code/flash-linear-attention/benchmarks/modules/benchmark_cross_entropy.py

import torch
import torch.nn as nn
import torch.nn.functional as F
import triton

from fla.modules import FusedCrossEntropyLoss, FusedLinearCrossEntropyLoss


@triton.testing.perf_report(
    triton.testing.Benchmark(
        # argument names to use as an x-axis for the plot
        x_names=['T'],
        # different possible values for `x_name`
        x_vals=[128 * 2 ** i for i in range(0, 8)],
        # argument name whose value corresponds to a different line in the plot
        line_arg='provider',
        # possible values for `line_arg``
        line_vals=['naive',  'fused', 'fused_linear', 'naive_bwd',  'fused_bwd', 'fused_linear_bwd'],
        # label name for the lines
        line_names=['naive',  'fused', 'fused_linear', 'naive_bwd',  'fused_bwd', 'fused_linear_bwd'],
        # line styles
        styles=[('green', '-'), ('blue', '--'), ('red', '-.'),
                ('cyan', ':'), ('yellow', 'dotted'), ('cyan', '--'), ('cyan', '-'), ('black', ':')],
        ylabel="Execution Time (ms)",  # label name for the y-axis
        # name for the plot. Used also as a file name for saving the plot.
        plot_name="Performance",
        args={},
    ),
)
def benchmark(T, provider):
    from fla.utils import device
    dtype = torch.bfloat16
    requires_grad = True
    B, H, V = 4, 4096, 120000

    x = torch.randn(B * T, H, device=device, requires_grad=requires_grad, dtype=dtype)
    target = torch.randint(0, V, (B * T,), device=device, dtype=torch.int64)
    w = torch.randn(V, H, device=device, requires_grad=requires_grad, dtype=dtype)
    b = torch.randn(V, device=device, requires_grad=requires_grad, dtype=dtype)

    quantiles = [0.5, 0.2, 0.8]
    results = 0, 0, 0
    if provider == 'naive':
        criterion = nn.CrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(F.linear(x, w, b), target), quantiles=quantiles)
    elif provider == 'naive_bwd':
        criterion = nn.CrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(F.linear(x, w, b), target).backward(), quantiles=quantiles)
    elif provider == 'fused':
        criterion = FusedCrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(F.linear(x, w, b), target), quantiles=quantiles)
    elif provider == 'fused_bwd':
        criterion = FusedCrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(F.linear(x, w, b), target).backward(), quantiles=quantiles)
    elif provider == 'fused_linear':
        criterion = FusedLinearCrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(x, target, w, b), quantiles=quantiles)
    elif provider == 'fused_linear_bwd':
        criterion = FusedLinearCrossEntropyLoss()
        results = triton.testing.do_bench(lambda: criterion(x, target, w, b).backward(), quantiles=quantiles)
    return results


if __name__ == '__main__':
    benchmark.run(print_data=True)