Jev-Style-Cascade-9B / validation /freeze_result.json
chaoliangUNSW's picture
Jev-Style-Cascade-9B: frozen cascade (Jev-Style-2B-Decision-v3 -> JevK5-9B), calibration record and evaluation
8c09135 verified
Raw History Blame Contribute Delete
3.22 kB
{
"calibration_sha256": "580962e4b2bc86b32b42731eda42536f57321c7d1ca9ac1e3b970bf7dbd7e2d2",
"questions": 4992,
"requests": 3758,
"predeclaration_sha256": "e427a671bbe6cc3a07c3c1819125f7169b6d547fc37d067f5f3ccff77ad4e0a6",
"singles": {
"0.8b-v3_bfloat16": {
"accuracy": 0.578926282051282,
"answered": 1.0,
"median_ms_warm": 46.29,
"median_ms_first_pass": 45.727
},
"0.8b-v3_float32": {
"accuracy": 0.5791266025641025,
"answered": 1.0,
"median_ms_warm": 42.731,
"median_ms_first_pass": 42.648
},
"0.8b-v3_graph": {
"accuracy": 0.5791266025641025,
"answered": 1.0,
"median_ms_warm": 11.41,
"median_ms_first_pass": 11.373
},
"2b-v3_bfloat16": {
"accuracy": 0.6606570512820513,
"answered": 1.0,
"median_ms_warm": 92.962,
"median_ms_first_pass": 93.484
},
"2b-v3_float32": {
"accuracy": 0.6618589743589743,
"answered": 1.0,
"median_ms_warm": 89.962,
"median_ms_first_pass": 89.226
},
"2b-v3_graph": {
"accuracy": 0.6618589743589743,
"answered": 1.0,
"median_ms_warm": 13.97,
"median_ms_first_pass": 13.958
},
"jevk5_9b": {
"accuracy": 0.7409855769230769,
"answered": 1.0,
"median_ms_warm": 28.779,
"median_ms_first_pass": 28.754
}
},
"path_0.8b-v3": {
"median_ms_float32": 42.731,
"bfloat16": {
"flips": 32,
"flip_rate": 0.00641,
"median_ms": 46.29,
"usable": false
},
"graph": {
"flips": 0,
"flip_rate": 0.0,
"median_ms": 11.41,
"usable": true
},
"chosen": "graph"
},
"path_2b-v3": {
"median_ms_float32": 89.962,
"bfloat16": {
"flips": 38,
"flip_rate": 0.00761,
"median_ms": 92.962,
"usable": false
},
"graph": {
"flips": 0,
"flip_rate": 0.0,
"median_ms": 13.97,
"usable": true
},
"chosen": "graph"
},
"speed_gate": {
"2b_median_ms": 13.97,
"jevk5_median_ms": 28.779,
"ratio": 0.4854,
"max_ratio": 0.6,
"pass": true
},
"jevk5_alone": {
"accuracy": 0.7409855769230769,
"cost_ms": 28.779,
"top_share": 1.0,
"calls_share": [
1.0
],
"answered_by": [
1.0
]
},
"accuracy_floor": 0.7309855769230769,
"glue2": {
"tiers": [
"2b-v3_graph",
"jevk5_9b"
],
"parity_mismatches": 0,
"thresholds": [
0.75
],
"accuracy": 0.7311698717948718,
"cost_ms": 29.93706093666844,
"top_share": 0.5723157051282052,
"calls_share": [
1.0,
0.5548163916977116
],
"answered_by": [
0.4276842948717949,
0.5723157051282052
]
},
"glue3": {
"tiers": [
"0.8b-v3_graph",
"2b-v3_graph",
"jevk5_9b"
],
"parity_mismatches": 0,
"thresholds": [
0.86,
0.88
],
"accuracy": 0.7311698717948718,
"cost_ms": 39.96178445981906,
"top_share": 0.6444310897435898,
"calls_share": [
1.0,
0.7698243746673763,
0.6184140500266099
],
"answered_by": [
0.21314102564102563,
0.1424278846153846,
0.6444310897435898
]
}
}