pirola/moe-strategy-35pct / SCD /stage6alt_eval.json
pirola's picture
download
raw
913 Bytes
{
"acc_norm_sum_gap": -0.025000000000000133,
"bpt_gap": 0.18680916770453448,
"corpus": {
"dataset": "wikitext",
"name": "wikitext",
"num_sequences": 64,
"sequence_length": 2048,
"split": "test",
"subset": "wikitext-2-raw-v1"
},
"lm_eval": {
"arc_easy_limit": 100,
"hellaswag_limit": 200
},
"mode": "thermometer",
"stage": "6alt",
"student_acc_norm_sum": 1.4649999999999999,
"student_arc_easy_acc_norm": 0.74,
"student_bpt": 2.911000749244174,
"student_hellaswag_acc_norm": 0.725,
"teacher_acc_norm_sum": 1.49,
"teacher_arc_easy_acc_norm": 0.75,
"teacher_bpt": 2.7241915815396394,
"teacher_cache": {
"hit": false,
"key": "c213c7c5d390ac8b236729041eb77281f074e112f36690b6120fd47042106f28",
"path": "/cache/ablations/_shared/thermometer_teacher_cache.json"
},
"teacher_hellaswag_acc_norm": 0.74,
"top1_agreement": 0.8253312706947327
}

Xet Storage Details

Size:
913 Bytes
·
Xet hash:
ae9490ed0c3afd5a51c440ceaa2ff0eccbccafa111d3e1c836d63454f627c972

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.