Spaces:
Runtime error
Runtime error
NorHsangPha
commited on
Commit
β’
77e720d
1
Parent(s):
cb45693
Initial: initial commit
Browse files- app.py +30 -0
- gpt2.py +108 -0
- requirements.txt +4 -0
app.py
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from gpt2 import generate_text, GENERATE_EXAMPLES
|
3 |
+
|
4 |
+
gpt_generate = gr.Interface(
|
5 |
+
fn=generate_text,
|
6 |
+
inputs=[
|
7 |
+
gr.Textbox(label="Input text"),
|
8 |
+
gr.Dropdown(
|
9 |
+
[
|
10 |
+
"sample_outputs",
|
11 |
+
"greedy_search",
|
12 |
+
"beem_search",
|
13 |
+
"top_k_search",
|
14 |
+
"top_p_search",
|
15 |
+
],
|
16 |
+
label="Search method",
|
17 |
+
value="sample_outputs",
|
18 |
+
),
|
19 |
+
],
|
20 |
+
outputs=gr.Textbox(label="Generated text"),
|
21 |
+
examples=GENERATE_EXAMPLES,
|
22 |
+
title="GPT-2 Text generator Demo",
|
23 |
+
description="Generate text using GPT-2.",
|
24 |
+
allow_flagging="never",
|
25 |
+
)
|
26 |
+
|
27 |
+
with gr.Blocks() as demo:
|
28 |
+
gpt_generate.render()
|
29 |
+
|
30 |
+
demo.launch()
|
gpt2.py
ADDED
@@ -0,0 +1,108 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
3 |
+
import torch
|
4 |
+
|
5 |
+
|
6 |
+
if torch.cuda.is_available():
|
7 |
+
device = torch.device("cuda")
|
8 |
+
elif (
|
9 |
+
hasattr(torch.backends, "mps")
|
10 |
+
and torch.backends.mps.is_available()
|
11 |
+
and torch.backends.mps.is_built()
|
12 |
+
):
|
13 |
+
device = torch.device("mps")
|
14 |
+
else:
|
15 |
+
device = torch.device("cpu")
|
16 |
+
|
17 |
+
print(f"running device: {device}")
|
18 |
+
auth_token = os.environ.get("TOKEN_READ_SECRET") or True
|
19 |
+
|
20 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
21 |
+
"NorHsangPha/shan_gpt2_news", token=auth_token
|
22 |
+
)
|
23 |
+
model = AutoModelForCausalLM.from_pretrained(
|
24 |
+
"NorHsangPha/shan_gpt2_news", pad_token_id=tokenizer.eos_token_id, token=auth_token
|
25 |
+
).to(device)
|
26 |
+
|
27 |
+
|
28 |
+
def greedy_search(model_inputs, max_new_tokens):
|
29 |
+
greedy_output = model.generate(**model_inputs, max_new_tokens=max_new_tokens)
|
30 |
+
|
31 |
+
return tokenizer.decode(greedy_output[0], skip_special_tokens=True)
|
32 |
+
|
33 |
+
|
34 |
+
def beem_search(model_inputs, max_new_tokens):
|
35 |
+
beam_output = model.generate(
|
36 |
+
**model_inputs,
|
37 |
+
max_new_tokens=max_new_tokens,
|
38 |
+
num_beams=5,
|
39 |
+
no_repeat_ngram_size=2, #
|
40 |
+
num_return_sequences=5, #
|
41 |
+
early_stopping=True,
|
42 |
+
)
|
43 |
+
|
44 |
+
return tokenizer.decode(beam_output[0], skip_special_tokens=True)
|
45 |
+
|
46 |
+
|
47 |
+
def sample_outputs(model_inputs, max_new_tokens):
|
48 |
+
sample_output = model.generate(
|
49 |
+
**model_inputs,
|
50 |
+
max_new_tokens=max_new_tokens,
|
51 |
+
do_sample=True,
|
52 |
+
top_k=0,
|
53 |
+
temperature=0.6,
|
54 |
+
)
|
55 |
+
|
56 |
+
return tokenizer.decode(sample_output[0], skip_special_tokens=True)
|
57 |
+
|
58 |
+
|
59 |
+
def top_k_search(model_inputs, max_new_tokens):
|
60 |
+
top_k_output = model.generate(
|
61 |
+
**model_inputs, max_new_tokens=max_new_tokens, do_sample=True, top_k=50
|
62 |
+
)
|
63 |
+
|
64 |
+
return tokenizer.decode(top_k_output[0], skip_special_tokens=True)
|
65 |
+
|
66 |
+
|
67 |
+
def top_p_search(model_inputs, max_new_tokens):
|
68 |
+
top_p_output = model.generate(
|
69 |
+
**model_inputs,
|
70 |
+
max_new_tokens=max_new_tokens,
|
71 |
+
do_sample=True,
|
72 |
+
top_p=0.92,
|
73 |
+
top_k=0,
|
74 |
+
)
|
75 |
+
|
76 |
+
return tokenizer.decode(top_p_output[0], skip_special_tokens=True)
|
77 |
+
|
78 |
+
|
79 |
+
def generate_text(input_text, search_method="sample_outputs"):
|
80 |
+
model_inputs = tokenizer(input_text, return_tensors="pt").to(device)
|
81 |
+
max_new_tokens = 120
|
82 |
+
|
83 |
+
match search_method:
|
84 |
+
case "greedy_search":
|
85 |
+
text = greedy_search(model_inputs, max_new_tokens)
|
86 |
+
|
87 |
+
case "beem_search":
|
88 |
+
text = beem_search(model_inputs, max_new_tokens)
|
89 |
+
|
90 |
+
case "top_k_search":
|
91 |
+
text = top_k_search(model_inputs, max_new_tokens)
|
92 |
+
|
93 |
+
case "top_p_search":
|
94 |
+
text = top_p_search(model_inputs, max_new_tokens)
|
95 |
+
|
96 |
+
case _:
|
97 |
+
text = sample_outputs(model_inputs, max_new_tokens)
|
98 |
+
|
99 |
+
return text
|
100 |
+
|
101 |
+
|
102 |
+
GENERATE_EXAMPLES = [
|
103 |
+
["αααΊααα―ααΊαΆαα", "sample_outputs"],
|
104 |
+
["αα’ααΊααα―α΅αΊαΈααα―α΅αΊαΈααα°ααΊ", "greedy_search"],
|
105 |
+
["αα’ααΊααα―α΅αΊαΈααα―α΅αΊαΈααα°ααΊ", "top_k_search"],
|
106 |
+
["αα’ααΊααα―α΅αΊαΈααα―α΅αΊαΈααα°ααΊ", "top_p_search"],
|
107 |
+
["αα’ααΊααα―α΅αΊαΈααα―α΅αΊαΈααα°ααΊ", "beem_search"],
|
108 |
+
]
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio
|
2 |
+
transformers
|
3 |
+
torch
|
4 |
+
torchaudio
|