Skip to content

bug: missing complexity analysis #155

Description

@aubler
  • need comprehensive model profiling
  • no flop count available
  • parameter counts not reported
  • memory usage unclear

fix:

pip install fvcore

then:

from fvcore.nn import FlopCountAnalysis, parameter_count_table

def profile_model(model, input_size=(32, 96, 7)):
    # Create sample input
    x = torch.randn(input_size)
    x_mark = torch.randn(input_size[0], input_size[1], 4)
    
    # Count parameters
    params = parameter_count_table(model)
    
    # Count FLOPs
    flops = FlopCountAnalysis(model, (x, x_mark, x, x_mark))
    
    return {
        "params": params,
        "flops": flops.total(),
        "flops_readable": f"{flops.total() / 1e9:.2f}G"
    }

add to model's repr method

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions