vh
a262477a61
stacks/intern-decision: compose (GPU 1, :8033, hard VRAM cap as the single .env knob, healthcheck, Homepage group 'AI - Eval & Retrieval'), .env.example and README. dns: intern-decision.fv.internal -> fv-ml1 (synced to ana/esh/nh3). acceptance on fv-ml1 GPU 3, 3 fresh processes: bit-identical to the Jev bench's native rows (pooled 240/259, Wyrd 79/84, 0/560 flips, Δp 0), negative control 10/122/14, 0 flips across restarts; largest accepted request 200 at a 10,134 MiB card peak under a 9.25 GiB cap; 503 and recovery proven at a tight cap. GPU 1 deploy held: nvidia-smi Free on GPU 1 is 15,442 MiB.
1486 lines
20 KiB
JSON
1486 lines
20 KiB
JSON
{
|
|
"ours": {
|
|
"r1": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
17,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
},
|
|
"a_vs_a_in_process_authored144": {
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
},
|
|
"r2": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
17,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
},
|
|
"a_vs_a_in_process_authored144": {
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
},
|
|
"r3": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
78,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
17,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
},
|
|
"a_vs_a_in_process_authored144": {
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
},
|
|
"oom": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 0,
|
|
"same_top_as_unrotated": 0,
|
|
"follows_description": 0,
|
|
"vs_original_gold": 0
|
|
},
|
|
"a_vs_a_in_process_authored144": {
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
}
|
|
}
|
|
},
|
|
"bench": {
|
|
"r1": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
}
|
|
},
|
|
"r2": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
16,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
}
|
|
},
|
|
"r3": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
16,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
}
|
|
},
|
|
"r4": {
|
|
"scores": {
|
|
"single/pooled": [
|
|
240,
|
|
259
|
|
],
|
|
"single/authored144": [
|
|
132,
|
|
144
|
|
],
|
|
"single/perturbations108": [
|
|
100,
|
|
108
|
|
],
|
|
"single/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"single/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"single/wyrd": [
|
|
79,
|
|
84
|
|
],
|
|
"single/wyrd:place": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"single/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"single/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"single/failures": 0,
|
|
"rotations/pooled": [
|
|
236,
|
|
259
|
|
],
|
|
"rotations/authored144": [
|
|
130,
|
|
144
|
|
],
|
|
"rotations/perturbations108": [
|
|
99,
|
|
108
|
|
],
|
|
"rotations/cicada-w1": [
|
|
29,
|
|
31
|
|
],
|
|
"rotations/cicada-w2": [
|
|
22,
|
|
31
|
|
],
|
|
"rotations/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"rotations/wyrd:place": [
|
|
18,
|
|
21
|
|
],
|
|
"rotations/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/wyrd:exit2": [
|
|
19,
|
|
21
|
|
],
|
|
"rotations/failures": 0,
|
|
"multifield/pooled": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd": [
|
|
77,
|
|
84
|
|
],
|
|
"multifield/wyrd:place": [
|
|
16,
|
|
21
|
|
],
|
|
"multifield/wyrd:place2": [
|
|
21,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/wyrd:exit2": [
|
|
20,
|
|
21
|
|
],
|
|
"multifield/failures": 0
|
|
},
|
|
"negative": {
|
|
"n": 144,
|
|
"same_top_as_unrotated": 10,
|
|
"follows_description": 122,
|
|
"vs_original_gold": 14
|
|
}
|
|
}
|
|
},
|
|
"vs_bench_row_by_row": {
|
|
"r1": {
|
|
"single": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"negative": {
|
|
"bench_repeat": "r1",
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"multifield": {
|
|
"bench_repeat": "r2",
|
|
"rows": 84,
|
|
"top_differs": 1,
|
|
"flipped_ids": [
|
|
"wyrd-scene-fletcher/f-leave-town/place"
|
|
],
|
|
"max_dp": 0.2942167277107136
|
|
}
|
|
},
|
|
"r2": {
|
|
"single": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"negative": {
|
|
"bench_repeat": "r1",
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"multifield": {
|
|
"bench_repeat": "r2",
|
|
"rows": 84,
|
|
"top_differs": 1,
|
|
"flipped_ids": [
|
|
"wyrd-scene-fletcher/f-leave-town/place"
|
|
],
|
|
"max_dp": 0.2942167277107136
|
|
}
|
|
},
|
|
"r3": {
|
|
"single": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"negative": {
|
|
"bench_repeat": "r1",
|
|
"rows": 144,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"multifield": {
|
|
"bench_repeat": "r2",
|
|
"rows": 84,
|
|
"top_differs": 1,
|
|
"flipped_ids": [
|
|
"wyrd-scene-fletcher/f-leave-town/place"
|
|
],
|
|
"max_dp": 0.2942167277107136
|
|
}
|
|
},
|
|
"oom": {
|
|
"single": {
|
|
"bench_repeat": "r1",
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"bench_repeat": "r1",
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
},
|
|
"negative": {
|
|
"bench_repeat": "r1",
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
},
|
|
"multifield": {
|
|
"bench_repeat": "r2",
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
}
|
|
}
|
|
},
|
|
"across_restarts": {
|
|
"oom~r1": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
}
|
|
},
|
|
"oom~r2": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
}
|
|
},
|
|
"oom~r3": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 0,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": null
|
|
}
|
|
},
|
|
"r1~r2": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
},
|
|
"r1~r3": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
},
|
|
"r2~r3": {
|
|
"single": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
},
|
|
"rotations": {
|
|
"rows": 560,
|
|
"top_differs": 0,
|
|
"flipped_ids": [],
|
|
"max_dp": 0.0
|
|
}
|
|
}
|
|
},
|
|
"summary": {
|
|
"multifield/pooled": {
|
|
"ours_median": 78,
|
|
"ours_all": [
|
|
78,
|
|
78,
|
|
78
|
|
],
|
|
"bench_all": [
|
|
77,
|
|
77,
|
|
77
|
|
],
|
|
"n": 84
|
|
},
|
|
"multifield/wyrd": {
|
|
"ours_median": 78,
|
|
"ours_all": [
|
|
78,
|
|
78,
|
|
78
|
|
],
|
|
"bench_all": [
|
|
77,
|
|
77,
|
|
77
|
|
],
|
|
"n": 84
|
|
},
|
|
"multifield/wyrd:exit": {
|
|
"ours_median": 20,
|
|
"ours_all": [
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"bench_all": [
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"n": 21
|
|
},
|
|
"multifield/wyrd:exit2": {
|
|
"ours_median": 20,
|
|
"ours_all": [
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"bench_all": [
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"n": 21
|
|
},
|
|
"multifield/wyrd:place": {
|
|
"ours_median": 17,
|
|
"ours_all": [
|
|
17,
|
|
17,
|
|
17
|
|
],
|
|
"bench_all": [
|
|
16,
|
|
16,
|
|
16
|
|
],
|
|
"n": 21
|
|
},
|
|
"multifield/wyrd:place2": {
|
|
"ours_median": 21,
|
|
"ours_all": [
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"bench_all": [
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"n": 21
|
|
},
|
|
"rotations/authored144": {
|
|
"ours_median": 130,
|
|
"ours_all": [
|
|
130,
|
|
130,
|
|
130
|
|
],
|
|
"bench_all": [
|
|
130,
|
|
130,
|
|
130,
|
|
130
|
|
],
|
|
"n": 144
|
|
},
|
|
"rotations/cicada-w1": {
|
|
"ours_median": 29,
|
|
"ours_all": [
|
|
29,
|
|
29,
|
|
29
|
|
],
|
|
"bench_all": [
|
|
29,
|
|
29,
|
|
29,
|
|
29
|
|
],
|
|
"n": 31
|
|
},
|
|
"rotations/cicada-w2": {
|
|
"ours_median": 22,
|
|
"ours_all": [
|
|
22,
|
|
22,
|
|
22
|
|
],
|
|
"bench_all": [
|
|
22,
|
|
22,
|
|
22,
|
|
22
|
|
],
|
|
"n": 31
|
|
},
|
|
"rotations/perturbations108": {
|
|
"ours_median": 99,
|
|
"ours_all": [
|
|
99,
|
|
99,
|
|
99
|
|
],
|
|
"bench_all": [
|
|
99,
|
|
99,
|
|
99,
|
|
99
|
|
],
|
|
"n": 108
|
|
},
|
|
"rotations/pooled": {
|
|
"ours_median": 236,
|
|
"ours_all": [
|
|
236,
|
|
236,
|
|
236
|
|
],
|
|
"bench_all": [
|
|
236,
|
|
236,
|
|
236,
|
|
236
|
|
],
|
|
"n": 259
|
|
},
|
|
"rotations/wyrd": {
|
|
"ours_median": 77,
|
|
"ours_all": [
|
|
77,
|
|
77,
|
|
77
|
|
],
|
|
"bench_all": [
|
|
77,
|
|
77,
|
|
77,
|
|
77
|
|
],
|
|
"n": 84
|
|
},
|
|
"rotations/wyrd:exit": {
|
|
"ours_median": 19,
|
|
"ours_all": [
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"bench_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"n": 21
|
|
},
|
|
"rotations/wyrd:exit2": {
|
|
"ours_median": 19,
|
|
"ours_all": [
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"bench_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"n": 21
|
|
},
|
|
"rotations/wyrd:place": {
|
|
"ours_median": 18,
|
|
"ours_all": [
|
|
18,
|
|
18,
|
|
18
|
|
],
|
|
"bench_all": [
|
|
18,
|
|
18,
|
|
18,
|
|
18
|
|
],
|
|
"n": 21
|
|
},
|
|
"rotations/wyrd:place2": {
|
|
"ours_median": 21,
|
|
"ours_all": [
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"bench_all": [
|
|
21,
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"n": 21
|
|
},
|
|
"single/authored144": {
|
|
"ours_median": 132.0,
|
|
"ours_all": [
|
|
132,
|
|
132,
|
|
132,
|
|
132
|
|
],
|
|
"bench_all": [
|
|
132,
|
|
132,
|
|
132,
|
|
132
|
|
],
|
|
"n": 144
|
|
},
|
|
"single/cicada-w1": {
|
|
"ours_median": 29.0,
|
|
"ours_all": [
|
|
29,
|
|
29,
|
|
29,
|
|
29
|
|
],
|
|
"bench_all": [
|
|
29,
|
|
29,
|
|
29,
|
|
29
|
|
],
|
|
"n": 31
|
|
},
|
|
"single/cicada-w2": {
|
|
"ours_median": 22.0,
|
|
"ours_all": [
|
|
22,
|
|
22,
|
|
22,
|
|
22
|
|
],
|
|
"bench_all": [
|
|
22,
|
|
22,
|
|
22,
|
|
22
|
|
],
|
|
"n": 31
|
|
},
|
|
"single/perturbations108": {
|
|
"ours_median": 100.0,
|
|
"ours_all": [
|
|
100,
|
|
100,
|
|
100,
|
|
100
|
|
],
|
|
"bench_all": [
|
|
100,
|
|
100,
|
|
100,
|
|
100
|
|
],
|
|
"n": 108
|
|
},
|
|
"single/pooled": {
|
|
"ours_median": 240.0,
|
|
"ours_all": [
|
|
240,
|
|
240,
|
|
240,
|
|
240
|
|
],
|
|
"bench_all": [
|
|
240,
|
|
240,
|
|
240,
|
|
240
|
|
],
|
|
"n": 259
|
|
},
|
|
"single/wyrd": {
|
|
"ours_median": 79.0,
|
|
"ours_all": [
|
|
79,
|
|
79,
|
|
79,
|
|
79
|
|
],
|
|
"bench_all": [
|
|
79,
|
|
79,
|
|
79,
|
|
79
|
|
],
|
|
"n": 84
|
|
},
|
|
"single/wyrd:exit": {
|
|
"ours_median": 19.0,
|
|
"ours_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"bench_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"n": 21
|
|
},
|
|
"single/wyrd:exit2": {
|
|
"ours_median": 20.0,
|
|
"ours_all": [
|
|
20,
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"bench_all": [
|
|
20,
|
|
20,
|
|
20,
|
|
20
|
|
],
|
|
"n": 21
|
|
},
|
|
"single/wyrd:place": {
|
|
"ours_median": 19.0,
|
|
"ours_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"bench_all": [
|
|
19,
|
|
19,
|
|
19,
|
|
19
|
|
],
|
|
"n": 21
|
|
},
|
|
"single/wyrd:place2": {
|
|
"ours_median": 21.0,
|
|
"ours_all": [
|
|
21,
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"bench_all": [
|
|
21,
|
|
21,
|
|
21,
|
|
21
|
|
],
|
|
"n": 21
|
|
}
|
|
}
|
|
}
|