Hugging Face's logo

Spaces:
isek-ai
/
LightNovel-Intro-RetNet-400M-Demo
Runtime error

App Files Community
LightNovel-Intro-RetNet-400M-Demo
File size: 3,556 Bytes 
617aba5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
import torch

from RetNet.retnet.modeling_retnet import RetNetForCausalLM
from transformers import AutoTokenizer

import gradio as gr

MODEL_NAME = "p1atdev/LightNovel-Intro-RetNet-400M"

DEFAULT_INPUT_TEXT = "目が覚めると、"

EXAMPLE_INPUTS = [
    DEFAULT_INPUT_TEXT,
    "冒険者ギルドには",
    "真っ白い部屋の中、そこには",
    "20XX年、",
    "「なんだって!?」",
    "どうやらトラックにはねられ、俺は",
]

tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = RetNetForCausalLM.from_pretrained(MODEL_NAME)
model.eval()


@torch.no_grad()
def generate(
    input_text,
    max_new_tokens=128,
    do_sample=True,
    temperature=1.0,
    top_p=0.95,
    top_k=20,
):
    if input_text.strip() == "":
        return ""

    inputs = tokenizer(input_text, return_tensors="pt", add_special_tokens=False)
    generated = model.custom_generate(
        **inputs,
        parallel_compute_prompt=True,
        max_new_tokens=max_new_tokens,
        do_sample=do_sample,
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
    )
    return tokenizer.batch_decode(generated)[0]


def continue_generate(
    input_text,
    *args,
):
    return input_text, generate(input_text, *args)


with gr.Blocks() as demo:
    gr.Markdown(
        """\
# LightNovel-Intro-RetNet-400M-Demo

ライトノベルの冒頭だけを学習した 400M パラメータの RetNet モデルのデモです。

### 参考:

- https://github.com/syncdoth/RetNet
"""
    )

    input_text = gr.Textbox(
        value=DEFAULT_INPUT_TEXT,
        placeholder="私の名前は...",
        lines=2,
    )
    output_text = gr.Textbox(
        value="",
        placeholder="ここに出力が表示されます...",
        lines=8,
        interactive=False,
    )

    with gr.Row():
        generate_btn = gr.Button("Generate ✒️", variant="primary")
        continue_btn = gr.Button("Continue ➡️", variant="secondary")
        clear_btn = gr.ClearButton(
            value="Clear 🧹",
            components=[input_text, output_text],
        )

    with gr.Accordion("Advanced settings", open=False):
        max_tokens = gr.Slider(
            label="Max tokens",
            minimum=8,
            maximum=512,
            value=128,
            step=4,
        )
        do_sample = gr.Checkbox(
            label="Do sample",
            value=True,
        )
        temperature = gr.Slider(
            label="Temperature",
            minimum=0,
            maximum=2,
            value=1,
            step=0.05,
        )
        top_p = gr.Slider(
            label="Top p",
            minimum=0,
            maximum=1,
            value=0.95,
            step=0.05,
        )
        top_k = gr.Slider(
            label="Top k",
            minimum=0,
            maximum=100,
            value=20,
            step=1,
        )

    gr.Examples(
        examples=EXAMPLE_INPUTS,
        inputs=input_text,
    )

    generate_btn.click(
        fn=generate,
        inputs=[
            input_text,
            max_tokens,
            do_sample,
            temperature,
            top_p,
            top_k,
        ],
        outputs=output_text,
        queue=False,
    )
    continue_btn.click(
        fn=continue_generate,
        inputs=[
            input_text,
            max_tokens,
            do_sample,
            temperature,
            top_p,
            top_k,
        ],
        outputs=[input_text, output_text],
        queue=False,
    )

demo.launch()