executive_summary
dict | detailed_results
dict | visualizations
list | reproducibility
dict |
|---|---|---|---|
{
"total_tests": 7,
"hfa_wins": 2,
"standard_wins": 0,
"key_findings": [
"HFA outperforms standard attention on pattern_recognition by 253.9%",
"HFA outperforms standard attention on computational_efficiency by 35695.6%"
]
}
|
{
"pattern_recognition": {
"hfa_performance": 0.528468887625606,
"standard_performance": 0.1493469230324704,
"improvement_ratio": 3.5385321431142467,
"sequence_lengths_tested": [
32,
64,
128
],
"metadata": {
"pattern_type": "alternating_with_markers"
}
},
"computational_efficiency": {
"hfa_performance": 0.19553671479225157,
"standard_performance": 0.0005462586879730225,
"improvement_ratio": 357.9562560694839,
"sequence_lengths_tested": [
32,
64,
128,
256
],
"metadata": {
"hfa_speed": "611",
"standard_speed": "467515",
"unit": "tokens_per_second"
}
}
}
|
[] |
{
"torch_version": "2.7.1+cu128",
"cuda_version": "12.8",
"random_seed": 42,
"hardware_info": {
"gpu_name": "NVIDIA GeForce RTX 3090 Ti",
"gpu_memory": 25294995456,
"cpu_count": 32,
"ram_total": 405372833792
}
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.