File size: 7,467 Bytes
1e57684
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
import argparse
import os
import shutil
import subprocess
import time
from pathlib import Path

from git import Repo


def clone_hf_with_git(username: str, model_name: str, saved_dir: str):
    full_model_name = f"{username}/{model_name}"
    url = f"https://huggingface.co/{full_model_name}"
    saved = f"{saved_dir}/{model_name}"

    # perform `git lfs install`
    subprocess.run(["git", "lfs", "install"])

    print(f"[INFO] Cloning {model_name} from {url} ...")
    Repo.clone_from(url, saved)


def download_hf_with_git(full_name: str, saved_dir: str):
    model_name = full_name.split("/")[1]
    url = f"git@hf.co:{full_name}"
    saved = f"{saved_dir}/{model_name}"

    # perform `git lfs install`
    subprocess.run(["git", "lfs", "install"])

    print(f"Cloning {model_name} from {url} ...")
    subprocess.run(["git", "clone", "--progress", url, saved])


def convert_hf_to_gguf(
    script_path: str,
    dir_raw_model: str,
    gguf_model_path: str,
    pad_vocab: bool = False,
):
    if pad_vocab is True:
        args = [
            "--outfile",
            gguf_model_path,
            # "--vocab-type",
            # "bpe",
            "--pad-vocab",
            dir_raw_model,
        ]
    else:
        args = ["--outfile", gguf_model_path, dir_raw_model]
        # convert.py for llama-3
        # args = ["--outfile", gguf_model_path, "--vocab-type", "bpe", dir_raw_model]
    res = subprocess.run(["python", script_path] + args)
    print(res)


def quantize_model(
    quantizer: str,
    f16_gguf_model_path: str,
    quantized_gguf_model_path: str,
    quant_type: str = "q4_0",
):
    print(f"[INFO] quantizer: {quantizer}")
    print(f"[INFO] quant_type: {quant_type}")
    print(f"[INFO] f16_gguf_model_path: {f16_gguf_model_path}")
    print(f"[INFO] quantized_model_filename: {quantized_gguf_model_path}")
    subprocess.run(
        [
            quantizer,
            f16_gguf_model_path,
            quantized_gguf_model_path,
            quant_type,
        ]
    )


def main():
    parser = argparse.ArgumentParser(description="Convert and quantize gguf models.")
    parser.add_argument(
        "--full-name",
        type=str,
        help="Huggingface model full name. e.g. `username/model_name`",
    )
    parser.add_argument(
        "-s",
        "--saved-dir",
        type=str,
        default="models",
        help="The directory to save the model.",
    )
    parser.add_argument(
        "--enable-converter",
        action="store_true",
        help="Enable the converter. Notice that `--converter` must be specified.",
    )
    parser.add_argument(
        "-c",
        "--converter",
        type=str,
        help="The path to the converter. Notice that `--enable-converter` must be specified if use this option.",
    )
    parser.add_argument(
        "--pad-vocab",
        action="store_true",
        help="Enable adding pad tokens when model vocab expects more than tokenizer metadata provides. Notice that `--enable-converter` must be specified.",
    )
    parser.add_argument(
        "--enable-quantizer",
        action="store_true",
        help="Enable the quantizer. Notice that `--quantizer` must be specified.",
    )
    parser.add_argument(
        "-q",
        "--quantizer",
        type=str,
        help="The path to the quantizer. Notice that `--enable-quantizer` must be specified if use this option.",
    )
    parser.add_argument(
        "-t",
        "--quant-type",
        type=str,
        default=None,
        help="The quantization type. Notice that `--enable-quantizer` must be specified if use this option.",
    )

    args = parser.parse_args()

    print(args)

    print("Download model ...")
    full_name = args.full_name
    username, model_name = full_name.split("/")
    saved_dir = args.saved_dir
    # try:
    #     download_hf_with_git(full_name, saved_dir)
    #     print(f"The raw model is saved in {saved_dir}.")

    # except Exception as e:
    #     print(f"Failed to download model. {e}")
    #     return

    if args.enable_converter is True:
        print("[CONVERTER] Convert model ...")
        converter = args.converter

        raw_model_dir = f"{saved_dir}/{model_name}"
        print(f"[CONVERTER] raw_model_dir: {raw_model_dir}")

        gguf_model_dir = Path(raw_model_dir).parent / f"{model_name}-gguf"
        if not gguf_model_dir.exists():
            gguf_model_dir.mkdir()
        f16_gguf_model_path = gguf_model_dir / f"{model_name}-f16.gguf"

        print(f"[CONVERTER] f16_gguf_model_path: {f16_gguf_model_path}")

        # try:
        #     convert_hf_to_gguf(
        #         converter,
        #         raw_model_dir,
        #         str(f16_gguf_model_path),
        #         args.pad_vocab,
        #     )
        #     print(f"The converted gguf model is saved in {f16_gguf_model_path}.")

        # except Exception as e:
        #     print(f"Failed to convert model. {e}")
        #     return

    if args.enable_quantizer is True:
        print("[QUANTIZER] Quantize model ...")
        quantizer = args.quantizer
        print(f"[QUANTIZER] quantizer: {quantizer}")

        if args.quant_type is not None:
            quant_type = args.quant_type
            quantized_gguf_model_path = (
                gguf_model_dir / f"{model_name}-{quant_type}.gguf"
            )

            print(f"[QUANTIZER] quant_type: {quant_type}")
            print(f"[QUANTIZER] quantized_model_filename: {quantized_gguf_model_path}")

            try:
                quantize_model(
                    quantizer,
                    str(f16_gguf_model_path),
                    str(quantized_gguf_model_path),
                    quant_type,
                )
                print(
                    f"The quantized gguf model is saved in {quantized_gguf_model_path}."
                )

            except Exception as e:
                print(e)
                print("Failed to quantize model.")
                return
        else:
            for quant_type in [
                # "Q2_K",
                # "Q3_K_L",
                # "Q3_K_M",
                # "Q3_K_S",
                # "Q4_0",
                # "Q4_K_M",
                # "Q4_K_S",
                # "Q5_0",
                "Q5_K_M",
                # "Q5_K_S",
                "Q6_K",
                "Q8_0",
            ]:
                quantized_gguf_model_path = (
                    gguf_model_dir / f"{model_name}-{quant_type}.gguf"
                )

                print(f"[QUANTIZER] quant_type: {quant_type}")
                print(
                    f"[QUANTIZER] quantized_model_filename: {quantized_gguf_model_path}"
                )

                try:
                    quantize_model(
                        quantizer,
                        str(f16_gguf_model_path),
                        str(quantized_gguf_model_path),
                        quant_type,
                    )
                    print(
                        f"The quantized gguf model is saved in {quantized_gguf_model_path}."
                    )

                except Exception as e:
                    print(e)
                    print("Failed to quantize model.")
                    return

        # # remove the raw model dir for saving space
        # print(f"The quantization is done. Remove {raw_model_dir}")
        # shutil.rmtree(raw_model_dir)

    print("Done.")


if __name__ == "__main__":
    main()