File size: 4,923 Bytes
8af190d
 
 
8913d2d
8af190d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Run the script and open the link in the browser.

import os
import json
import gradio as gr
import streamlit as st
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# scratch with latbert tokenizer
CHECKPOINT_PATH= 'scratch_2-nodes_tokenizer_latbert-original_packing_fcocchi/'
CHECKPOINT_PATH= 'itserr/latin_llm_alpha'

print(f"Loading model from: {CHECKPOINT_PATH}")
tokenizer = AutoTokenizer.from_pretrained(CHECKPOINT_PATH, token=os.environ['HF_TOKEN'])
model = AutoModelForCausalLM.from_pretrained(CHECKPOINT_PATH, token=os.environ['HF_TOKEN'])

description="""
This is a Latin Language Model (LLM) based on GPT-2 and it was trained on a large corpus of Latin texts and can generate text in Latin. \n
Demo instructions:
- Enter a prompt in Latin in the Input Text box.
- Select the temperature value to control the randomness of the generated text (higher value produce a more creative and unstable answer).
- Click the 'Generate Text' button to trigger model generation.
- (Optional) insert a Feedback text in the box.
- Click the 'Like' or 'Dislike' button to judge the generation correctness. 
"""
title= "(L<sup>2</sup>) - Latin Language Model"
article= "hello world ..."
examples= ['Accidere ex una scintilla', 'Audacter calumniare,', 'Consolatium misero comites']
logo_image= 'ITSERR_row_logo.png'

def generate_text(prompt, slider):
    if torch.cuda.is_available(): device = torch.device("cuda")      
    else: 
        device = torch.device("cpu")
        print("No GPU available")
    
    print("***** Generate *****")
    text_generator = pipeline("text-generation", model=model, tokenizer=tokenizer, device=device)
    #generated_text = text_generator(prompt, max_length=100)
    generated_text = text_generator(prompt, max_length=50, do_sample=True, temperature=slider, repetition_penalty=2.0, truncation=True)
    return generated_text[0]['generated_text']

# Function to handle user preferences
def handle_preference(preference, input, output, feedback, temp_value, preferences_file="preferences.json"):
    """
    Format values stored in preferences:
        - input text
        - output generated text
        - user feedback
        - float temperature value
    """
    
    if os.path.exists(preferences_file):
        with open(preferences_file, "r") as file:
            preferences = json.load(file)
    else:
        preferences = {"like": [], "dislike": [], "count_like": 0, "count_dislike": 0}
    
    if input == output:
        output_tuple= ("", "", feedback)
    else:
        output_tuple= (input, output.split(input)[-1], feedback, temp_value)
    if preference == "like":
        preferences["like"].append(output_tuple)
        if output_tuple[1] != "" :
            preferences["count_like"] += 1
    elif preference == "dislike":
        preferences["dislike"].append(output_tuple)
        if output_tuple[1] != "" :
            preferences["count_dislike"] += 1
    
    with open(preferences_file, "w") as file:
        json.dump(preferences, file)
    
    print(f"Admin log: like: {preferences['count_like']} and dislike: {preferences['count_dislike']}")
    return f"You select '{preference}' as answer of the model generation. Thank you for your time!"

custom_css = """
#logo {
    display: block;
    margin-left: auto;
    margin-right: auto;
    width: 280px;
    height: 140px;
}
"""

with gr.Blocks(css=custom_css) as demo:
    gr.Image(logo_image, elem_id="logo")
    gr.Markdown(f"<h1 style='text-align: center;'>{title}</h1>")
    gr.Markdown(description)
    
    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(lines=5, placeholder="Enter latin text here...", label="Input Text")
        with gr.Column():
            output_text = gr.Textbox(lines=5, placeholder="Output text will appear here...", label="Output Text")

    gr.Examples(examples=examples, inputs=input_text)
    temperature_slider = gr.Slider(minimum=0.1, maximum=5.0, step=0.1, value=1.0, label="Temperature")
    
    clean_button = gr.Button("Generate Text")
    clean_button.click(fn=generate_text, inputs=[input_text, temperature_slider], outputs=output_text)
    feedback_output = gr.Textbox(lines=1, placeholder="If you want to provide a feedback, please fill this box ...", label="Feedback")

    with gr.Row():
        like_button = gr.Button("Like")
        dislike_button = gr.Button("Dislike")

    button_output = gr.Textbox(lines=1, placeholder="Please submit your choice", label="Latin Language Model Demo")
    like_button.click(fn=lambda x,y,z,v: handle_preference("like", x, y, z, v), inputs=[input_text, output_text, feedback_output, temperature_slider], outputs=button_output)
    dislike_button.click(fn=lambda x,y,z,v: handle_preference("dislike", x, y, z, v), inputs=[input_text, output_text, feedback_output, temperature_slider], outputs=button_output)
    #gr.Markdown(article)

demo.launch(share=True)