laya-coreml / reports /benchmark-jev-coreml.json
alexwengg's picture
Add five-task Jev comparison reports
40e4036 verified
Raw
History Blame Contribute Delete
2.54 kB
{
"chip" : "Apple M5 Pro",
"elapsed_s" : 2.7073659896850586,
"latency_ms" : {
"p50" : 4.4866669999999997,
"p95" : 9.6365829999999999
},
"lengths" : [
128,
256,
512,
1024
],
"load_s" : 2.0176939964294434,
"questions" : 500,
"suites" : {
"emotion" : {
"accuracy" : 0.57999999999999996,
"buckets" : {
"L128" : 100
},
"dropped" : 0,
"latency_ms" : {
"p50" : 4.0849580000000003,
"p95" : 4.5919999999999996
},
"max_probability_delta_vs_reference" : 0.014793992042541504,
"n" : 100,
"reference_argmax_agreement" : 1,
"reference_compared" : 100,
"state_truncated" : 0
},
"news_topic" : {
"accuracy" : 0.96999999999999997,
"buckets" : {
"L128" : 86,
"L256" : 14
},
"dropped" : 0,
"latency_ms" : {
"p50" : 5.5224159999999998,
"p95" : 9.2874169999999996
},
"max_probability_delta_vs_reference" : 0.010490596294403076,
"n" : 100,
"reference_argmax_agreement" : 1,
"reference_compared" : 100,
"state_truncated" : 0
},
"prompt_injection" : {
"accuracy" : 0.65000000000000002,
"buckets" : {
"L128" : 96,
"L256" : 4
},
"dropped" : 0,
"latency_ms" : {
"p50" : 4.3983749999999997,
"p95" : 6.5922919999999996
},
"max_probability_delta_vs_reference" : 0.02795100212097168,
"n" : 100,
"reference_argmax_agreement" : 0.98999999999999999,
"reference_compared" : 100,
"state_truncated" : 0
},
"review_stars" : {
"accuracy" : 0.34999999999999998,
"buckets" : {
"L128" : 28,
"L256" : 35,
"L512" : 26,
"L1024" : 11
},
"dropped" : 0,
"latency_ms" : {
"p50" : 5.9012919999999998,
"p95" : 18.642083
},
"max_probability_delta_vs_reference" : 0.0017490386962890625,
"n" : 100,
"reference_argmax_agreement" : 1,
"reference_compared" : 100,
"state_truncated" : 0
},
"sms_spam" : {
"accuracy" : 0.57999999999999996,
"buckets" : {
"L128" : 100
},
"dropped" : 0,
"latency_ms" : {
"p50" : 4.1685420000000004,
"p95" : 5.0857080000000003
},
"max_probability_delta_vs_reference" : 0.020002603530883789,
"n" : 100,
"reference_argmax_agreement" : 1,
"reference_compared" : 100,
"state_truncated" : 0
}
}
}