|
109 |
+
iface = gr.Interface(
|
110 |
+
fn=generate_answer,
|
111 |
+
inputs=[
|
112 |
+
Textbox(label="Question"),
|
113 |
+
Textbox(label="Context"),
|
114 |
+
Textbox(label="Ground truth"),
|
115 |
+
Checkbox(label="Include pretrained model's answer"),
|
116 |
+
Checkbox(label="Include natural answer"),
|
117 |
+
Checkbox(label="Include pretrained model's natural answer")
|
118 |
+
],
|
119 |
+
outputs=[
|
120 |
+
Textbox(label="Generated Answer"),
|
121 |
+
Textbox(label="Natural Answer"),
|
122 |
+
Textbox(label="Pretrained Model's Answer"),
|
123 |
+
Textbox(label="Pretrained Model's Natural Answer")
|
124 |
+
],
|
125 |
+
examples=list_examples()
|
126 |
+
)
|
127 |
+
|
128 |
+
# Launch the Gradio interface
|
129 |
+
iface.launch()
|