File size: 2,893 Bytes
7edda8b
2bede7c
 
7edda8b
 
2bede7c
 
75b770e
2bede7c
 
 
75b770e
2bede7c
 
 
75b770e
 
2bede7c
 
 
 
 
 
 
 
 
 
 
 
 
 
7edda8b
2bede7c
 
 
 
 
 
 
 
75b770e
 
 
 
 
 
 
 
2bede7c
75b770e
2bede7c
 
 
7edda8b
 
 
7cd57ad
 
 
 
2bede7c
 
 
 
 
7edda8b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2bede7c
7cd57ad
 
 
 
 
 
 
 
2bede7c
 
 
 
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
import shutil
import subprocess

import gradio as gr

from huggingface_hub import create_repo, HfApi
from huggingface_hub import snapshot_download
from huggingface_hub import whoami

api = HfApi()

def process_model(model_id, q_method, hf_token):
    
    MODEL_NAME = model_id.split('/')[-1]
    fp16 = f"{MODEL_NAME}/{MODEL_NAME.lower()}.fp16.bin"

    username = whoami(hf_token)["name"]
    
    snapshot_download(repo_id=model_id, local_dir = f"{MODEL_NAME}", local_dir_use_symlinks=False)
    print("Model downloaded successully!")
    
    fp16_conversion = f"python llama.cpp/convert.py {MODEL_NAME} --outtype f16 --outfile {fp16}"
    subprocess.run(fp16_conversion, shell=True)
    print("Model converted to fp16 successully!")

    qtype = f"{MODEL_NAME}/{MODEL_NAME.lower()}.{q_method.upper()}.gguf"
    quantise_ggml = f"./llama.cpp/quantize {fp16} {qtype} {q_method}"
    subprocess.run(quantise_ggml, shell=True)
    print("Quantised successfully!")

    # Create empty repo
    repo_url = create_repo(
        repo_id = f"{username}/{MODEL_NAME}-{q_method}-GGUF",
        repo_type="model",
        exist_ok=True,
        token=hf_token
    )
    print("Empty repo created successfully!")

    # Upload gguf files
    # api.upload_folder(
    #     folder_path=MODEL_NAME,
    #     repo_id=f"{username}/{MODEL_NAME}-{q_method}-GGUF",
    #     allow_patterns=["*.gguf"],
    #     token=hf_token
    # )
    api.upload_file(
        path_or_fileobj=qtype,
        repo_id=f"{username}/{MODEL_NAME}-{q_method}-GGUF",
        repo_type="model",
    )
    print("Uploaded successfully!")

    shutil.rmtree(MODEL_NAME)
    print("Folder cleaned up successfully!")

    return (
        f'Find your repo <a href=\'{repo_url}\' target="_blank" style="text-decoration:underline">here</a>',
        "llama.png",
    )    

# Create Gradio interface
iface = gr.Interface(
    fn=process_model, 
    inputs=[
        gr.Textbox(
            lines=1, 
            label="Hub Model ID",
            info="Model repo ID"
        ),
        gr.Dropdown(
            ["Q2_K", "Q3_K_S", "Q3_K_M", "Q3_K_L", "Q4_0", "Q4_K_S", "Q4_K_M", "Q5_0", "Q5_K_S", "Q5_K_M", "Q6_K", "Q8_0"], 
            label="Quantization Method", 
            info="GGML quantisation type"
        ),
        gr.Textbox(
            lines=1, 
            label="HF Write Token",
            info="https://hf.co/settings/token"
        )
    ], 
    outputs=[
        gr.Markdown(label="output"),
        gr.Image(show_label=False),
    ],
    title="Create your own GGUF Quants!",
    description="Create GGUF quants from any Hugging Face repository! You need to specify a write token obtained in https://hf.co/settings/tokens.",
    article="<p>Find your write token at <a href='https://huggingface.co/settings/tokens' target='_blank'>token settings</a></p>",
    
)

# Launch the interface
iface.launch(debug=True)