jspr's picture
Add forward self-model checkpoints (Llama 3.2 1B + toy GPT)
302675f verified
Raw
History Blame Contribute Delete
12.9 kB
{
"source_layer": 7,
"target_layer": 8,
"d_model": 2048,
"seq_len": 2048,
"fwd_n_layer": 1,
"fwd_n_head": 1,
"fwd_d_head": 128,
"fwd_mlp_mult": 2.0,
"fwd_use_swiglu": true,
"fwd_n_params": 26235264,
"main_n_params": 1235814400,
"capacity_ratio": 0.021229129552139868,
"n_steps": 10000,
"batch_size": 16,
"fwd_lr": 0.0001,
"n_train_shards": 99,
"n_val_shards": 2,
"n_val_sequences": 491,
"final_val_mse": 0.001788996768394305,
"final_val_cosine": 0.9372754942986273,
"final_val_residual_norm": 1.8941153287887573,
"residual_pca": {
"effective_rank": 1908.2322998046875,
"max_rank": 2048,
"rank_50": 485,
"rank_90": 1419,
"rank_95": 1650,
"top1_variance": 0.005640974268317223
},
"training_seconds": 12340.371667385101,
"history": {
"fwd_mse": [
[
0,
0.05361234396696091
],
[
200,
0.005762346554547548
],
[
400,
0.004786794073879719
],
[
600,
0.003985430113971233
],
[
800,
0.0037297275848686695
],
[
1000,
0.0033700044732540846
],
[
1200,
0.003214194206520915
],
[
1400,
0.003064497606828809
],
[
1600,
0.0030410615727305412
],
[
1800,
0.002859783126041293
],
[
2000,
0.0029426398687064648
],
[
2200,
0.0027721929363906384
],
[
2400,
0.002723717363551259
],
[
2600,
0.0027984725311398506
],
[
2800,
0.002662141341716051
],
[
3000,
0.0026135139632970095
],
[
3200,
0.0025119641795754433
],
[
3400,
0.002608730923384428
],
[
3600,
0.002473900094628334
],
[
3800,
0.002471000887453556
],
[
4000,
0.002470198320224881
],
[
4200,
0.0024023596197366714
],
[
4400,
0.0023067130241543055
],
[
4600,
0.002266491763293743
],
[
4800,
0.0022853089030832052
],
[
5000,
0.0022200047969818115
],
[
5200,
0.002257999498397112
],
[
5400,
0.0022179423831403255
],
[
5600,
0.002188409212976694
],
[
5800,
0.0021320050582289696
],
[
6000,
0.0020800644997507334
],
[
6200,
0.002177655231207609
],
[
6400,
0.0020370068959891796
],
[
6600,
0.0020786449313163757
],
[
6800,
0.001991601660847664
],
[
7000,
0.00202348455786705
],
[
7200,
0.001995587255805731
],
[
7400,
0.001970375655218959
],
[
7600,
0.001978542422875762
],
[
7800,
0.001997516490519047
],
[
8000,
0.0018945611082017422
],
[
8200,
0.0019200685201212764
],
[
8400,
0.0019268350442871451
],
[
8600,
0.0018792669288814068
],
[
8800,
0.0018499521538615227
],
[
9000,
0.0019255225779488683
],
[
9200,
0.0017902057152241468
],
[
9400,
0.0018028962658718228
],
[
9600,
0.0018486585468053818
],
[
9800,
0.0018617857713252306
],
[
9999,
0.0017257111612707376
]
],
"val_fwd_mse": [
[
0,
0.03311384047071139
],
[
200,
0.005512383347377181
],
[
400,
0.0045879316671440995
],
[
600,
0.003962893248535692
],
[
800,
0.003686679759994149
],
[
1000,
0.003359384679545959
],
[
1200,
0.0032044988125562667
],
[
1400,
0.003068833821453154
],
[
1600,
0.002978292355934779
],
[
1800,
0.002922045214412113
],
[
2000,
0.0028643588225046794
],
[
2200,
0.0028142516734078527
],
[
2400,
0.0027492424395556252
],
[
2600,
0.002699357736855745
],
[
2800,
0.002640491467900574
],
[
3000,
0.0026000263014187416
],
[
3200,
0.0025552748236805202
],
[
3400,
0.0025211286110182604
],
[
3600,
0.002512407225246231
],
[
3800,
0.00245760358714809
],
[
4000,
0.0024378761183470486
],
[
4200,
0.002386905951425433
],
[
4400,
0.0023764715219537417
],
[
4600,
0.0023403250457098086
],
[
4800,
0.00231518279761076
],
[
5000,
0.0022747613489627837
],
[
5200,
0.0022313310919950405
],
[
5400,
0.0022013659744213026
],
[
5600,
0.0021793171996250748
],
[
5800,
0.0021587318119903405
],
[
6000,
0.002127136965282261
],
[
6200,
0.002119861962273717
],
[
6400,
0.002091070303382973
],
[
6600,
0.0020678113680332897
],
[
6800,
0.0020570720623557764
],
[
7000,
0.0020369751223673423
],
[
7200,
0.0020223997300490735
],
[
7400,
0.001986565021798015
],
[
7600,
0.0019637826131656765
],
[
7800,
0.0019431026380819579
],
[
8000,
0.0019247106043621897
],
[
8200,
0.0019088692807902893
],
[
8400,
0.001896107786645492
],
[
8600,
0.0018856525033091505
],
[
8800,
0.0018764173689608773
],
[
9000,
0.0018630559517381093
],
[
9200,
0.001854456685638676
],
[
9400,
0.0018435753881931305
],
[
9600,
0.0018305608769878744
],
[
9800,
0.0018034361147632203
],
[
9999,
0.001789538178127259
]
],
"val_cosine_sim": [
[
0,
0.4243091017007828
],
[
200,
0.8082238614559174
],
[
400,
0.8376828610897065
],
[
600,
0.8584890842437745
],
[
800,
0.8687650005022685
],
[
1000,
0.8799132625261943
],
[
1200,
0.8853773395220439
],
[
1400,
0.8903553942839305
],
[
1600,
0.8936428467432658
],
[
1800,
0.8957931141058604
],
[
2000,
0.8978019853432974
],
[
2200,
0.8997232178846996
],
[
2400,
0.9020942270755767
],
[
2600,
0.9039487282435099
],
[
2800,
0.9061655422051748
],
[
3000,
0.9076851467291515
],
[
3200,
0.9093430797259013
],
[
3400,
0.9105890274047852
],
[
3600,
0.9109355668226878
],
[
3800,
0.912908277908961
],
[
4000,
0.91365660627683
],
[
4200,
0.9155096729596456
],
[
4400,
0.9158911069234212
],
[
4600,
0.9172228117783864
],
[
4800,
0.9181304196516673
],
[
5000,
0.9196795304616292
],
[
5200,
0.9212888975938162
],
[
5400,
0.922396965821584
],
[
5600,
0.9231784681479136
],
[
5800,
0.9239337881406148
],
[
6000,
0.9250807682673137
],
[
6200,
0.9253247896830241
],
[
6400,
0.9263921101888021
],
[
6600,
0.9272292335828145
],
[
6800,
0.9276153564453125
],
[
7000,
0.9283371845881144
],
[
7200,
0.9288679758707682
],
[
7400,
0.9302072445551555
],
[
7600,
0.9310539762179056
],
[
7800,
0.9318127512931824
],
[
8000,
0.9324680864810944
],
[
8200,
0.9330347994963328
],
[
8400,
0.9334850251674652
],
[
8600,
0.9338701903820038
],
[
8800,
0.9342047671477
],
[
9000,
0.9346590379873911
],
[
9200,
0.9349598467350007
],
[
9400,
0.9353463292121887
],
[
9600,
0.9358185668786366
],
[
9800,
0.9368309001127879
],
[
9999,
0.9373501121997834
]
],
"val_residual_norm": [
[
0,
8.19957447052002
],
[
200,
3.3171073754628497
],
[
400,
3.038803553581238
],
[
600,
2.8268452167510985
],
[
800,
2.7267664591471354
],
[
1000,
2.601663025220235
],
[
1200,
2.540529958407084
],
[
1400,
2.485924998919169
],
[
1600,
2.4484671036402386
],
[
1800,
2.42530730565389
],
[
2000,
2.400950050354004
],
[
2200,
2.379831910133362
],
[
2400,
2.3519990046819053
],
[
2600,
2.3303396622339885
],
[
2800,
2.3045204798380534
],
[
3000,
2.2867403507232664
],
[
3200,
2.2670581499735514
],
[
3400,
2.251586365699768
],
[
3600,
2.2480901082356772
],
[
3800,
2.222928269704183
],
[
4000,
2.2138659000396728
],
[
4200,
2.190362556775411
],
[
4400,
2.1856037537256876
],
[
4600,
2.1689730246861774
],
[
4800,
2.156961464881897
],
[
5000,
2.1379854599634807
],
[
5200,
2.11747457186381
],
[
5400,
2.102999965349833
],
[
5600,
2.092522080739339
],
[
5800,
2.082704210281372
],
[
6000,
2.067051116625468
],
[
6200,
2.063450400034587
],
[
6400,
2.049210047721863
],
[
6600,
2.03757913907369
],
[
6800,
2.032305860519409
],
[
7000,
2.0222584287325542
],
[
7200,
2.015055950482686
],
[
7400,
1.996809720993042
],
[
7600,
1.9852852423985798
],
[
7800,
1.9748168031374613
],
[
8000,
1.9652122696240744
],
[
8200,
1.9570548454920451
],
[
8400,
1.9504539052645364
],
[
8600,
1.9450278560320535
],
[
8800,
1.940179415543874
],
[
9000,
1.9331526279449462
],
[
9200,
1.9286702315012614
],
[
9400,
1.922955584526062
],
[
9600,
1.9158894975980123
],
[
9800,
1.9016074856122336
],
[
9999,
1.8942763884862264
]
]
}
}