Update app.py
Browse files
app.py
CHANGED
@@ -2,9 +2,6 @@ import torch
|
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import gradio as gr
|
4 |
|
5 |
-
# Check if GPU is available, otherwise use CPU
|
6 |
-
# device = "cuda" if torch.cuda.is_available() else "cpu"
|
7 |
-
|
8 |
# Load pre-trained GPT-2 model and tokenizer
|
9 |
model_name = "gpt2-large"
|
10 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
@@ -13,14 +10,12 @@ model = AutoModelForCausalLM.from_pretrained(model_name)
|
|
13 |
def generate_text(input_text, max_length=16, num_beams=5, do_sample=False, no_repeat_ngram_size=2):
|
14 |
"""
|
15 |
Generate text based on the given input text.
|
16 |
-
|
17 |
Parameters:
|
18 |
- input_text (str): The input text to start generation from.
|
19 |
- max_length (int): Maximum length of the generated text.
|
20 |
- num_beams (int): Number of beams for beam search.
|
21 |
- do_sample (bool): Whether to use sampling or not.
|
22 |
- no_repeat_ngram_size (int): Size of the n-gram to avoid repetition.
|
23 |
-
|
24 |
Returns:
|
25 |
- generated_text (str): The generated text.
|
26 |
"""
|
@@ -36,13 +31,11 @@ def generate_text(input_text, max_length=16, num_beams=5, do_sample=False, no_re
|
|
36 |
def generate_text_with_nucleus_search(input_text, max_length=16, do_sample=True, top_p=0.9):
|
37 |
"""
|
38 |
Generate text with nucleus sampling based on the given input text.
|
39 |
-
|
40 |
Parameters:
|
41 |
- input_text (str): The input text to start generation from.
|
42 |
- max_length (int): Maximum length of the generated text.
|
43 |
- do_sample (bool): Whether to use sampling or not.
|
44 |
- top_p (float): Nucleus sampling parameter.
|
45 |
-
|
46 |
Returns:
|
47 |
- generated_text (str): The generated text.
|
48 |
"""
|
@@ -54,17 +47,25 @@ def generate_text_with_nucleus_search(input_text, max_length=16, do_sample=True,
|
|
54 |
generated_text = tokenizer.decode(output[0])
|
55 |
return generated_text
|
56 |
|
57 |
-
# Create Gradio
|
58 |
input_text_interface = gr.Textbox(lines=5, label="Input Text", placeholder="Enter text for generation...")
|
59 |
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
|
|
66 |
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
allow_flagging="never").launch(share=True)
|
|
|
2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
import gradio as gr
|
4 |
|
|
|
|
|
|
|
5 |
# Load pre-trained GPT-2 model and tokenizer
|
6 |
model_name = "gpt2-large"
|
7 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
|
10 |
def generate_text(input_text, max_length=16, num_beams=5, do_sample=False, no_repeat_ngram_size=2):
|
11 |
"""
|
12 |
Generate text based on the given input text.
|
|
|
13 |
Parameters:
|
14 |
- input_text (str): The input text to start generation from.
|
15 |
- max_length (int): Maximum length of the generated text.
|
16 |
- num_beams (int): Number of beams for beam search.
|
17 |
- do_sample (bool): Whether to use sampling or not.
|
18 |
- no_repeat_ngram_size (int): Size of the n-gram to avoid repetition.
|
|
|
19 |
Returns:
|
20 |
- generated_text (str): The generated text.
|
21 |
"""
|
|
|
31 |
def generate_text_with_nucleus_search(input_text, max_length=16, do_sample=True, top_p=0.9):
|
32 |
"""
|
33 |
Generate text with nucleus sampling based on the given input text.
|
|
|
34 |
Parameters:
|
35 |
- input_text (str): The input text to start generation from.
|
36 |
- max_length (int): Maximum length of the generated text.
|
37 |
- do_sample (bool): Whether to use sampling or not.
|
38 |
- top_p (float): Nucleus sampling parameter.
|
|
|
39 |
Returns:
|
40 |
- generated_text (str): The generated text.
|
41 |
"""
|
|
|
47 |
generated_text = tokenizer.decode(output[0])
|
48 |
return generated_text
|
49 |
|
50 |
+
# Create Gradio input interface
|
51 |
input_text_interface = gr.Textbox(lines=5, label="Input Text", placeholder="Enter text for generation...")
|
52 |
|
53 |
+
# Create Gradio output interfaces
|
54 |
+
output_text_interface1 = gr.Textbox(label="Generated Text (Regular)", placeholder="Generated text will appear here...")
|
55 |
+
output_text_interface2 = gr.Textbox(label="Generated Text (Nucleus Sampling)", placeholder="Generated text will appear here...")
|
56 |
+
|
57 |
+
# Interface for regular text generation
|
58 |
+
interface1 = gr.Interface(generate_text, input_text_interface, output_text_interface1,
|
59 |
+
title="Text Generation with GPT-2",
|
60 |
+
description="Generate text using the GPT-2 model with regular generation method.",
|
61 |
+
allow_flagging="never")
|
62 |
|
63 |
+
# Interface for text generation with nucleus sampling
|
64 |
+
interface2 = gr.Interface(generate_text_with_nucleus_search, input_text_interface, output_text_interface2,
|
65 |
+
title="Text Generation with Nucleus Sampling",
|
66 |
+
description="Generate text using nucleus sampling with the GPT-2 model.",
|
67 |
+
allow_flagging="never")
|
68 |
|
69 |
+
# Launch both interfaces
|
70 |
+
interface1.launch(share=True)
|
71 |
+
interface2.launch(share=True)
|
|