Slayer149-checkpoints / reports /selection.json
kacperwikiel's picture
Update experiment measurements and status
6669ad2 verified
Raw History Blame Contribute Delete
3.77 kB
{
"objective": "rank_1_efficiency_after_full_confirmation",
"selection_version": "combined-development-v2-before-pilots",
"results": [
{
"source": "fineweb_edu",
"effects": [
{
"seed": 1337,
"arc_delta_pp": 0.17543859649122806,
"grammar_delta_pp": 2.430555555555556,
"wiki_ppl_ratio": 0.9998499487147989,
"wiki_normalized_delta": 0.0026825473015649644,
"combined_dev_delta": 0.869558899782783
},
{
"seed": 2027,
"arc_delta_pp": 0.3508771929824561,
"grammar_delta_pp": 0.6944444444444444,
"wiki_ppl_ratio": 0.9999204588938312,
"wiki_normalized_delta": 0.001421948878126593,
"combined_dev_delta": 0.3489145287683424
}
],
"mean_combined_dev_delta": 0.6092367142755627,
"paired_dev_99pct_interval": [
-0.2904952545353535,
1.7977113731449583
],
"eligible_for_extension": false
},
{
"source": "dclm_edu",
"effects": [
{
"seed": 1337,
"arc_delta_pp": 0.17543859649122806,
"grammar_delta_pp": 1.3888888888888888,
"wiki_ppl_ratio": 0.9999958367943211,
"wiki_normalized_delta": 7.442243152627325e-05,
"combined_dev_delta": 0.5214673026038811
},
{
"seed": 2027,
"arc_delta_pp": 0.3508771929824561,
"grammar_delta_pp": 0.0,
"wiki_ppl_ratio": 0.9998027158537773,
"wiki_normalized_delta": 0.003527037797255739,
"combined_dev_delta": 0.11813474359323728
}
],
"mean_combined_dev_delta": 0.3198010230985592,
"paired_dev_99pct_interval": [
-0.40827828976693786,
1.1902993174455379
],
"eligible_for_extension": false
},
{
"source": "cosmopedia",
"effects": [
{
"seed": 1337,
"arc_delta_pp": 0.3508771929824561,
"grammar_delta_pp": 1.7361111111111112,
"wiki_ppl_ratio": 1.0022786467949207,
"wiki_normalized_delta": -0.04068719618935495,
"combined_dev_delta": 0.6821003693014039
},
{
"seed": 2027,
"arc_delta_pp": -0.8771929824561403,
"grammar_delta_pp": 0.3472222222222222,
"wiki_ppl_ratio": 1.0025389677110237,
"wiki_normalized_delta": -0.04532956111627206,
"combined_dev_delta": -0.1917667737833967
}
],
"mean_combined_dev_delta": 0.24516679775900363,
"paired_dev_99pct_interval": [
-0.7168184605255966,
1.2868486934705048
],
"eligible_for_extension": false
},
{
"source": "finemath",
"effects": [
{
"seed": 1337,
"arc_delta_pp": -0.17543859649122806,
"grammar_delta_pp": 2.083333333333333,
"wiki_ppl_ratio": 0.9999358419833166,
"wiki_normalized_delta": 0.0011469380111162695,
"combined_dev_delta": 0.6363472249510737
},
{
"seed": 2027,
"arc_delta_pp": 0.0,
"grammar_delta_pp": 0.6944444444444444,
"wiki_ppl_ratio": 0.9999406018055441,
"wiki_normalized_delta": 0.0010618452438650554,
"combined_dev_delta": 0.23183542989610317
}
],
"mean_combined_dev_delta": 0.4340913274235884,
"paired_dev_99pct_interval": [
-0.26400808778108986,
1.3345603820044878
],
"eligible_for_extension": false
}
],
"selected_source": null,
"note": "Development-only surrogate; narrow grammar proxy; interval conditional on fixed WikiText validation corpus. Two seeds do not establish population variance. No official GLINT scores."
}