char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Sigbjørn SkjæretRefactor gguf scripts to improve metadata handling (#11909)69050a11b

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
1.6 KiB49 linesraw
1#!/usr/bin/env python3
2import logging
3import sys
4from pathlib import Path
5
6logger = logging.getLogger("reader")
7
8# Necessary to load the local gguf package
9sys.path.insert(0, str(Path(__file__).parent.parent))
10
11from gguf.gguf_reader import GGUFReader
12
13
14def read_gguf_file(gguf_file_path):
15    """
16    Reads and prints key-value pairs and tensor information from a GGUF file in an improved format.
17
18    Parameters:
19    - gguf_file_path: Path to the GGUF file.
20    """
21
22    reader = GGUFReader(gguf_file_path)
23
24    # List all key-value pairs in a columnized format
25    print("Key-Value Pairs:") # noqa: NP100
26    max_key_length = max(len(key) for key in reader.fields.keys())
27    for key, field in reader.fields.items():
28        value = field.parts[field.data[0]]
29        print(f"{key:{max_key_length}} : {value}") # noqa: NP100
30    print("----") # noqa: NP100
31
32    # List all tensors
33    print("Tensors:") # noqa: NP100
34    tensor_info_format = "{:<30} | Shape: {:<15} | Size: {:<12} | Quantization: {}"
35    print(tensor_info_format.format("Tensor Name", "Shape", "Size", "Quantization")) # noqa: NP100
36    print("-" * 80) # noqa: NP100
37    for tensor in reader.tensors:
38        shape_str = "x".join(map(str, tensor.shape))
39        size_str = str(tensor.n_elements)
40        quantization_str = tensor.tensor_type.name
41        print(tensor_info_format.format(tensor.name, shape_str, size_str, quantization_str)) # noqa: NP100
42
43
44if __name__ == '__main__':
45    if len(sys.argv) < 2:
46        logger.info("Usage: reader.py <path_to_gguf_file>")
47        sys.exit(1)
48    gguf_file_path = sys.argv[1]
49    read_gguf_file(gguf_file_path)