longformer-qasper-document-qa / evaluation /baseline_comparison.json
anmol-unitmole's picture
Upload evaluation/baseline_comparison.json with huggingface_hub
c8129ba verified
Raw
History Blame Contribute Delete
1.4 kB
[
{
"model_name": "BERT truncated 512",
"examples": 200,
"exact_match": 0.015,
"token_f1": 0.07365558071885409,
"evidence_recovered": 0.26,
"evidence_token_recall": 0.4133843754471053,
"average_latency_seconds": 0.028679482998559252,
"p95_latency_seconds": 0.03750500498281326,
"throughput_examples_per_second": 34.86813203885984,
"peak_gpu_memory_mb": 457.69970703125,
"average_window_count": 1.0
},
{
"model_name": "Longformer SQuAD sliding windows",
"examples": 200,
"exact_match": 0.06,
"token_f1": 0.16157914224641423,
"evidence_recovered": 0.3,
"evidence_token_recall": 0.4588296618391169,
"average_latency_seconds": 0.25616708650166403,
"p95_latency_seconds": 0.4064437099703352,
"throughput_examples_per_second": 3.903702125266995,
"peak_gpu_memory_mb": 768.369140625,
"average_window_count": 3.28
},
{
"model_name": "Longformer QASPER fine-tuned",
"examples": 200,
"exact_match": 0.125,
"token_f1": 0.26664085976583785,
"evidence_recovered": 0.49,
"evidence_token_recall": 0.6014359159875786,
"average_latency_seconds": 0.26518189649854323,
"p95_latency_seconds": 0.44097381501051114,
"throughput_examples_per_second": 3.7709964865776326,
"peak_gpu_memory_mb": 768.369140625,
"average_window_count": 3.28
}
]