tokenfold-select-v2 / evidence /development-compression.json
snchimata's picture
Release Tokenfold Select v2 ONNX and Safetensors preview
78736d9 verified
Raw History Blame Contribute Delete
10.9 kB
{
"all_successful_Select_payloads_exact_and_required": true,
"candidate_changed_payloads_vs_FP32": {
"code:0.5": 6,
"logs:0.25": 6,
"logs:0.5": 6
},
"candidate_execution_profile": "cuda-fp16-cast-v1",
"candidate_revision": "4dbdb5071e7d6316d8eb70dc161b1f2f77f555012f6616d7fa34311188e24cec",
"cold_latency_measured": false,
"complete_original_denominator": 180,
"compression_clock_excludes": "Model loading/warmup, file IO and offline audit",
"contiguous_pipeline_latency_measured": false,
"controls_payloads_and_failures_match_previous_FP32": true,
"controls_precision": "FP32 unchanged",
"counts": {
"completed": 132,
"healthy_budget_refusal_not_compression_pass": 48
},
"downstream_quality_measured": false,
"failed_predecessor_preserved_not_pooled": true,
"full_denominator_audit_sha256": "8ec14628c27041b3b0f4c47d1b5073f98064c5f082b5c415c1c4303ac2f62414",
"independent_post_audit_sha256": "fcb250093d5ab2dd185605395052cdb4cbbf9f34ce85830f45ff18c292118426",
"kind": "Already-exposed-source warm DEV evidence only",
"latency_superiority_claim": false,
"native_allocator_sha256": "d2dc740ce81c313c39f2493fe62561db660f82da300913bc4f1aa33c737c1793",
"per_workload_arm_budget": [
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.25,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "code"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.5,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "code"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 447.7370000004157,
"failure_aware_empirical_p95_ms": 451.22790000095847,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "code"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 451.9159999999829,
"failure_aware_empirical_p95_ms": 466.93560000130674,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "code"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 271.70590000059747,
"failure_aware_empirical_p95_ms": 281.2317999996594,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "code"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 268.71290000053705,
"failure_aware_empirical_p95_ms": 282.4414999995497,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "code"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.25,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "json"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.5,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "json"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 209.81700000083947,
"failure_aware_empirical_p95_ms": 220.7605000003241,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "json"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 207.644000000073,
"failure_aware_empirical_p95_ms": 215.1427999997395,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "json"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 149.75730000151088,
"failure_aware_empirical_p95_ms": 164.40569999940635,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "json"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 150.19930000016757,
"failure_aware_empirical_p95_ms": 161.9959999989078,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "json"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.25,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "logs"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.5,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "logs"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 3740.212100001372,
"failure_aware_empirical_p95_ms": 3808.8632999988477,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "logs"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 3760.490300001038,
"failure_aware_empirical_p95_ms": 3788.0289999993693,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "logs"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 1822.0676999990246,
"failure_aware_empirical_p95_ms": 1840.2031000005081,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "logs"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 1842.7443999989919,
"failure_aware_empirical_p95_ms": 1862.3588999998901,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "logs"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.25,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "multi-turn"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 94.57739999925252,
"failure_aware_empirical_p95_ms": 97.53429999909713,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "multi-turn"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 560.2236000013363,
"failure_aware_empirical_p95_ms": 564.5395999999892,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "multi-turn"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 563.0398000012065,
"failure_aware_empirical_p95_ms": 568.8818000016909,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "multi-turn"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 265.19890000054147,
"failure_aware_empirical_p95_ms": 267.4618000000919,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "multi-turn"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 267.0509999989008,
"failure_aware_empirical_p95_ms": 271.35810000072524,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "multi-turn"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": "infinite",
"failure_aware_empirical_p95_ms": "infinite",
"ratio": 0.25,
"successful_requests": 0,
"unique_payload_hashes": 0,
"workload": "retrieval"
},
{
"arm": "kompress-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 2733.64120000042,
"failure_aware_empirical_p95_ms": 2751.18029999976,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "retrieval"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 2045.554300000731,
"failure_aware_empirical_p95_ms": 2058.284700000513,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "retrieval"
},
{
"arm": "select-v1",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 2046.0320000001957,
"failure_aware_empirical_p95_ms": 2070.1850000004924,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "retrieval"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 763.7071000008291,
"failure_aware_empirical_p95_ms": 783.8761000002705,
"ratio": 0.25,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "retrieval"
},
{
"arm": "select-v2",
"attempted_requests": 6,
"failure_aware_empirical_p50_ms": 773.4466999991128,
"failure_aware_empirical_p95_ms": 803.6979000007705,
"ratio": 0.5,
"successful_requests": 6,
"unique_payload_hashes": 1,
"workload": "retrieval"
}
],
"protocol_sha256": "4429ace708e98261769412d6b7b7011c8571b89bc3c36796d49103bd0085f41a",
"qualification": false,
"repeats_are_not_independent_clusters": true,
"repeats_per_source_arm_budget": 6,
"report_sha256": "c9175773caebe0ce4492e671cc0c5c41a177ab3868120350bba44b6758cc5104",
"source_units": 5,
"source_units_per_family": 1,
"timing_profile": "caller-grouping-inclusive-v3"
}