-
Notifications
You must be signed in to change notification settings - Fork 0
/
app.py
191 lines (158 loc) · 4.89 KB
/
app.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
import argparse
import gradio as gr
import torch
from src.models.GPT2 import model_getter
from src.utils.generation_utils import TextGenerator
def parse():
parser = argparse.ArgumentParser(description="Gradio Inference App")
parser.add_argument("--model-size", default="medium", type=str)
parser.add_argument("--share", default=False, action="store_true")
parser.add_argument("--bit-quantize", default=False, action="store_true")
args = parser.parse_args()
return args
DEVICE = "cpu"
if torch.cuda.is_available():
DEVICE = "cuda"
BNB_FLAG = False
try:
import bitsandbytes as bnb
BNB_FLAG = True
except Exception as e:
pass
generator = TextGenerator(seq_len=512, tokenizer=None)
def model_creator(size: str) -> torch.nn.Module:
save_paths = {
"base*": "checkpoints/127_weights.pth.tar",
"medium*": "checkpoints/303_weights.pth.tar",
"XL*": "checkpoints/1B8bit_weights.pth.tar"
if BNB_FLAG
else "checkpoints/1B_weights_noBNB.pth.tar",
"medium": "checkpoints/354_weights.pth.tar",
}
if "*" in size:
model = model_getter(
size,
vocab_size=50257,
num_ctx=512,
model_checkpoint=save_paths[size],
**{
"fused_residuals": True,
"num_head": 8,
"use_alibi": True,
"quantized_state": True if "XL" in size else False,
},
)
elif size == "medium":
model = model_getter(
"medium",
vocab_size=50257,
num_ctx=1024,
model_checkpoint=save_paths[size],
**{"fused_residuals": False, "use_alibi": False},
)
model.to(DEVICE)
model.eval()
return model
def generate_text(
prompt,
steps,
temperature,
top_k,
top_p,
tau,
repetition_penalty,
epsilon,
sampling_choice,
):
if sampling_choice == "Top-k":
sampling_method = "topk"
elif sampling_choice == "Nucleus":
sampling_method = "nucleus"
elif sampling_choice == "Typical":
sampling_method = "typical"
elif sampling_choice == "Greedy":
sampling_method = "greedy"
elif sampling_choice == "$\eta$":
sampling_method = "eta"
generated_text, new_gen, logprobs = generator.generate_text_from_prompt(
model=model,
prompt=prompt,
steps=int(steps),
temperature=temperature,
top_k=top_k,
top_p=top_p,
tau=tau,
repetition_penalty=repetition_penalty,
epsilon=epsilon,
sampling_method=sampling_method,
device=DEVICE,
)
original_gen_length = len(generated_text) - len(new_gen)
return [
(generated_text[:original_gen_length], None),
(generated_text[original_gen_length:], "Generated Text"),
]
if __name__ == "__main__":
args = parse()
assert args.model_size in ["base*", "medium*", "medium", "XL*"]
model = model_creator(args.model_size)
from src.utils.gradio_utils import DESCRIPTION_MAP
description = DESCRIPTION_MAP[args.model_size]
iface = gr.Interface(
fn=generate_text,
inputs=[
gr.inputs.Textbox(lines=10, label="Enter your text here"),
gr.inputs.Slider(
0, 1000, default=100, label="Number of tokens to generate"
),
gr.inputs.Slider(0, 2, default=0.70, label="Temperature"),
gr.inputs.Slider(
0,
50,
default=40,
label="k (Top-k Sampling)",
),
gr.inputs.Slider(
0,
1,
default=0.96,
label="p (Nucleus Sampling)",
),
gr.inputs.Slider(
0,
1,
default=0.2,
label="Tau (Typical Sampling)",
),
gr.inputs.Slider(
0.0,
1.3,
default=1.2,
label="Repetition Penalty",
),
gr.inputs.Slider(
0.0,
0.001,
default=0.0006,
label="$\epsilon$",
),
gr.inputs.Radio(
choices=["Top-k", "Nucleus", "Typical", "Greedy", "$\eta$"],
label="Sampling Method",
default="Nucleus",
),
],
outputs=gr.HighlightedText(
label="Generated Text",
combine_adjacent=True,
color_map=["Generated Text", "blue"],
),
live=False,
title="GPT-* 🤖"
if args.model_size in ["base*", "medium*", "XL*"]
else "GPT-354M 🤖",
description=description,
article="For more details check out the model repo [here](https://github.com/fattorib/Little-GPT)",
allow_flagging="never",
)
iface.launch(share=args.share)