File size: 5,343 Bytes
3cf1901
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
282b88e
3cf1901
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
os.system("pip install transformers")
os.system("pip3 install torch==1.10.1+cpu torchvision==0.11.2+cpu torchaudio==0.10.1+cpu -f "
          "https://download.pytorch.org/whl/cpu/torch_stable.html")
os.system("pip install mtranslate")
os.system("pip install requests")
os.system("pip install random")

import transformers
import json
import random
import requests

from mtranslate import translate
import streamlit as st


MODELS = {
    "GPT-2 Model Recycled From English": {
        "url": "https://api-inference.huggingface.co/models/GroNLP/gpt2-small-dutch"
    },
}

PROMPT_LIST = {
    "Er was eens...": ["Er was eens..."],
    "Dag.": ["Hallo, mijn naam is "],
    "Te zijn of niet te zijn?": ["Naar mijn mening is 'zijn'"],
}


def query(payload, model_name):
    data = json.dumps(payload)
    print("model url:", MODELS[model_name]["url"])
    response = requests.request(
        "POST", MODELS[model_name]["url"], headers={}, data=data
    )
    return json.loads(response.content.decode("utf-8"))


def process(
    text: str, model_name: str, max_len: int, temp: float, top_k: int, top_p: float
):
    payload = {
        "inputs": text,
        "parameters": {
            "max_new_tokens": max_len,
            "top_k": top_k,
            "top_p": top_p,
            "temperature": temp,
            "repetition_penalty": 2.0,
        },
        "options": {
            "use_cache": True,
        },
    }
    return query(payload, model_name)


# Page
st.set_page_config(page_title="Dutch GPT-2 Demo")
st.title("Dutch GPT-2")


# Sidebar
st.sidebar.subheader("Configurable parameters")

max_len = st.sidebar.number_input(
    "Maximum length",
    value=100,
    help="The maximum length of the sequence to be generated.",
)

temp = st.sidebar.slider(
    "Temperature",
    value=1.0,
    min_value=0.1,
    max_value=100.0,
    help="The value used to module the next token probabilities.",
)

top_k = st.sidebar.number_input(
    "Top k",
    value=10,
    help="The number of highest probability vocabulary tokens to keep for top-k-filtering.",
)

top_p = st.sidebar.number_input(
    "Top p",
    value=0.95,
    help=" If set to float < 1, only the most probable tokens with probabilities that add up to top_p or higher are kept for generation.",
)

do_sample = st.sidebar.selectbox(
    "Sampling?",
    (True, False),
    help="Whether or not to use sampling; use greedy decoding otherwise.",
)


# Body
st.markdown(
    """
    Dutch GPT-2 model (small) is based on the English GPT-2 model:

    Researches [Wietse de Vries](https://www.semanticscholar.org/author/Wietse-de-Vries/144611157) and [M. Nissim](https://www.semanticscholar.org/author/M.-Nissim/2742475)
    obtained this model by transfering the English GPT-2 model in multiple procedure while exploiting genetic closeness between Dutch and English. 

    During this process, they retrained the lexical embeddings of the original English GPT-2 model and did additional fine-tuning of the full Dutch model
    for better text generation. 
    
    For more information on the model:
    
    [arXiv](https://arxiv.org/abs/2012.05628)
    [GitHub](https://github.com/wietsedv/gpt2-recycle)
    
    """
)

model_name = st.selectbox("Model", (list(MODELS.keys())))

ALL_PROMPTS = list(PROMPT_LIST.keys()) + ["Custom"]
prompt = st.selectbox("Prompt", ALL_PROMPTS, index=len(ALL_PROMPTS) - 1)
if prompt == "Custom":
    prompt_box = "Enter your text here"
else:
    prompt_box = random.choice(PROMPT_LIST[prompt])

text = st.text_area("Enter text", prompt_box)

if st.button("Run"):
    with st.spinner(text="Getting results..."):
        st.subheader("Result")
        print(f"maxlen:{max_len}, temp:{temp}, top_k:{top_k}, top_p:{top_p}")
        result = process(
            text=text,
            model_name=model_name,
            max_len=int(max_len),
            temp=temp,
            top_k=int(top_k),
            top_p=float(top_p),
        )
        print("result:", result)
        if "error" in result:
            if type(result["error"]) is str:
                st.write(f'{result["error"]}.', end=" ")
                if "estimated_time" in result:
                    st.write(
                        f'Please try again in about {result["estimated_time"]:.0f} seconds.'
                    )
            else:
                if type(result["error"]) is list:
                    for error in result["error"]:
                        st.write(f"{error}")
        else:
            result = result[0]["generated_text"]
            st.write(result.replace("\n", "  \n"))
            st.text("English translation")
            st.write(translate(result, "en", "nl").replace("\n", "  \n"))
 
st.subheader("References:") 
st.markdown(
    """
```    
@inproceedings{de-vries-nissim-2021-good,
    title = "As Good as New. How to Successfully Recycle {E}nglish {GPT}-2 to Make Models for Other Languages",
    author = "de Vries, Wietse  and
      Nissim, Malvina",
    booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
    month = aug,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-acl.74",
    doi = "10.18653/v1/2021.findings-acl.74",
    pages = "836--846",
}
```
    
    
    """
)