[{"family":"256x8","key":"pt5b_sft3_full","label":"5B PT targets + all ordinary SFT \u00d73","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5124293190450496,"token_accuracy":0.8052098751068115,"format_rate":0.9813766891891892,"pass_at_k":{"1":0.19113175675675675,"10":0.5198494073494074,"11":0.5311584249084249,"12":0.5412169587169587,"13":0.5502606177606179,"14":0.5584740990990992,"15":0.5660050675675675,"16":0.572972972972973,"2":0.2906362612612613,"3":0.352566361003861,"4":0.39583308583308585,"5":0.4283175861300862,"6":0.4538944163944164,"7":0.47472683566433566,"8":0.49213596463596465,"9":0.5069811944811945},"evaluated_prompts":1480},{"family":"256x8","key":"pt2p5b_sft3_first","label":"First 2.5B PT targets + first ordinary-SFT half","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5250549227713442,"token_accuracy":0.7984380722045898,"format_rate":0.9799831081081081,"pass_at_k":{"1":0.13851351351351351,"10":0.45111349798849804,"11":0.4636634058509058,"12":0.47486560736560735,"13":0.4849481177606177,"14":0.49409346846846847,"15":0.5024493243243243,"16":0.5101351351351351,"2":0.2214414414414414,"3":0.2783506274131274,"4":0.3204829967329967,"5":0.3533084595584596,"6":0.37985764235764236,"7":0.4019428038178038,"8":0.42071095571095574,"9":0.4369244353619354},"evaluated_prompts":1480},{"family":"256x8","key":"pt2p5b_sft3_second_plain","label":"Second 2.5B PT targets + second ordinary-SFT half","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5074321794623302,"token_accuracy":0.8054999113082886,"format_rate":0.9839527027027027,"pass_at_k":{"1":0.18234797297297298,"10":0.5157620251370252,"11":0.5280419327294327,"12":0.5390781853281853,"13":0.5490830115830115,"14":0.5582150900900901,"15":0.5665962837837838,"16":0.5743243243243243,"2":0.28094031531531527,"3":0.3433373552123552,"4":0.3871164983664984,"5":0.4201479123354124,"6":0.4463715338715338,"7":0.46795661264411254,"8":0.48620366870366877,"9":0.5019494188244188},"evaluated_prompts":1480},{"family":"256x8","key":"rl_pt5b_source","label":"RL1500 from the complete 5B parent","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5275179100033462,"token_accuracy":0.8032954931259155,"format_rate":0.9990287162162163,"pass_at_k":{"1":0.36925675675675673,"10":0.5266008484758484,"11":0.5317723802098802,"12":0.5364244876744877,"13":0.5406455115830116,"14":0.5445045045045045,"15":0.5480574324324324,"16":0.5513513513513514,"2":0.4249774774774775,"3":0.4534326737451737,"4":0.47238416988416987,"5":0.48642615830115826,"6":0.49747524435024437,"7":0.5065320237195238,"8":0.5141824166824167,"9":0.5207922061047061},"evaluated_prompts":1480},{"family":"256x8","key":"rl_pt2p5b_source","label":"RL1500 from the first 2.5B parent","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5402892917952817,"token_accuracy":0.796013355255127,"format_rate":0.9984375,"pass_at_k":{"1":0.30789695945945944,"10":0.46912310662310663,"11":0.47459023240273235,"12":0.47951106326106324,"13":0.48397683397683394,"14":0.48805743243243244,"15":0.4918074324324324,"16":0.49527027027027026,"2":0.3639076576576577,"3":0.39317205598455596,"4":0.4126288238788238,"5":0.4270723505098505,"6":0.43852085414585407,"7":0.4479808282933283,"8":0.456019236019236,"9":0.462988152050652},"evaluated_prompts":1480},{"family":"256x8","key":"trace03_shuffled","label":"One shuffled pass over full-parent successful RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5370426717426243,"token_accuracy":0.8038429021835327,"format_rate":0.9997043918918919,"pass_at_k":{"1":0.36114864864864865,"10":0.5458359208359208,"11":0.552702083952084,"12":0.5590280665280665,"13":0.5649034749034749,"14":0.5703941441441441,"15":0.5755489864864864,"16":0.5804054054054054,"2":0.4259009009009008,"3":0.4586414092664093,"4":0.48032967032967033,"5":0.4965521668646668,"6":0.5095667001917003,"7":0.520488033925534,"8":0.5299405174405174,"9":0.538306322056322},"evaluated_prompts":1480},{"family":"256x8","key":"trace04_chronological","label":"One chronological pass over full-parent successful RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5461314942062178,"token_accuracy":0.8028436899185181,"format_rate":0.9997043918918919,"pass_at_k":{"1":0.3769847972972973,"10":0.5327352546102546,"11":0.5379919067419068,"12":0.5427858627858628,"13":0.5472019787644787,"14":0.5513063063063064,"15":0.555152027027027,"16":0.5587837837837838,"2":0.43316441441441444,"3":0.46125844594594595,"4":0.4796848084348085,"5":0.49327046827046833,"6":0.5039816602316602,"7":0.5128148625023625,"8":0.5203388353388353,"9":0.5269030547155547},"evaluated_prompts":1480},{"family":"256x8","key":"trace05_shuffled","label":"One shuffled pass over half-parent successful RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5529008129733451,"token_accuracy":0.7965162992477417,"format_rate":0.999535472972973,"pass_at_k":{"1":0.3027027027027027,"10":0.4863351175851176,"11":0.4922537063162063,"12":0.4975694238194239,"13":0.5024022683397683,"14":0.5068468468468469,"15":0.5109797297297297,"16":0.5148648648648648,"2":0.3663963963963964,"3":0.400141167953668,"4":0.4227071577071577,"5":0.439384405009405,"6":0.4524451055701056,"7":0.4630848374598375,"8":0.47200681450681453,"9":0.4796569055944056},"evaluated_prompts":1480},{"family":"256x8","key":"trace06_chronological","label":"One chronological pass over half-parent successful RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5603667393549017,"token_accuracy":0.7952938079833984,"format_rate":0.9995777027027027,"pass_at_k":{"1":0.30743243243243246,"10":0.46717108904608906,"11":0.47320299351549355,"12":0.47869616869616866,"13":0.483730694980695,"14":0.4883671171171171,"15":0.49265202702702704,"16":0.4966216216216216,"2":0.3609684684684685,"3":0.38925916988416986,"4":0.4084281259281259,"5":0.42296090733590735,"6":0.43468288468288463,"7":0.4445229517104517,"8":0.453015498015498,"9":0.46049216830466827},"evaluated_prompts":1480},{"family":"256x8","key":"trace05_then_pt_second","label":"Second 2.5B PT + SFT half after shuffled trace training","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5073571630482313,"token_accuracy":0.8055938482284546,"format_rate":0.9844594594594595,"pass_at_k":{"1":0.17723817567567568,"10":0.5073753948753948,"11":0.5187252499752499,"12":0.5288030888030888,"13":0.5378318050193049,"14":0.5459797297297297,"15":0.5533783783783783,"16":0.5601351351351351,"2":0.27552364864864864,"3":0.338296332046332,"4":0.38228615978615976,"5":0.41520208395208397,"6":0.44103033790533797,"7":0.46201852201852195,"8":0.47953133203133197,"9":0.4944508363258363},"evaluated_prompts":1480},{"family":"256x8","key":"trace06_then_pt_second","label":"Second 2.5B PT + SFT half after chronological trace training","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5075848827646329,"token_accuracy":0.8055107593536377,"format_rate":0.9833614864864865,"pass_at_k":{"1":0.17820945945945946,"10":0.5118163086913087,"11":0.5239018723393724,"12":0.5347089397089397,"13":0.544460666023166,"14":0.5533277027027027,"15":0.5614442567567568,"16":0.5689189189189189,"2":0.2758220720720721,"3":0.3383747586872587,"4":0.3826362488862489,"5":0.4160892176517176,"6":0.442569001944002,"7":0.46425823332073335,"8":0.4824962724962725,"9":0.4981524640899641},"evaluated_prompts":1480},{"family":"256x8","key":"trace07_mixed_pt_second","label":"Stable interleaving of second PT, ordinary SFT, and RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5109790352094215,"token_accuracy":0.8041523694992065,"format_rate":0.9980152027027027,"pass_at_k":{"1":0.3349662162162162,"10":0.5194968882468882,"11":0.5258254133254133,"12":0.5315503415503415,"13":0.5367676158301158,"14":0.5415484234234235,"15":0.5459459459459459,"16":0.55,"2":0.4003265765765766,"3":0.433334942084942,"4":0.4552364864864865,"5":0.47156036531036527,"6":0.48453649391149395,"7":0.4952862762237762,"8":0.5044517944517944,"9":0.5124335546210547},"evaluated_prompts":1480},{"family":"256x8","key":"trace08_global_shuffle_pt_second","label":"Global record permutation of second PT, ordinary SFT, and RL traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5097308494736996,"token_accuracy":0.8050708770751953,"format_rate":0.9975506756756757,"pass_at_k":{"1":0.3256756756756757,"10":0.5225065475065475,"11":0.5289350992475993,"12":0.5346688446688447,"13":0.5398274613899614,"14":0.5445045045045045,"15":0.5487753378378378,"16":0.5527027027027027,"2":0.3947635135135135,"3":0.4301918436293436,"4":0.45389664389664386,"5":0.47160708098208104,"6":0.4856366606366606,"7":0.49716316622566625,"8":0.5068782943782943,"9":0.5152256189756189},"evaluated_prompts":1480},{"family":"256x8","key":"rl01_final","label":"Final RL3000","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5323746171850748,"token_accuracy":0.8024080991744995,"format_rate":0.9993243243243243,"pass_at_k":{"1":0.37863175675675675,"10":0.5234978703728703,"11":0.5284349940599941,"12":0.5329339916839917,"13":0.5370789092664092,"14":0.5409346846846848,"15":0.5445523648648649,"16":0.547972972972973,"2":0.4303716216216216,"3":0.4563416988416989,"4":0.4735224235224235,"5":0.4862708518958519,"6":0.49637060574560576,"7":0.5047170312795313,"8":0.5118231168231168,"9":0.5180113045738046},"evaluated_prompts":1480},{"family":"256x8","key":"rl02_final","label":"Final RL3000","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5231766922428465,"token_accuracy":0.8031466007232666,"format_rate":0.9994087837837838,"pass_at_k":{"1":0.3793918918918919,"10":0.533245976995977,"11":0.538634633947134,"12":0.5434682209682209,"13":0.5478378378378378,"14":0.5518130630630631,"15":0.5554476351351352,"16":0.5587837837837838,"2":0.4318355855855856,"3":0.4590082046332046,"4":0.47755643005643006,"5":0.4916337181962182,"6":0.5029271404271405,"7":0.5123031444906444,"8":0.5202766752766752,"9":0.5271812984312985},"evaluated_prompts":1480},{"family":"256x8","key":"rl03_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5461733712666348,"token_accuracy":0.8021718263626099,"format_rate":0.9997466216216216,"pass_at_k":{"1":0.37635135135135134,"10":0.5145612833112834,"11":0.5188522485397485,"12":0.5227138402138403,"13":0.5262210424710424,"14":0.5294313063063063,"15":0.5323902027027027,"16":0.5351351351351351,"2":0.4268806306306306,"3":0.45244932432432433,"4":0.46897572022572015,"5":0.48096367315117317,"6":0.49028142465642466,"7":0.49786542005292,"8":0.5042438942438943,"9":0.509739415989416},"evaluated_prompts":1480},{"family":"256x8","key":"rl04_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.554019728558302,"token_accuracy":0.8013458251953125,"format_rate":0.9997466216216216,"pass_at_k":{"1":0.38488175675675673,"10":0.5191689222939223,"11":0.5232964248589248,"12":0.5269642857142857,"13":0.5302569980694981,"14":0.5332432432432432,"15":0.5359797297297297,"16":0.5385135135135135,"2":0.4336486486486486,"3":0.4582806467181467,"4":0.4743206118206118,"5":0.48606016731016727,"6":0.49525373275373274,"7":0.5027641277641278,"8":0.5090724290724291,"9":0.5144747566622566},"evaluated_prompts":1480},{"family":"256x8","key":"rl05_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5183104446439364,"token_accuracy":0.8039599657058716,"format_rate":0.9987331081081081,"pass_at_k":{"1":0.366089527027027,"10":0.5296803703053703,"11":0.5353329806454807,"12":0.5404755717255717,"13":0.5451978764478764,"14":0.5495664414414414,"15":0.5536317567567568,"16":0.5574324324324325,"2":0.42335585585585583,"3":0.4527256274131274,"4":0.4722965547965548,"5":0.4869216228591228,"6":0.4985411547911548,"7":0.5081389387639388,"8":0.5162977487977488,"9":0.5233937346437346},"evaluated_prompts":1480},{"family":"256x8","key":"rl06_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5175058060627593,"token_accuracy":0.8040586709976196,"format_rate":0.9992398648648648,"pass_at_k":{"1":0.3663006756756757,"10":0.5213069025569026,"11":0.5268916753291752,"12":0.5319969557469558,"13":0.5366831563706563,"14":0.5409966216216215,"15":0.5449746621621622,"16":0.5486486486486486,"2":0.4221846846846846,"3":0.44979488416988417,"4":0.4677940302940303,"5":0.4811113998613999,"6":0.49172693860193856,"7":0.5006199796824797,"8":0.5083269808269809,"9":0.5151593507843507},"evaluated_prompts":1480},{"family":"256x8","key":"rl07_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5202202623991886,"token_accuracy":0.8023027181625366,"format_rate":0.9994932432432433,"pass_at_k":{"1":0.37136824324324325,"10":0.5179534148284148,"11":0.5231033746658746,"12":0.5277394564894565,"13":0.5319461872586873,"14":0.5357882882882883,"15":0.5393158783783784,"16":0.5425675675675675,"2":0.4212162162162163,"3":0.44738537644787646,"4":0.46512065637065636,"5":0.47849795193545186,"6":0.4891933235683236,"7":0.49806635087885087,"8":0.5056204981204981,"9":0.5121770459270458},"evaluated_prompts":1480},{"family":"256x8","key":"rl08_final","label":"Final RL1500","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5190419370586861,"token_accuracy":0.8031327724456787,"format_rate":0.9993243243243243,"pass_at_k":{"1":0.3645692567567568,"10":0.505928784053784,"11":0.5114413176913177,"12":0.5165299227799228,"13":0.5212632722007722,"14":0.5256925675675675,"15":0.5298564189189189,"16":0.5337837837837838,"2":0.41214527027027026,"3":0.4371525096525096,"4":0.454024354024354,"5":0.46676179957429964,"6":0.47703698328698324,"7":0.4856959341334341,"8":0.4932182532182532,"9":0.4998984123984124},"evaluated_prompts":1480},{"family":"256x8","key":"trace09_direct_rl_then_pt_second_v3","label":"PT2 + SFT half from RL1 policy, fresh AdamW","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5078258783167922,"token_accuracy":0.8055297136306763,"format_rate":0.9835726351351352,"pass_at_k":{"1":0.17846283783783784,"10":0.505760962010962,"11":0.5167766495891496,"12":0.5265321502821503,"13":0.5352557915057915,"14":0.5431193693693693,"15":0.5502533783783784,"16":0.5567567567567567,"2":0.27676801801801804,"3":0.3391457528957529,"4":0.382732773982774,"5":0.4152890184140184,"6":0.4407774657774658,"7":0.4614277783027783,"8":0.4785997710997711,"9":0.493176915989416},"evaluated_prompts":1480},{"family":"256x8","key":"trace10_direct_rl_then_pt_replay_second_v2","label":"PT2 + SFT half from RL1 policy with late RL replay","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5085713656066267,"token_accuracy":0.8055074214935303,"format_rate":0.9925253378378378,"pass_at_k":{"1":0.2633868243243243,"10":0.5339570902070901,"11":0.5431993119493119,"12":0.5515592515592516,"13":0.5591686776061776,"14":0.5661261261261261,"15":0.572508445945946,"16":0.5783783783783784,"2":0.3572297297297298,"3":0.40713441119691124,"4":0.4398904811404811,"5":0.4639896421146421,"6":0.48298711423711416,"7":0.4986627646002646,"8":0.5120150045150045,"9":0.5236497708372708},"evaluated_prompts":1480},{"family":"256x8","key":"rl11_12_shared_continuous","label":"Shared RL1500 with PT1 AdamW state","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5397322934730641,"token_accuracy":0.7959716320037842,"format_rate":0.9987753378378378,"pass_at_k":{"1":0.3153293918918919,"10":0.4740577834327834,"11":0.47944525913275915,"12":0.48433917433917434,"13":0.4888416988416988,"14":0.4930292792792793,"15":0.49695945945945946,"16":0.5006756756756757,"2":0.3697747747747748,"3":0.398456805019305,"4":0.4178627116127116,"5":0.43238571676071674,"6":0.443862471987472,"7":0.45326887639387636,"8":0.4611980511980512,"9":0.46803865053865057},"evaluated_prompts":1480},{"family":"256x8","key":"trace11_continuous_then_pt_second","label":"PT2 + SFT half with continuous AdamW, no replay","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5077274095899283,"token_accuracy":0.805488109588623,"format_rate":0.9845439189189189,"pass_at_k":{"1":0.1810810810810811,"10":0.5086600055350055,"11":0.5203002796752797,"12":0.5308033858033858,"13":0.5403764478764479,"14":0.5491722972972973,"15":0.5573057432432432,"16":0.5648648648648649,"2":0.27977477477477475,"3":0.3419980694980695,"4":0.3852598752598752,"5":0.4175052284427285,"6":0.44279057091557095,"7":0.46339692402192395,"8":0.4807046557046557,"9":0.4955963546588546},"evaluated_prompts":1480},{"family":"256x8","key":"trace12_continuous_then_pt_replay_second","label":"PT2 + SFT half with continuous AdamW and late RL replay","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5079516245986893,"token_accuracy":0.8054080009460449,"format_rate":0.9931165540540541,"pass_at_k":{"1":0.2622043918918919,"10":0.5369362900612901,"11":0.5463235385110385,"12":0.5547757647757648,"13":0.5624396718146718,"14":0.5694313063063063,"15":0.5758445945945946,"16":0.5817567567567568,"2":0.3579954954954954,"3":0.408362693050193,"4":0.44125631125631126,"5":0.4655570921195922,"6":0.4848478379728379,"7":0.5008531586656587,"8":0.514519887019887,"9":0.5264222854847855},"evaluated_prompts":1480},{"family":"256x8","key":"rl09_final","label":"Final RL1500 with fresh AdamW","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5191726169911883,"token_accuracy":0.8037725687026978,"format_rate":0.9986064189189189,"pass_at_k":{"1":0.36355574324324325,"10":0.5205846518346519,"11":0.5251969173844173,"12":0.5293321205821206,"13":0.5330791505791506,"14":0.5365033783783784,"15":0.5396537162162162,"16":0.5425675675675675,"2":0.42153153153153156,"3":0.4508542471042471,"4":0.46979915354915347,"5":0.4834363553113553,"6":0.49392558455058455,"7":0.5023654554904555,"8":0.5093843318843319,"9":0.5153720941220942},"evaluated_prompts":1480},{"family":"256x8","key":"rl10_final","label":"Final RL1500 with fresh AdamW after late replay","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5198341243606072,"token_accuracy":0.8034602403640747,"format_rate":0.9992820945945946,"pass_at_k":{"1":0.3658783783783784,"10":0.5081244768744769,"11":0.5130460474210474,"12":0.5175434362934364,"13":0.5216855694980694,"14":0.5255236486486486,"15":0.5290962837837838,"16":0.5324324324324324,"2":0.41769707207207213,"3":0.44315516409266414,"4":0.45964619839619847,"5":0.47184591872091874,"6":0.4815588971838972,"7":0.48964698308448307,"8":0.49658980658980656,"9":0.5026838026838026},"evaluated_prompts":1480},{"family":"256x8","key":"rl11_final","label":"Final RL1500 with continuous AdamW, no replay","kind":"rl","rl_prompt_rows":53225,"log_loss":0.5201870615706012,"token_accuracy":0.8037136793136597,"format_rate":0.9988597972972973,"pass_at_k":{"1":0.36803209459459457,"10":0.5625359437859437,"11":0.5693919537669538,"12":0.5756225868725868,"13":0.5813236003861004,"14":0.5865709459459459,"15":0.5914273648648649,"16":0.595945945945946,"2":0.43683558558558555,"3":0.47174710424710425,"4":0.494707825957826,"5":0.5117604383229384,"6":0.5253446722196722,"7":0.5366647372897373,"8":0.5463908313908313,"9":0.5549278255528255},"evaluated_prompts":1480},{"family":"256x8","key":"rl12_final","label":"Final RL1500 with continuous AdamW after late replay","kind":"rl","rl_prompt_rows":53225,"log_loss":0.518223166032792,"token_accuracy":0.8039038181304932,"format_rate":0.9990287162162163,"pass_at_k":{"1":0.37394425675675674,"10":0.5402617652617653,"11":0.5459705412830413,"12":0.5511037273537274,"13":0.5557601351351351,"14":0.560016891891892,"15":0.5639358108108108,"16":0.5675675675675675,"2":0.4310360360360361,"3":0.4607589285714286,"4":0.4809095634095634,"5":0.49607449138699133,"6":0.5081435456435457,"7":0.5181044462294462,"8":0.5265460215460215,"9":0.5338466854091853},"evaluated_prompts":1480},{"family":"256x8","key":"trace13_pt1_adam_ordered_pt_second","label":"Reset to PT1 weights + PT1 AdamW; ordered PT2 PT/SFT + all RL1 traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5112567177627684,"token_accuracy":0.8040008544921875,"format_rate":0.9976351351351351,"pass_at_k":{"1":0.3356418918918919,"10":0.5187625718875719,"11":0.5245110323235324,"12":0.5297044847044847,"13":0.5344486003861004,"14":0.5388175675675676,"15":0.5428631756756757,"16":0.5466216216216216,"2":0.40135135135135136,"3":0.4349046814671814,"4":0.4570823433323434,"5":0.473366653054153,"6":0.4860517860517861,"7":0.49633735352485353,"8":0.5049369474369474,"9":0.5123091098091098},"evaluated_prompts":1480},{"family":"256x8","key":"trace14_pt1_adam_global_pt_second","label":"Reset to PT1 weights + PT1 AdamW; globally shuffled PT2 PT/SFT/RL1 traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5102192551025809,"token_accuracy":0.805042028427124,"format_rate":0.9978462837837838,"pass_at_k":{"1":0.328125,"10":0.5227622883872883,"11":0.5298137251262252,"12":0.5362782150282149,"13":0.5422659266409267,"14":0.5478603603603603,"15":0.553125,"16":0.5581081081081081,"2":0.39594594594594595,"3":0.4304645270270271,"4":0.45354841104841104,"5":0.4708778833778834,"6":0.4847504691254691,"7":0.49632205632205634,"8":0.5062577962577963,"9":0.5149770246645247},"evaluated_prompts":1480},{"family":"256x8","key":"rl13_final","label":"Final RL1500 with fresh AdamW","kind":"rl","rl_prompt_rows":53225,"log_loss":0.5190891723958282,"token_accuracy":0.802220344543457,"format_rate":0.9992398648648648,"pass_at_k":{"1":0.3781672297297297,"10":0.5287906350406351,"11":0.5342048435798435,"12":0.5391940154440155,"13":0.5438296332046332,"14":0.5481644144144144,"15":0.5522381756756757,"16":0.5560810810810811,"2":0.4314921171171171,"3":0.4585304054054054,"4":0.4763791951291952,"5":0.4895830239580239,"6":0.5000505406755407,"7":0.508754725004725,"8":0.5162441987441988,"9":0.5228515994140994},"evaluated_prompts":1480},{"family":"256x8","key":"rl14_final","label":"Final RL1500 with fresh AdamW","kind":"rl","rl_prompt_rows":53225,"log_loss":0.5197699205505696,"token_accuracy":0.8031389713287354,"format_rate":0.9991976351351352,"pass_at_k":{"1":0.37960304054054056,"10":0.5343849562599562,"11":0.539655263092763,"12":0.5443892931392932,"13":0.5486763996138996,"14":0.5525844594594594,"15":0.5561655405405406,"16":0.5594594594594594,"2":0.43286036036036035,"3":0.4607070463320463,"4":0.47946688446688446,"5":0.4935298794673794,"6":0.5046981734481735,"7":0.5139079923454923,"8":0.5217112917112917,"9":0.5284585853335854},"evaluated_prompts":1480},{"family":"256x8","key":"trace15_direct_rl_all_trace_ordered_continue_adam_pt_second","label":"RL1 weights + RL1 AdamW; ordered PT2 PT/SFT + all RL1 traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5117043960247888,"token_accuracy":0.8040602207183838,"format_rate":0.9975506756756757,"pass_at_k":{"1":0.34083614864864864,"10":0.5224894193644193,"11":0.5292738652113652,"12":0.5354900504900505,"13":0.5412234555984556,"14":0.5465371621621622,"15":0.5514780405405405,"16":0.5560810810810811,"2":0.4040146396396396,"3":0.43576978764478763,"4":0.45716958716958717,"5":0.4733453061578061,"6":0.48635629573129574,"7":0.49726676195426195,"8":0.5066937041937042,"9":0.5150186637686637},"evaluated_prompts":1480},{"family":"256x8","key":"trace16_direct_rl_all_trace_ordered_fresh_adam_pt_second","label":"RL1 weights + fresh AdamW; ordered PT2 PT/SFT + all RL1 traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5104697200463045,"token_accuracy":0.804161548614502,"format_rate":0.9978885135135135,"pass_at_k":{"1":0.3403716216216216,"10":0.5245608276858277,"11":0.5307812964062963,"12":0.5363498663498664,"13":0.5413803088803089,"14":0.5459628378378378,"15":0.550168918918919,"16":0.5540540540540541,"2":0.4054898648648649,"3":0.4381141409266409,"4":0.45991758241758246,"5":0.4763018513018513,"6":0.48940603315603315,"7":0.5002852721602722,"8":0.5095381570381571,"9":0.5175432337932339},"evaluated_prompts":1480},{"family":"256x8","key":"rl15_final","label":"Final RL1500 with fresh AdamW","kind":"rl","rl_prompt_rows":53225,"log_loss":0.5217091242832407,"token_accuracy":0.8021767139434814,"format_rate":0.9993665540540541,"pass_at_k":{"1":0.38809121621621623,"10":0.5343109593109593,"11":0.5390318718443718,"12":0.5432528957528958,"13":0.547065637065637,"14":0.5505461711711712,"15":0.5537584459459459,"16":0.5567567567567567,"2":0.4409290540540541,"3":0.467405888030888,"4":0.48471190971190975,"5":0.4974592552717552,"6":0.5075281137781138,"7":0.5158366211491212,"8":0.5228858753858754,"9":0.5289773317898318},"evaluated_prompts":1480},{"family":"256x8","key":"rl16_final","label":"Final RL1500 with fresh AdamW","kind":"rl","rl_prompt_rows":53225,"log_loss":0.5188898985355062,"token_accuracy":0.8024859428405762,"format_rate":0.9993665540540541,"pass_at_k":{"1":0.38044763513513513,"10":0.5322754104004104,"11":0.5377841612216613,"12":0.5428073953073953,"13":0.5474263996138996,"14":0.5517060810810811,"15":0.5557010135135135,"16":0.5594594594594594,"2":0.433338963963964,"3":0.4602304536679537,"4":0.4781868131868132,"5":0.49167440104940097,"6":0.5024964056214056,"7":0.5115475571725572,"8":0.5193350343350343,"9":0.526174458986959},"evaluated_prompts":1480},{"family":"256x8","key":"rl11_final_prompts28419","label":"Final RL1500 with continuous AdamW \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5207206121111173,"token_accuracy":0.8035929203033447,"format_rate":0.9980574324324324,"pass_at_k":{"1":0.36896114864864865,"10":0.5278241184491185,"11":0.5330808830808831,"12":0.5377390852390852,"13":0.5418955115830115,"14":0.545625,"15":0.5489864864864865,"16":0.552027027027027,"2":0.4227252252252252,"3":0.45145149613899616,"4":0.47103170478170475,"5":0.48572604197604197,"6":0.49736665361665355,"7":0.5069189425439425,"8":0.5149547974547974,"9":0.5218405074655075},"evaluated_prompts":1480},{"family":"256x8","key":"rl12_final_prompts28419","label":"Final RL1500 with continuous AdamW after late replay \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5192720130639066,"token_accuracy":0.8038878440856934,"format_rate":0.9991131756756757,"pass_at_k":{"1":0.36925675675675673,"10":0.5048024610524611,"11":0.5096461674586675,"12":0.5140521978021978,"13":0.5180839768339769,"14":0.5217905405405405,"15":0.5252111486486486,"16":0.5283783783783784,"2":0.41692004504504504,"3":0.4410123069498069,"4":0.4570845708345708,"5":0.46907685501435503,"6":0.47862871925371925,"7":0.48658435314685317,"8":0.4934236334236334,"9":0.49943548006048005},"evaluated_prompts":1480},{"family":"256x8","key":"rl13_final_prompts28419","label":"Final RL1500 with fresh AdamW \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5205067803496429,"token_accuracy":0.8023970127105713,"format_rate":0.9991976351351352,"pass_at_k":{"1":0.37533783783783786,"10":0.5070122445122445,"11":0.5107288263538263,"12":0.514025839025839,"13":0.5169799710424711,"14":0.5196509009009009,"15":0.5220861486486487,"16":0.5243243243243243,"2":0.42407657657657655,"3":0.4489273648648649,"4":0.46493465993465993,"5":0.47637965919215924,"6":0.4851125226125226,"7":0.49209447646947646,"8":0.49787170037170037,"9":0.5027740502740503},"evaluated_prompts":1480},{"family":"256x8","key":"rl14_final_prompts28419","label":"Final RL1500 with fresh AdamW \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5186762721954696,"token_accuracy":0.8034354448318481,"format_rate":0.9991554054054054,"pass_at_k":{"1":0.3711148648648649,"10":0.5064874314874315,"11":0.5108757796257796,"12":0.5148507573507574,"13":0.5184785231660232,"14":0.5218074324324324,"15":0.5248733108108108,"16":0.5277027027027027,"2":0.4204391891891892,"3":0.445209942084942,"4":0.4612689337689337,"5":0.47299261830511835,"6":0.48216690404190404,"7":0.48968696843696846,"8":0.4960592935592935,"9":0.5015930944055944},"evaluated_prompts":1480},{"family":"256x8","key":"rl15_final_prompts28419","label":"Final RL1500 with fresh AdamW \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5213666554886016,"token_accuracy":0.8023989200592041,"format_rate":0.9994087837837838,"pass_at_k":{"1":0.37622466216216216,"10":0.5143665456165456,"11":0.5187998094248095,"12":0.5227613602613602,"13":0.5263260135135135,"14":0.5295495495495496,"15":0.5324746621621622,"16":0.5351351351351351,"2":0.42581081081081085,"3":0.4509206081081081,"4":0.4674053311553311,"5":0.4795580264330265,"6":0.4891323203823204,"7":0.49699560574560575,"8":0.5036367311367311,"9":0.5093594665469665},"evaluated_prompts":1480},{"family":"256x8","key":"rl16_final_prompts28419","label":"Final RL1500 with fresh AdamW \u00b7 28,419-prompt set","kind":"rl","rl_prompt_rows":28419,"log_loss":0.5206610285389062,"token_accuracy":0.8024152517318726,"format_rate":0.9993665540540541,"pass_at_k":{"1":0.3801097972972973,"10":0.5204257735507736,"11":0.5255981771606772,"12":0.5303333828333828,"13":0.534698359073359,"14":0.5387387387387388,"15":0.5424831081081081,"16":0.5459459459459459,"2":0.42942567567567563,"3":0.4540250965250965,"4":0.4704232254232254,"5":0.4827879046629046,"6":0.49275420525420527,"7":0.5011237833112833,"8":0.5083510783510783,"9":0.5147212837837838},"evaluated_prompts":1480},{"family":"64x16","key":"pt5b_sft3_full","label":"PT 5B + SFT \u00d73","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5124293190450496,"token_accuracy":0.8052098751068115,"format_rate":0.9813766891891892,"pass_at_k":{"1":0.19113175675675675,"10":0.5198494073494074,"11":0.5311584249084249,"12":0.5412169587169587,"13":0.5502606177606179,"14":0.5584740990990992,"15":0.5660050675675675,"16":0.572972972972973,"2":0.2906362612612613,"3":0.352566361003861,"4":0.39583308583308585,"5":0.4283175861300862,"6":0.4538944163944164,"7":0.47472683566433566,"8":0.49213596463596465,"9":0.5069811944811945},"evaluated_prompts":1480},{"family":"64x16","key":"pt2p5b_sft3_first","label":"PT 2.5B + SFT half","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5250549227713442,"token_accuracy":0.7984380722045898,"format_rate":0.9799831081081081,"pass_at_k":{"1":0.13851351351351351,"10":0.45111349798849804,"11":0.4636634058509058,"12":0.47486560736560735,"13":0.4849481177606177,"14":0.49409346846846847,"15":0.5024493243243243,"16":0.5101351351351351,"2":0.2214414414414414,"3":0.2783506274131274,"4":0.3204829967329967,"5":0.3533084595584596,"6":0.37985764235764236,"7":0.4019428038178038,"8":0.42071095571095574,"9":0.4369244353619354},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e01_rl3000_full","label":"RL 3000 \u00b7 full eligible set","kind":"rl","rl_prompt_rows":53156,"log_loss":0.5347762000126298,"token_accuracy":0.8018784523010254,"format_rate":0.9986486486486487,"pass_at_k":{"1":0.3819256756756757,"10":0.5581329987579988,"11":0.5640911172161172,"12":0.5693967181467181,"13":0.5741517857142857,"14":0.5784346846846846,"15":0.5823057432432432,"16":0.5858108108108108,"2":0.44333896396396394,"3":0.47455236486486485,"4":0.4956055093555094,"5":0.5114939114939114,"6":0.5242037523287524,"7":0.5347363447363448,"8":0.5436736386736387,"9":0.5513877929502929},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e02_rl1_a","label":"RL 1500 \u00b7 dataset A","kind":"rl","rl_prompt_rows":26578,"log_loss":0.526596832665568,"token_accuracy":0.8029974699020386,"format_rate":0.9977618243243244,"pass_at_k":{"1":0.3619932432432432,"10":0.5619907963657964,"11":0.568909947034947,"12":0.5751878526878528,"13":0.580945945945946,"14":0.5862781531531531,"15":0.5912584459459459,"16":0.595945945945946,"2":0.43110360360360356,"3":0.46697635135135135,"4":0.4911293436293437,"5":0.509222477972478,"6":0.5235918642168642,"7":0.5354543682668682,"8":0.5455275205275204,"9":0.5542696914571914},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e02_trace_chronological","label":"Chronological trace training \u00b7 645,635 successful traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5380134813721839,"token_accuracy":0.8028364181518555,"format_rate":0.9992820945945946,"pass_at_k":{"1":0.36786317567567567,"10":0.5572517853767853,"11":0.5636363355113355,"12":0.5693885506385507,"13":0.5746138996138996,"14":0.5793918918918919,"15":0.5837837837837838,"16":0.5878378378378378,"2":0.43459459459459465,"3":0.46855453667953667,"4":0.4911768636768637,"5":0.5080754690129691,"6":0.521492358992359,"7":0.5325652641277642,"8":0.5419577269577269,"9":0.5500918422793423},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e02_rl2_b","label":"RL 1500 \u00b7 dataset B","kind":"rl","rl_prompt_rows":26578,"log_loss":0.5428753650416162,"token_accuracy":0.8020278215408325,"format_rate":0.9993665540540541,"pass_at_k":{"1":0.38627533783783785,"10":0.557360376110376,"11":0.5633930118305118,"12":0.5688238788238789,"13":0.5737282818532818,"14":0.5781644144144145,"15":0.582179054054054,"16":0.5858108108108108,"2":0.4473198198198198,"3":0.4775880791505791,"4":0.49751447876447874,"5":0.5124161592911592,"6":0.5243776493776493,"7":0.5344029956529956,"8":0.5430422205422205,"9":0.5506244093744094},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e03_rl1_full","label":"RL 1500 \u00b7 full eligible set","kind":"rl","rl_prompt_rows":53156,"log_loss":0.5240711558384967,"token_accuracy":0.803371787071228,"format_rate":0.9983530405405405,"pass_at_k":{"1":0.36241554054054054,"10":0.5725731194481195,"11":0.5797568000693001,"12":0.5862704187704187,"13":0.5922212837837838,"14":0.5976914414414414,"15":0.6027449324324324,"16":0.6074324324324324,"2":0.43715653153153144,"3":0.47509049227799227,"4":0.500025245025245,"5":0.5184887325512325,"6":0.5331178112428112,"7":0.5452255599130599,"8":0.5555581805581805,"9":0.5645746314496315},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e03_trace_chronological","label":"Chronological trace training \u00b7 623,988 successful traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5356731697004479,"token_accuracy":0.8030999898910522,"format_rate":0.9997043918918919,"pass_at_k":{"1":0.3658783783783784,"10":0.5609015140265141,"11":0.567081538956539,"12":0.5726540689040689,"13":0.5777195945945945,"14":0.5823536036036036,"15":0.5866131756756757,"16":0.5905405405405405,"2":0.43793918918918917,"3":0.47351954633204635,"4":0.4964241164241164,"5":0.5131605162855163,"6":0.5262780294030294,"7":0.5370262592137592,"8":0.5461134211134211,"9":0.5539768002268003},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e03_rl2_full","label":"RL 1500 \u00b7 full eligible set","kind":"rl","rl_prompt_rows":53156,"log_loss":0.5437107646460433,"token_accuracy":0.8016986846923828,"format_rate":0.9997466216216216,"pass_at_k":{"1":0.38196790540540543,"10":0.5517814955314955,"11":0.557212775962776,"12":0.5620953370953371,"13":0.5665238899613899,"14":0.5705686936936937,"15":0.5742820945945946,"16":0.5777027027027027,"2":0.44317004504504504,"3":0.4740202702702702,"4":0.4943187555687556,"5":0.5092715139590139,"6":0.5210098178848179,"7":0.5306125968625969,"8":0.5387026712026711,"9":0.5456718956718957},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e04_rl1_a","label":"RL 1500 \u00b7 dataset A","kind":"rl","rl_prompt_rows":26578,"log_loss":0.5369070223467831,"token_accuracy":0.7963087558746338,"format_rate":0.9977195945945946,"pass_at_k":{"1":0.2875422297297297,"10":0.49452481639981644,"11":0.5021006583506583,"12":0.508994653994654,"13":0.5153269787644787,"14":0.5211880630630631,"15":0.5266469594594595,"16":0.5317567567567567,"2":0.355197072072072,"3":0.39261462355212357,"4":0.418256237006237,"5":0.4375694547569548,"6":0.4529660373410373,"7":0.465727946040446,"8":0.4766132741132741,"9":0.4861043517293518},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e04_pt2_trace_mixed","label":"PT 2.5B + SFT half + chronological traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5099554604997958,"token_accuracy":0.8044548034667969,"format_rate":0.9950591216216216,"pass_at_k":{"1":0.29472128378378376,"10":0.5487376643626644,"11":0.557399483961984,"12":0.5652171814671814,"13":0.5723395270270271,"14":0.5788795045045045,"15":0.5849239864864865,"16":0.5905405405405405,"2":0.38184684684684694,"3":0.4280574324324325,"4":0.45899168399168394,"5":0.4820274601524601,"6":0.5002594533844533,"7":0.5152790115290115,"8":0.5280094080094081,"9":0.5390323190323191},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e04_rl2_b","label":"RL 1500 \u00b7 dataset B","kind":"rl","rl_prompt_rows":26578,"log_loss":0.522744822953588,"token_accuracy":0.8017675876617432,"format_rate":0.9983952702702703,"pass_at_k":{"1":0.36157094594594597,"10":0.5405776655776656,"11":0.5465220027720028,"12":0.5519041431541432,"13":0.5568315637065637,"14":0.5613851351351351,"15":0.565625,"16":0.5695945945945946,"2":0.42518018018018017,"3":0.4577171814671815,"4":0.4790870953370953,"5":0.4948858714483715,"6":0.5073734542484543,"7":0.5176607092232092,"8":0.5263793338793339,"9":0.5339280145530145},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e05_rl1_full","label":"RL 1500 \u00b7 full eligible set","kind":"rl","rl_prompt_rows":53156,"log_loss":0.5369719017227399,"token_accuracy":0.7963387966156006,"format_rate":0.995945945945946,"pass_at_k":{"1":0.29172297297297295,"10":0.4941685004185004,"11":0.5006026631026631,"12":0.5063246213246213,"13":0.5114671814671814,"14":0.5161317567567567,"15":0.5203969594594594,"16":0.5243243243243243,"2":0.3615146396396397,"3":0.39872828185328185,"4":0.4236286011286011,"5":0.4421048967923968,"6":0.4566349191349192,"7":0.4684842184842185,"8":0.47839454839454837,"9":0.4868454687204687},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e05_pt2_trace_mixed","label":"PT 2.5B + SFT half + chronological traces","kind":"supervised","rl_prompt_rows":null,"log_loss":0.5098884055534942,"token_accuracy":0.804680585861206,"format_rate":0.9927364864864865,"pass_at_k":{"1":0.3024493243243243,"10":0.5531439000189,"11":0.5610714595089594,"12":0.5681972081972082,"13":0.5746730212355212,"14":0.5806137387387388,"15":0.5861064189189189,"16":0.5912162162162162,"2":0.39100788288288285,"3":0.43738899613899607,"4":0.46797408672408675,"5":0.4903932778932779,"6":0.5078705078705079,"7":0.5220791792666793,"8":0.5339901089901089,"9":0.5442105698355698},"evaluated_prompts":1480},{"family":"64x16","key":"r64_e05_rl2_full","label":"RL 1500 \u00b7 full eligible set","kind":"rl","rl_prompt_rows":53156,"log_loss":0.5207001141675528,"token_accuracy":0.8025496006011963,"format_rate":0.9985219594594594,"pass_at_k":{"1":0.3609375,"10":0.5546627359127358,"11":0.5620249542124542,"12":0.5687299524799525,"13":0.5748708976833977,"14":0.5805180180180181,"15":0.5857263513513513,"16":0.5905405405405405,"2":0.42555180180180174,"3":0.45991916023166024,"4":0.48323507573507574,"5":0.5008756868131868,"6":0.5151053338553339,"7":0.5270663626913626,"8":0.5374039774039774,"9":0.5465156633906634},"evaluated_prompts":1480}]