|
| 1 | +#!/usr/bin/env python3 |
| 2 | +import logging |
| 3 | +import argparse |
| 4 | +import os |
| 5 | +import sys |
| 6 | +import json |
| 7 | +from pathlib import Path |
| 8 | + |
| 9 | +import numpy as np |
| 10 | +from typing import Any, Mapping, Sequence |
| 11 | + |
| 12 | +# Necessary to load the local gguf package |
| 13 | +if "NO_LOCAL_GGUF" not in os.environ and (Path(__file__).parent.parent.parent / 'gguf-py').exists(): |
| 14 | + sys.path.insert(0, str(Path(__file__).parent.parent)) |
| 15 | + |
| 16 | +import gguf |
| 17 | + |
| 18 | +logger = logging.getLogger("gguf-new-metadata") |
| 19 | + |
| 20 | + |
| 21 | +def get_byteorder(reader: gguf.GGUFReader) -> gguf.GGUFEndian: |
| 22 | + if np.uint32(1) == np.uint32(1).newbyteorder("<"): |
| 23 | + # Host is little endian |
| 24 | + host_endian = gguf.GGUFEndian.LITTLE |
| 25 | + swapped_endian = gguf.GGUFEndian.BIG |
| 26 | + else: |
| 27 | + # Sorry PDP or other weird systems that don't use BE or LE. |
| 28 | + host_endian = gguf.GGUFEndian.BIG |
| 29 | + swapped_endian = gguf.GGUFEndian.LITTLE |
| 30 | + |
| 31 | + if reader.byte_order == "S": |
| 32 | + return swapped_endian |
| 33 | + else: |
| 34 | + return host_endian |
| 35 | + |
| 36 | + |
| 37 | +def decode_field(field: gguf.ReaderField) -> Any: |
| 38 | + if field and field.types: |
| 39 | + main_type = field.types[0] |
| 40 | + |
| 41 | + if main_type == gguf.GGUFValueType.ARRAY: |
| 42 | + sub_type = field.types[-1] |
| 43 | + |
| 44 | + if sub_type == gguf.GGUFValueType.STRING: |
| 45 | + return [str(bytes(field.parts[idx]), encoding='utf8') for idx in field.data] |
| 46 | + else: |
| 47 | + return [pv for idx in field.data for pv in field.parts[idx].tolist()] |
| 48 | + if main_type == gguf.GGUFValueType.STRING: |
| 49 | + return str(bytes(field.parts[-1]), encoding='utf8') |
| 50 | + else: |
| 51 | + return field.parts[-1][0] |
| 52 | + |
| 53 | + return None |
| 54 | + |
| 55 | + |
| 56 | +def get_field_data(reader: gguf.GGUFReader, key: str) -> Any: |
| 57 | + field = reader.get_field(key) |
| 58 | + |
| 59 | + return decode_field(field) |
| 60 | + |
| 61 | + |
| 62 | +def copy_with_new_metadata(reader: gguf.GGUFReader, writer: gguf.GGUFWriter, new_metadata: Mapping[str, str], remove_metadata: Sequence[str]) -> None: |
| 63 | + for field in reader.fields.values(): |
| 64 | + # Suppress virtual fields and fields written by GGUFWriter |
| 65 | + if field.name == gguf.Keys.General.ARCHITECTURE or field.name.startswith('GGUF.'): |
| 66 | + logger.debug(f'Suppressing {field.name}') |
| 67 | + continue |
| 68 | + |
| 69 | + # Skip old chat templates if we have new ones |
| 70 | + if field.name.startswith(gguf.Keys.Tokenizer.CHAT_TEMPLATE) and gguf.Keys.Tokenizer.CHAT_TEMPLATE in new_metadata: |
| 71 | + logger.debug(f'Skipping {field.name}') |
| 72 | + continue |
| 73 | + |
| 74 | + if field.name in remove_metadata: |
| 75 | + logger.debug(f'Removing {field.name}') |
| 76 | + continue |
| 77 | + |
| 78 | + old_val = decode_field(field) |
| 79 | + val = new_metadata.get(field.name, old_val) |
| 80 | + |
| 81 | + if field.name in new_metadata: |
| 82 | + logger.debug(f'Modifying {field.name}: "{old_val}" -> "{val}"') |
| 83 | + del new_metadata[field.name] |
| 84 | + elif val is not None: |
| 85 | + logger.debug(f'Copying {field.name}') |
| 86 | + |
| 87 | + if val is not None: |
| 88 | + writer.add_key(field.name) |
| 89 | + writer.add_val(val, field.types[0]) |
| 90 | + |
| 91 | + if gguf.Keys.Tokenizer.CHAT_TEMPLATE in new_metadata: |
| 92 | + logger.debug('Adding chat template(s)') |
| 93 | + writer.add_chat_template(new_metadata[gguf.Keys.Tokenizer.CHAT_TEMPLATE]) |
| 94 | + del new_metadata[gguf.Keys.Tokenizer.CHAT_TEMPLATE] |
| 95 | + |
| 96 | + # TODO: Support other types than string? |
| 97 | + for key, val in new_metadata.items(): |
| 98 | + logger.debug(f'Adding {key}: {val}') |
| 99 | + writer.add_key(key) |
| 100 | + writer.add_val(val, gguf.GGUFValueType.STRING) |
| 101 | + |
| 102 | + for tensor in reader.tensors: |
| 103 | + # Dimensions are written in reverse order, so flip them first |
| 104 | + shape = np.flipud(tensor.shape) |
| 105 | + writer.add_tensor_info(tensor.name, shape, tensor.data.dtype, tensor.data.nbytes, tensor.tensor_type) |
| 106 | + |
| 107 | + writer.write_header_to_file() |
| 108 | + writer.write_kv_data_to_file() |
| 109 | + writer.write_ti_data_to_file() |
| 110 | + |
| 111 | + for tensor in reader.tensors: |
| 112 | + writer.write_tensor_data(tensor.data) |
| 113 | + |
| 114 | + writer.close() |
| 115 | + |
| 116 | + |
| 117 | +def main() -> None: |
| 118 | + parser = argparse.ArgumentParser(description="Make a copy of a GGUF file with new metadata") |
| 119 | + parser.add_argument("input", type=Path, help="GGUF format model input filename") |
| 120 | + parser.add_argument("output", type=Path, help="GGUF format model output filename") |
| 121 | + parser.add_argument("--general-name", type=str, help="The models general.name") |
| 122 | + parser.add_argument("--general-description", type=str, help="The models general.description") |
| 123 | + parser.add_argument("--chat-template", type=str, help="Chat template string (or JSON string containing templates)") |
| 124 | + parser.add_argument("--chat-template-config", type=Path, help="Config file (tokenizer_config.json) containing chat template(s)") |
| 125 | + parser.add_argument("--remove-metadata", action="append", type=str, help="Remove metadata (by key name) from output model") |
| 126 | + parser.add_argument("--force", action="store_true", help="Bypass warnings without confirmation") |
| 127 | + parser.add_argument("--verbose", action="store_true", help="Increase output verbosity") |
| 128 | + args = parser.parse_args(None if len(sys.argv) > 2 else ["--help"]) |
| 129 | + |
| 130 | + logging.basicConfig(level=logging.DEBUG if args.verbose else logging.INFO) |
| 131 | + |
| 132 | + new_metadata = {} |
| 133 | + remove_metadata = args.remove_metadata or [] |
| 134 | + |
| 135 | + if args.general_name: |
| 136 | + new_metadata[gguf.Keys.General.NAME] = args.general_name |
| 137 | + |
| 138 | + if args.general_description: |
| 139 | + new_metadata[gguf.Keys.General.DESCRIPTION] = args.general_description |
| 140 | + |
| 141 | + if args.chat_template: |
| 142 | + new_metadata[gguf.Keys.Tokenizer.CHAT_TEMPLATE] = json.loads(args.chat_template) if args.chat_template.startswith('[') else args.chat_template |
| 143 | + |
| 144 | + if args.chat_template_config: |
| 145 | + with open(args.chat_template_config, 'r') as fp: |
| 146 | + config = json.load(fp) |
| 147 | + template = config.get('chat_template') |
| 148 | + if template: |
| 149 | + new_metadata[gguf.Keys.Tokenizer.CHAT_TEMPLATE] = template |
| 150 | + |
| 151 | + if remove_metadata: |
| 152 | + logger.warning('*** Warning *** Warning *** Warning **') |
| 153 | + logger.warning('* Most metadata is required for a fully functional GGUF file,') |
| 154 | + logger.warning('* removing crucial metadata may result in a corrupt output file!') |
| 155 | + |
| 156 | + if not args.force: |
| 157 | + logger.warning('* Enter exactly YES if you are positive you want to proceed:') |
| 158 | + response = input('YES, I am sure> ') |
| 159 | + if response != 'YES': |
| 160 | + logger.info("You didn't enter YES. Okay then, see ya!") |
| 161 | + sys.exit(0) |
| 162 | + |
| 163 | + logger.info(f'* Loading: {args.input}') |
| 164 | + reader = gguf.GGUFReader(args.input, 'r') |
| 165 | + |
| 166 | + arch = get_field_data(reader, gguf.Keys.General.ARCHITECTURE) |
| 167 | + endianess = get_byteorder(reader) |
| 168 | + |
| 169 | + if os.path.isfile(args.output) and not args.force: |
| 170 | + logger.warning('*** Warning *** Warning *** Warning **') |
| 171 | + logger.warning(f'* The "{args.output}" GGUF file already exists, it will be overwritten!') |
| 172 | + logger.warning('* Enter exactly YES if you are positive you want to proceed:') |
| 173 | + response = input('YES, I am sure> ') |
| 174 | + if response != 'YES': |
| 175 | + logger.info("You didn't enter YES. Okay then, see ya!") |
| 176 | + sys.exit(0) |
| 177 | + |
| 178 | + logger.info(f'* Writing: {args.output}') |
| 179 | + writer = gguf.GGUFWriter(args.output, arch=arch, endianess=endianess) |
| 180 | + |
| 181 | + alignment = get_field_data(reader, gguf.Keys.General.ALIGNMENT) |
| 182 | + if alignment is not None: |
| 183 | + logger.debug(f'Setting custom alignment: {alignment}') |
| 184 | + writer.data_alignment = alignment |
| 185 | + |
| 186 | + copy_with_new_metadata(reader, writer, new_metadata, remove_metadata) |
| 187 | + |
| 188 | + |
| 189 | +if __name__ == '__main__': |
| 190 | + main() |
0 commit comments