SwiGLU Activation Benchmarks - Aggregated Results

This document combines benchmark results from multiple SwiGLU activation implementations.

Combined Summary and Visualization

▶ code ▼ output ▶ uv-logs | Cell: combine | 4.32s | Raw

# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "numpy",
#     "torch==2.8.0",
#     "kernels-benchmark-tools",
#     "matplotlib",
# ]
#
# [tool.uv.sources]
# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
# ///
from kernels_benchmark_tools.core.visuals import generate_combined_results

# Map display names to uvnote environment variables
cache_env_map = {
    "HF Kernels SwiGLU": "UVNOTE_FILE_HF_KERNELS_SWIGLU_BENCHMARK",
    "PyTorch SwiGLU": "UVNOTE_FILE_TORCH_SWIGLU_BENCHMARK",
    # "Compiled SwiGLU": "UVNOTE_FILE_COMPILED_SWIGLU_BENCHMARK",
}

# Generate combined results with visualization
generate_combined_results(
    cache_env_map=cache_env_map,
    output_filename="activation.jsonl",
    svg_filename="latency.svg"
)

======================================================================
LOADING BENCHMARK DATA
======================================================================
✓ HF Kernels SwiGLU             : /__w/kernels-benchmarks/kernels-benchmarks/benches/activation/impls/.uvnote/cache/2775e6386f1caf1fda935a997130c06dcaf7641efb0db21560c35301fdabfd9b
✓ PyTorch SwiGLU                : /__w/kernels-benchmarks/kernels-benchmarks/benches/activation/impls/.uvnote/cache/661ca38adec8893d7c284140e922da661f0afcea4aaff6a3bf48a6494ce7c6eb

  ✓ Found HF Kernels SwiGLU
     Path: /__w/kernels-benchmarks/kernels-benchmarks/benches/activation/impls/.uvnote/cache/2775e6386f1caf1fda935a997130c06dcaf7641efb0db21560c35301fdabfd9b/activation.jsonl
  ✓ Found PyTorch SwiGLU
     Path: /__w/kernels-benchmarks/kernels-benchmarks/benches/activation/impls/.uvnote/cache/661ca38adec8893d7c284140e922da661f0afcea4aaff6a3bf48a6494ce7c6eb/activation.jsonl

======================================================================
Summary: 2 found, 0 skipped, 0 missing
======================================================================

COMBINED BENCHMARK SUMMARY

impl                     wl                  p50(ms)  ok
hf_kernels_swiglu        cuda_T128_D1024        0.03  True
hf_kernels_swiglu        cuda_T128_D2048        0.03  True
hf_kernels_swiglu        cuda_T128_D768         0.02  True
hf_kernels_swiglu        cuda_T256_D1024        0.03  True
hf_kernels_swiglu        cuda_T256_D2048        0.03  True
hf_kernels_swiglu        cuda_T256_D768         0.03  True
hf_kernels_swiglu        cuda_T512_D1024        0.03  True
hf_kernels_swiglu        cuda_T512_D2048        0.03  True
hf_kernels_swiglu        cuda_T512_D768         0.03  True
torch_eager              cuda_T128_D1024        0.05  True
torch_eager              cuda_T128_D2048        0.05  True
torch_eager              cuda_T128_D768         0.04  True
torch_eager              cuda_T256_D1024        0.05  True
torch_eager              cuda_T256_D2048        0.05  True
torch_eager              cuda_T256_D768         0.05  True
torch_eager              cuda_T512_D1024        0.05  True
torch_eager              cuda_T512_D2048        0.05  True
torch_eager              cuda_T512_D768         0.05  True

GENERATING COMBINED VISUALIZATION

Loaded 18 records
✓ Visualization saved as latency.svg
Saved latency.png
✓ Visualization saved as latency.svg
✓ SVG visualization ready!

ANALYSIS COMPLETE
Total implementations analyzed: 2

Implementations included:
  ✓ HF Kernels SwiGLU
  ✓ PyTorch SwiGLU

▶ UV Install Logs

Artifacts:

latency.svg