File size: 3,160 Bytes
199a0ec
 
 
 
 
de9ee5d
199a0ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de9ee5d
199a0ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de9ee5d
 
199a0ec
 
 
 
 
de9ee5d
 
199a0ec
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
from transformers import (
    pipeline,
    AutoModelForSpeechSeq2Seq,
    AutoProcessor,
    AutoModelForCausalLM,
    AutoTokenizer, 
    BitsAndBytesConfig,
)
import torch
import os
import random

def yt2mp3(url, outputMp3F):
    tmpVideoF=random.random()
    os.system(f"./bin/youtube-dl -o /tmp/{tmpVideoF} --verbose " + url)
    os.system(f"ffmpeg -y -i /tmp/{tmpVideoF}.* -vn -ar 44100 -ac 2 -b:a 192k {outputMp3F}")


def speech2text(mp3_file):
    device = 'cuda:0'

    torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

    model_id = "distil-whisper/distil-large-v2"

    model = AutoModelForSpeechSeq2Seq.from_pretrained(
        model_id, 
        torch_dtype=torch_dtype, 
        low_cpu_mem_usage=True, 
        use_safetensors=True,
        use_flash_attention_2=True
    )

    model.to(device)

    processor = AutoProcessor.from_pretrained(model_id)

    pipe = pipeline(
        "automatic-speech-recognition",
        model=model,
        tokenizer=processor.tokenizer,
        feature_extractor=processor.feature_extractor,
        max_new_tokens=128,
        chunk_length_s=15,
        batch_size=16,
        torch_dtype=torch_dtype,
        device=device,
    )

    result = pipe(mp3_file)

    text_from_video = result["text"]

    return text_from_video


def chat(system_prompt, text):
    model_name = "meta-llama/Llama-2-7b-chat-hf"
    token = os.environ['HUGGINGFACE_TOKEN']

    bnb_config = BitsAndBytesConfig(
        load_in_8bit=True
    )

    device_map = {"": 0}

    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        quantization_config=bnb_config,
        device_map=device_map,
        use_auth_token=token
    )

    tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=token)

    llama_pipeline = pipeline(task="text-generation", model=model, tokenizer=tokenizer)

    text = f"""
    <s>[INST] <<SYS>>
    {system_prompt}
    <</SYS>>
    {text}[/INST]
    """

    sequences = llama_pipeline(
        text,
        do_sample=True,
        top_k=10,
        num_return_sequences=1,
        eos_token_id=tokenizer.eos_token_id,
        max_length=32000
    )

    generated_text = sequences[0]["generated_text"]
    generated_text = generated_text[generated_text.find('[/INST]')+len('[/INST]'):]

    return generated_text

def summarize(text):
    input_len = 10000

    while True:
        summary = chat("", "Summarize the following: " + text[0:input_len])

        if len(text) < input_len:
            return summary
        
        text = summary + " " + text[input_len:]

import gradio as gr

def summarize_from_youtube(url):
    outputMp3F = "./files/audio.mp3"
    yt2mp3(url=url, outputMp3F=outputMp3F)
    transcribed = speech2text(mp3_file=outputMp3F)
    summary = summarize(transcribed)
    return summary

youtube_url = gr.inputs.Textbox(lines=1, label="Enter YouTube URL")
output_text = gr.outputs.Textbox(label="Summary")

gr.Interface(
    fn=summarize_from_youtube, 
    inputs=youtube_url, 
    outputs=output_text, 
    title="YouTube Summarizer",
    description="Enter a YouTube URL to summarize its content."
).launch()