Skip to content

scripts: add bpw per layer and model #14703

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 15, 2025
Merged

scripts: add bpw per layer and model #14703

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 24 additions & 1 deletion gguf-py/gguf/scripts/gguf_dump.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -234,6 +234,8 @@ def dump_markdown_metadata(reader: GGUFReader, args: argparse.Namespace) -> None
markdown_content += '## Key Value Metadata Store\n\n'
markdown_content += f'There are {len(reader.fields)} key-value pairs in this file\n'
markdown_content += '\n'
total_model_bytes = 0
total_model_elements = 0

kv_dump_table: list[dict[str, str | int]] = []
for n, field in enumerate(reader.fields.values(), 1):
Expand Down Expand Up @@ -377,6 +379,8 @@ def escape_markdown_inline_code(value_string):
tensors = tensor_groups[group]
group_elements = sum(tensor.n_elements for tensor in tensors)
group_percentage = group_elements / total_elements * 100
total_group_bytes = 0
total_group_elements = 0
markdown_content += f"### <a name=\"{group.replace('.', '_')}\">{translate_tensor_name(group)} Tensor Group : {element_count_rounded_notation(group_elements)} Elements</a>\n\n"

# Precalculate column sizing for visual consistency
Expand All @@ -397,7 +401,13 @@ def escape_markdown_inline_code(value_string):
element_count_est = f"({element_count_rounded_notation(tensor.n_elements):>{prettify_element_est_count_size}})"
element_count_string = f"{element_count_est} {tensor.n_elements:>{prettify_element_count_size}}"
type_name_string = f"{tensor.tensor_type.name}"
tensor_dump_table.append({"t_id":tensor_name_to_key[tensor.name], "layer_name":tensor.name, "human_layer_name":human_friendly_name, "element_count":element_count_string, "pretty_dimension":pretty_dimension, "tensor_type":type_name_string})
if tensor.n_elements > 0:
bpw = (tensor.n_bytes * 8) / tensor.n_elements
else:
bpw = float('nan')
tensor_dump_table.append({"t_id":tensor_name_to_key[tensor.name], "layer_name":tensor.name, "human_layer_name":human_friendly_name, "element_count":element_count_string, "pretty_dimension":pretty_dimension, "tensor_type":type_name_string, "bpw": f"{bpw:.4f}"})
total_group_bytes += tensor.n_bytes
total_group_elements += tensor.n_elements

tensor_dump_table_header_map = [
{'key_name':'t_id', 'header_name':'T_ID', 'align':'right'},
Expand All @@ -406,15 +416,28 @@ def escape_markdown_inline_code(value_string):
{'key_name':'element_count', 'header_name':'Elements', 'align':'left'},
{'key_name':'pretty_dimension', 'header_name':'Shape', 'align':'left'},
{'key_name':'tensor_type', 'header_name':'Type', 'align':'left'},
{'key_name':'bpw', 'header_name':'BPW', 'align':'right'},
]

markdown_content += markdown_table_with_alignment_support(tensor_dump_table_header_map, tensor_dump_table)

markdown_content += "\n"
markdown_content += f"- Total elements in {group}: ({element_count_rounded_notation(group_elements):>4}) {group_elements}\n"
markdown_content += f"- Percentage of total elements: {group_percentage:.2f}%\n"
if total_group_elements > 0:
total_group_bpw = (total_group_bytes * 8) / total_group_elements
markdown_content += f"- Bits per Weight (BPW) for {group}: {total_group_bpw:.4f} bits\n"
else:
markdown_content += f"- Bits per Weight (BPW) for {group}: undefined (no elements)\n"
markdown_content += "\n\n"
total_model_bytes += total_group_bytes
total_model_elements += total_group_elements

if total_model_elements > 0:
total_model_bpw = (total_model_bytes * 8) / total_model_elements
markdown_content += f"Total BPW for {os.path.basename(args.model)}: {total_model_bpw:.4f} bits"
else:
markdown_content += f"Total BPW for {os.path.basename(args.model)}: undefined (no elements)"
print(markdown_content) # noqa: NP100


Expand Down