File size: 4,435 Bytes
f32f4fc
 
 
 
 
 
 
e5e68c0
 
f32f4fc
 
 
 
 
 
 
 
 
 
1629dbf
e5e68c0
 
 
 
 
 
 
1629dbf
 
 
 
e5e68c0
 
 
 
 
 
 
 
 
1629dbf
 
 
 
 
 
e5e68c0
1629dbf
e5e68c0
1629dbf
 
 
 
 
 
 
e5e68c0
1629dbf
 
 
 
e5e68c0
 
 
 
 
 
 
 
1629dbf
55356d5
e5e68c0
 
1629dbf
e5e68c0
 
1629dbf
 
 
 
 
 
e5e68c0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f32f4fc
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
#!/usr/bin/env python
# coding: utf-8

# ### Keywords to Title Generator
# - https://huggingface.co/EnglishVoice/t5-base-keywords-to-headline?text=diabetic+diet+plan
# - Apache 2.0

# In[7]:


import torch
from transformers import T5ForConditionalGeneration,T5Tokenizer

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = T5ForConditionalGeneration.from_pretrained("EnglishVoice/t5-base-keywords-to-headline")
tokenizer = T5Tokenizer.from_pretrained("EnglishVoice/t5-base-keywords-to-headline", clean_up_tokenization_spaces=True, legacy=False)
model = model.to(device)



# In[37]:


def title_gen(keywords, diversity, temp):

 if keywords!= "":
    text =  "headline: " + keywords
    encoding = tokenizer.encode_plus(text, return_tensors = "pt")
    input_ids = encoding["input_ids"].to(device)
    attention_masks = encoding["attention_mask"].to(device)
    
    if diversity:
            num_beams = 20,
            num_beam_groups = 20,
            diversity_penalty=0.8,
            early_stopping = True,
    else:
             penalty_alpha = 0.8,
     
    beam_outputs = model.generate(
        input_ids = input_ids,
        attention_mask = attention_masks,
        max_new_tokens = 30,
        do_sample = True,
        num_return_sequences = 5,
        temperature = temp,
        top_k = 15,
        no_repeat_ngram_size = 3,
        #top_p = 0.60,
    )

    titles = ""
    
    for i in range(len(beam_outputs)):
        result = tokenizer.decode(beam_outputs[i], skip_special_tokens=True)
        titles += f"<p align=center><b>{result}</b></p>" #Create string with titles and <br> tag for html reading in gradio html
    
    return titles


# In[8]:


import gradio as gr


# In[40]:


iface = gr.Interface(fn=title_gen, 
                     inputs=[gr.Textbox(label="Paste one or more keywords searated by a comma and hit 'Submit'.", lines=1), "checkbox", gr.Slider(0.1, 1.9, 1.2)],
                     outputs=[gr.HTML(label="Title suggestions:")],
                     title="AI Keywords to Title Generator", 
                     #description="Turn keywords into creative suggestions",
                     article="<div align=left><h1>AI Creative Title Generator</h1><li>With just keywords, generate a list of creative titles.</li><li>Click on Submit to generate more title options.</li><li>Tweak slider for less or more creative titles</li><li>Check 'diversity' to turn on diversity beam search</li><p>AI Model:<br><li>T5 Model trained on a dataset of titles and related keywords</li><li>Original model id: EnglishVoice/t5-base-keywords-to-headline by English Voice AI Labs</li></p><p>Default parameter details:<br><li>Temperature = 1.2, no_repeat_ngram_size=3, top_k = 15, penalty_alpha = 0.8, max_new_tokens = 30</li><p>Diversity beam search params:<br><li>num_beams=20, diversity_penalty=0.8, num_beam_groups=20</li></div>",
                     flagging_mode='never'
                    )

iface.launch()


# In[ ]:


'''
#Create a four button panel for changing parameters with one click

def fn(text):
    return ("Hello gradio!")

with gr.Blocks () as demo:

    with gr.Row(variant='compact') as PanelRow1: #first row: top
        
        with gr.Column(scale=0, min_width=180) as PanelCol5:
            gr.HTML("")
        with gr.Column(scale=0) as PanelCol4:
            submit = gr.Button("Temp++", scale=0)
        with gr.Column(scale=1) as PanelCol5:
            gr.HTML("")
        
       
    with gr.Row(variant='compact') as PanelRow2: #2nd row: left, right, middle

        with gr.Column(min_width=100) as PanelCol1:
            submit = gr.Button("Contrastive")
        with gr.Column(min_width=100) as PanelCol2:
            submit = gr.Button("Re-generate")
        with gr.Column(min_width=100) as PanelCol3:
            submit = gr.Button("Diversity Beam")
        
        with gr.Column(min_width=100) as PanelCol5:
            gr.HTML("")
        with gr.Column(min_width=100) as PanelCol5:
            gr.HTML("")
        with gr.Column(scale=0) as PanelCol5:
            gr.HTML("")
            
    with gr.Row(variant='compact') as PanelRow3: #last row: down
        with gr.Column(scale=0, min_width=180) as PanelCol7:
            gr.HTML("")
        with gr.Column(scale=1) as PanelCol6:
            submit = gr.Button("Temp--", scale=0)
            
        with gr.Column(scale=0) as PanelCol5:
            gr.HTML("")

demo.launch()
'''