Upload folder using huggingface_hub

scaling_sdpa_granularity_2/hd_4_seed_1339/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "self", "log_dir": "scaling_sdpa_granularity_2/hd_4_seed_1339", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 64, "n_embd": 256, "head_dim": 4, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 4375, "warmup_steps": 250, "group": "scaling_sdpa_granularity_2", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1339, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "disable_selection_head_linear_combo_bias": false, "assert_latent_matches_no_head": false, "n_latent_masks": null, "init_latent_masks_to_identity": false, "init_latent_masks_to_inverse": false, "latent_mask_scale": null, "latent_mask_runtime_multiplier": null, "latent_mask_sigmoid": false, "S_layernorm": false, "one_head_per_latent_mask": false, "att_conv": false, "att_conv_init": "none", "att_conv_scale": 1.0, "att_conv_weight_decay": false, "protection_kind": "none", "leaky_relu_alpha": null, "leaky_relu_bias": null, "use_compile": true, "use_mini_model": false, "upload_to_hf": true, "seq_len": 256, "batch_size": 128, "total_batch_size": 131072, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": true, "disable_selection": false, "mup_enable_coord_check_logging": false, "max_lr": 0.003, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "mup_zero_init": true, "l1_loss": false, "debugpy": false, "key": "hd_4", "latent_mask_precision": "bfloat16", "att_conv_precision": "bfloat16", "attn_precision": "bfloat16", "profile_kind": "none", "dense_attention_kind": "mha", "head_dim_value": 4, "sdpa_iter_size": null, "stabilize_attn_scores": false, "override_use_sdpa": false, "autocast_precision": "bfloat16", "simulate_micro_bs": null, "simulate_micro_bs_2": null, "ckpt_attn": false, "c_proj_scale_init": 1.0}

scaling_sdpa_granularity_2/hd_4_seed_1339/dataloader_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e9422c936c4d62660261ba5451b4b72bd5055f233420bd7b0797ede3ccedd1e
+size 964

scaling_sdpa_granularity_2/hd_4_seed_1339/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 11.3322
+0 val perplexity 83466.1328
+0 train 11.334930 (lr=4.1958e-06) (hash(x)=161979327)
+10 train 9.911111 (lr=4.6154e-05) (hash(x)=165941070)
+20 train 9.239918 (lr=8.8112e-05) (hash(x)=152918033)
+30 train 8.536682 (lr=1.3007e-04) (hash(x)=169931607)
+40 train 7.754634 (lr=1.7203e-04) (hash(x)=174342675)
+50 train 7.436986 (lr=2.1399e-04) (hash(x)=163872554)
+60 train 7.417516 (lr=2.5594e-04) (hash(x)=161534908)
+70 train 7.017256 (lr=2.9790e-04) (hash(x)=156776712)
+80 train 7.014252 (lr=3.3986e-04) (hash(x)=164996929)
+90 train 6.848983 (lr=3.8182e-04) (hash(x)=152962169)
+100 val loss 6.7726
+100 val perplexity 873.5637
+100 train 6.631457 (lr=4.2378e-04) (hash(x)=147111701)
+110 train 6.627344 (lr=4.6573e-04) (hash(x)=167595270)
+120 train 6.517109 (lr=5.0769e-04) (hash(x)=152205426)
+130 train 6.403903 (lr=5.4965e-04) (hash(x)=158427335)
+140 train 6.409801 (lr=5.9161e-04) (hash(x)=174588384)
+150 train 6.514221 (lr=6.3357e-04) (hash(x)=157067270)
+160 train 6.474116 (lr=6.7552e-04) (hash(x)=161859595)
+170 train 6.389406 (lr=7.1748e-04) (hash(x)=162739537)
+180 train 6.345077 (lr=7.5944e-04) (hash(x)=170785304)
+190 train 6.270652 (lr=8.0140e-04) (hash(x)=168658088)
+200 val loss 6.1854
+200 val perplexity 485.6113
+200 train 6.217596 (lr=8.4336e-04) (hash(x)=150749086)
+210 train 6.142254 (lr=8.8531e-04) (hash(x)=148789576)
+220 train 6.079651 (lr=9.2727e-04) (hash(x)=153164408)
+230 train 6.213134 (lr=9.6923e-04) (hash(x)=184594366)
+240 train 5.940273 (lr=1.0112e-03) (hash(x)=157817319)
+250 train 6.003999 (lr=1.0531e-03) (hash(x)=149451562)
+260 train 5.889643 (lr=1.0951e-03) (hash(x)=165738161)
+270 train 5.856654 (lr=1.1371e-03) (hash(x)=165724764)
+280 train 5.795928 (lr=1.1790e-03) (hash(x)=155414136)
+290 train 5.731412 (lr=1.2210e-03) (hash(x)=165090903)
+300 val loss 5.8145
+300 val perplexity 335.1081
+300 train 5.587214 (lr=1.2629e-03) (hash(x)=154445545)
+310 train 5.539246 (lr=1.3049e-03) (hash(x)=158949603)
+320 train 5.630632 (lr=1.3469e-03) (hash(x)=161412973)
+330 train 5.703120 (lr=1.3888e-03) (hash(x)=164572249)
+340 train 5.741727 (lr=1.4308e-03) (hash(x)=162202191)
+350 train 5.720638 (lr=1.4727e-03) (hash(x)=158135392)
+360 train 5.648538 (lr=1.5147e-03) (hash(x)=163752391)
+370 train 5.683677 (lr=1.5566e-03) (hash(x)=172758799)
+380 train 5.479600 (lr=1.5986e-03) (hash(x)=144569488)
+390 train 5.456264 (lr=1.6406e-03) (hash(x)=150429229)
+400 val loss 5.4630
+400 val perplexity 235.8104
+400 train 5.447008 (lr=1.6825e-03) (hash(x)=154261530)
+410 train 5.430774 (lr=1.7245e-03) (hash(x)=159172506)
+420 train 5.398603 (lr=1.7664e-03) (hash(x)=154419867)
+430 train 5.433768 (lr=1.8084e-03) (hash(x)=167337099)
+440 train 5.328567 (lr=1.8503e-03) (hash(x)=154535255)
+450 train 5.305804 (lr=1.8923e-03) (hash(x)=153077071)
+460 train 5.243974 (lr=1.9343e-03) (hash(x)=150853070)
+470 train 5.093634 (lr=1.9762e-03) (hash(x)=160701113)
+480 train 5.183717 (lr=2.0182e-03) (hash(x)=154690870)
+490 train 5.005250 (lr=2.0601e-03) (hash(x)=163834286)
+500 val loss 5.2770
+500 val perplexity 195.7720
+500 train 5.068475 (lr=2.1021e-03) (hash(x)=162651717)
+510 train 5.189492 (lr=2.1441e-03) (hash(x)=160448311)
+520 train 5.219241 (lr=2.1860e-03) (hash(x)=165741643)
+530 train 5.352086 (lr=2.2280e-03) (hash(x)=153201264)
+540 train 5.254241 (lr=2.2699e-03) (hash(x)=159027160)
+550 train 5.222506 (lr=2.3119e-03) (hash(x)=170769095)
+560 train 5.264449 (lr=2.3538e-03) (hash(x)=153606190)
+570 train 5.226569 (lr=2.3958e-03) (hash(x)=159708689)
+580 train 5.139729 (lr=2.4378e-03) (hash(x)=179108359)
+590 train 5.202787 (lr=2.4797e-03) (hash(x)=157134241)
+600 val loss 5.1516
+600 val perplexity 172.7156
+600 train 5.178465 (lr=2.5217e-03) (hash(x)=162178356)
+610 train 5.106272 (lr=2.5636e-03) (hash(x)=167646928)
+620 train 5.061795 (lr=2.6056e-03) (hash(x)=160979601)
+630 train 4.960162 (lr=2.6476e-03) (hash(x)=158349468)
+640 train 4.959280 (lr=2.6895e-03) (hash(x)=150370348)
+650 train 4.918564 (lr=2.7315e-03) (hash(x)=152810075)
+660 train 4.892381 (lr=2.7734e-03) (hash(x)=145786262)
+670 train 4.856458 (lr=2.8154e-03) (hash(x)=160779604)
+680 train 4.775396 (lr=2.8573e-03) (hash(x)=152343688)
+690 train 4.785786 (lr=2.8993e-03) (hash(x)=166507600)
+700 val loss 5.1132
+700 val perplexity 166.2053
+700 train 4.855059 (lr=2.9413e-03) (hash(x)=162623700)
+710 train 5.135530 (lr=2.9832e-03) (hash(x)=162293403)
+720 train 5.029587 (lr=3.0000e-03) (hash(x)=159564657)
+730 train 5.041409 (lr=2.9999e-03) (hash(x)=154209107)
+740 train 5.010594 (lr=2.9997e-03) (hash(x)=156485709)
+750 train 4.996478 (lr=2.9994e-03) (hash(x)=160834702)
+760 train 4.977389 (lr=2.9990e-03) (hash(x)=152240275)
+770 train 4.984237 (lr=2.9985e-03) (hash(x)=168903075)
+780 train 5.051382 (lr=2.9979e-03) (hash(x)=166843738)
+790 train 5.090503 (lr=2.9972e-03) (hash(x)=166358104)
+800 val loss 4.9181
+800 val perplexity 136.7418
+800 train 4.909531 (lr=2.9964e-03) (hash(x)=153477747)
+810 train 4.846362 (lr=2.9955e-03) (hash(x)=157954385)
+820 train 4.825951 (lr=2.9945e-03) (hash(x)=174383505)
+830 train 4.872127 (lr=2.9934e-03) (hash(x)=145226181)
+840 train 4.851125 (lr=2.9922e-03) (hash(x)=157665743)
+850 train 4.848803 (lr=2.9909e-03) (hash(x)=158565836)
+860 train 4.847145 (lr=2.9896e-03) (hash(x)=171168105)
+870 train 5.231100 (lr=2.9881e-03) (hash(x)=159017777)
+880 train 5.017185 (lr=2.9865e-03) (hash(x)=168156768)
+890 train 4.933808 (lr=2.9848e-03) (hash(x)=145645245)
+900 val loss 4.8135
+900 val perplexity 123.1614
+900 train 4.747865 (lr=2.9830e-03) (hash(x)=151245521)
+910 train 4.769773 (lr=2.9811e-03) (hash(x)=182614672)
+920 train 4.772495 (lr=2.9792e-03) (hash(x)=167377432)
+930 train 4.676403 (lr=2.9771e-03) (hash(x)=171347228)
+940 train 4.722653 (lr=2.9749e-03) (hash(x)=171588755)
+950 train 4.812825 (lr=2.9726e-03) (hash(x)=159560168)
+960 train 4.708476 (lr=2.9703e-03) (hash(x)=171801357)
+970 train 4.689244 (lr=2.9678e-03) (hash(x)=172513884)
+980 train 4.732615 (lr=2.9652e-03) (hash(x)=170659685)
+990 train 4.746790 (lr=2.9626e-03) (hash(x)=165436358)
+1000 val loss 4.7158
+1000 val perplexity 111.6972
+1000 train 4.819874 (lr=2.9598e-03) (hash(x)=167684690)
+1010 train 4.743705 (lr=2.9570e-03) (hash(x)=127225492)
+1020 train 4.808373 (lr=2.9540e-03) (hash(x)=167575323)
+1030 train 4.728828 (lr=2.9510e-03) (hash(x)=161266302)
+1040 train 4.691758 (lr=2.9478e-03) (hash(x)=155607661)
+1050 train 4.728576 (lr=2.9446e-03) (hash(x)=165685359)
+1060 train 4.574750 (lr=2.9412e-03) (hash(x)=160417824)
+1070 train 4.702698 (lr=2.9378e-03) (hash(x)=173313882)
+1080 train 4.667741 (lr=2.9343e-03) (hash(x)=165485047)
+1090 train 4.549989 (lr=2.9307e-03) (hash(x)=159848305)
+1100 val loss 4.6562
+1100 val perplexity 105.2372
+1100 train 4.743488 (lr=2.9270e-03) (hash(x)=170890601)
+1110 train 4.633878 (lr=2.9231e-03) (hash(x)=172689983)
+1120 train 4.691858 (lr=2.9192e-03) (hash(x)=154762859)
+1130 train 4.596595 (lr=2.9152e-03) (hash(x)=168196871)
+1140 train 4.640957 (lr=2.9112e-03) (hash(x)=156715577)
+1150 train 4.748505 (lr=2.9070e-03) (hash(x)=166381788)
+1160 train 4.664018 (lr=2.9027e-03) (hash(x)=162095130)
+1170 train 4.652794 (lr=2.8983e-03) (hash(x)=166094750)
+1180 train 4.579928 (lr=2.8939e-03) (hash(x)=154965753)
+1190 train 4.571191 (lr=2.8893e-03) (hash(x)=172766447)
+1200 val loss 4.5638
+1200 val perplexity 95.9444
+1200 train 4.493141 (lr=2.8847e-03) (hash(x)=161185334)
+1210 train 4.682182 (lr=2.8800e-03) (hash(x)=167224144)
+1220 train 4.493692 (lr=2.8751e-03) (hash(x)=167252866)
+1230 train 4.493360 (lr=2.8702e-03) (hash(x)=155856924)
+1240 train 4.515785 (lr=2.8652e-03) (hash(x)=155956446)
+1250 train 4.390602 (lr=2.8601e-03) (hash(x)=140794796)
+1260 train 4.528267 (lr=2.8550e-03) (hash(x)=166863869)
+1270 train 4.592518 (lr=2.8497e-03) (hash(x)=166021892)
+1280 train 4.606199 (lr=2.8443e-03) (hash(x)=170503648)
+1290 train 4.534116 (lr=2.8389e-03) (hash(x)=164012719)
+1300 val loss 4.5323
+1300 val perplexity 92.9744
+1300 train 4.582257 (lr=2.8333e-03) (hash(x)=156519834)
+1310 train 4.546476 (lr=2.8277e-03) (hash(x)=165146652)
+1320 train 4.577162 (lr=2.8220e-03) (hash(x)=174391990)
+1330 train 4.572653 (lr=2.8162e-03) (hash(x)=161131201)
+1340 train 4.537876 (lr=2.8103e-03) (hash(x)=162240912)
+1350 train 4.467605 (lr=2.8044e-03) (hash(x)=152462931)
+1360 train 4.454934 (lr=2.7983e-03) (hash(x)=150443278)
+1370 train 4.423910 (lr=2.7922e-03) (hash(x)=153721711)
+1380 train 4.545514 (lr=2.7860e-03) (hash(x)=158070628)
+1390 train 4.553819 (lr=2.7797e-03) (hash(x)=163318827)
+1400 val loss 4.5078
+1400 val perplexity 90.7226
+1400 train 4.467959 (lr=2.7733e-03) (hash(x)=154210336)
+1410 train 4.530313 (lr=2.7668e-03) (hash(x)=162393933)
+1420 train 4.519830 (lr=2.7603e-03) (hash(x)=156659308)
+1430 train 4.505915 (lr=2.7536e-03) (hash(x)=171243439)
+1440 train 4.627336 (lr=2.7469e-03) (hash(x)=178459521)
+1450 train 4.493991 (lr=2.7401e-03) (hash(x)=159166350)
+1460 train 4.430307 (lr=2.7333e-03) (hash(x)=173554569)
+1470 train 4.510086 (lr=2.7263e-03) (hash(x)=158362521)
+1480 train 4.470243 (lr=2.7193e-03) (hash(x)=165899207)
+1490 train 4.509147 (lr=2.7121e-03) (hash(x)=156470939)
+1500 val loss 4.4599
+1500 val perplexity 86.4768
+1500 train 4.479420 (lr=2.7050e-03) (hash(x)=164197780)
+1510 train 4.349643 (lr=2.6977e-03) (hash(x)=146709896)
+1520 train 4.364938 (lr=2.6903e-03) (hash(x)=151304838)
+1530 train 4.406455 (lr=2.6829e-03) (hash(x)=163187333)
+1540 train 4.370675 (lr=2.6754e-03) (hash(x)=153630537)
+1550 train 4.371236 (lr=2.6678e-03) (hash(x)=157299103)
+1560 train 4.286015 (lr=2.6602e-03) (hash(x)=153786477)
+1570 train 4.616540 (lr=2.6525e-03) (hash(x)=155037209)
+1580 train 4.501076 (lr=2.6447e-03) (hash(x)=146157308)
+1590 train 4.515359 (lr=2.6368e-03) (hash(x)=147823317)
+1600 val loss 4.4303
+1600 val perplexity 83.9567
+1600 train 4.534293 (lr=2.6289e-03) (hash(x)=169140159)
+1610 train 4.286151 (lr=2.6208e-03) (hash(x)=157811493)
+1620 train 4.398427 (lr=2.6128e-03) (hash(x)=168940851)
+1630 train 4.361331 (lr=2.6046e-03) (hash(x)=151149213)
+1640 train 4.477731 (lr=2.5964e-03) (hash(x)=161822009)
+1650 train 4.885291 (lr=2.5881e-03) (hash(x)=151056313)
+1660 train 4.412240 (lr=2.5797e-03) (hash(x)=156920306)
+1670 train 4.308738 (lr=2.5713e-03) (hash(x)=159695815)
+1680 train 4.335822 (lr=2.5628e-03) (hash(x)=160232907)
+1690 train 4.309434 (lr=2.5542e-03) (hash(x)=158099357)
+1700 val loss 4.4248
+1700 val perplexity 83.4947
+1700 train 4.365559 (lr=2.5455e-03) (hash(x)=164127381)
+1710 train 4.545897 (lr=2.5368e-03) (hash(x)=156592040)
+1720 train 4.461895 (lr=2.5281e-03) (hash(x)=160220265)
+1730 train 4.426192 (lr=2.5192e-03) (hash(x)=156162204)
+1740 train 4.416493 (lr=2.5103e-03) (hash(x)=152767893)
+1750 train 4.381948 (lr=2.5014e-03) (hash(x)=162707306)
+1760 train 4.399451 (lr=2.4924e-03) (hash(x)=154472004)
+1770 train 4.253201 (lr=2.4833e-03) (hash(x)=160374234)
+1780 train 4.409150 (lr=2.4741e-03) (hash(x)=159709225)
+1790 train 4.476162 (lr=2.4649e-03) (hash(x)=152095944)
+1800 val loss 4.3733
+1800 val perplexity 79.3067
+1800 train 4.421177 (lr=2.4556e-03) (hash(x)=157920255)
+1810 train 4.169903 (lr=2.4463e-03) (hash(x)=167237680)
+1820 train 4.277070 (lr=2.4369e-03) (hash(x)=161255774)
+1830 train 4.332584 (lr=2.4275e-03) (hash(x)=154275481)
+1840 train 4.255128 (lr=2.4180e-03) (hash(x)=152398178)
+1850 train 4.379836 (lr=2.4084e-03) (hash(x)=147646022)
+1860 train 4.407181 (lr=2.3988e-03) (hash(x)=158895559)
+1870 train 4.328145 (lr=2.3891e-03) (hash(x)=162742598)
+1880 train 4.343457 (lr=2.3794e-03) (hash(x)=169911029)
+1890 train 4.349477 (lr=2.3696e-03) (hash(x)=172068270)
+1900 val loss 4.3730
+1900 val perplexity 79.2829
+1900 train 4.444389 (lr=2.3598e-03) (hash(x)=159975457)
+1910 train 4.440850 (lr=2.3499e-03) (hash(x)=178775667)
+1920 train 4.308937 (lr=2.3400e-03) (hash(x)=151180831)
+1930 train 4.320154 (lr=2.3300e-03) (hash(x)=154239412)
+1940 train 4.322514 (lr=2.3200e-03) (hash(x)=139144465)
+1950 train 4.292586 (lr=2.3099e-03) (hash(x)=194482413)
+1960 train 4.304648 (lr=2.2998e-03) (hash(x)=155812344)
+1970 train 4.320321 (lr=2.2896e-03) (hash(x)=157577031)
+1980 train 4.375261 (lr=2.2793e-03) (hash(x)=152045027)
+1990 train 4.241688 (lr=2.2691e-03) (hash(x)=141742221)
+2000 val loss 4.3226
+2000 val perplexity 75.3862
+2000 train 4.380366 (lr=2.2588e-03) (hash(x)=160555812)
+2010 train 4.325287 (lr=2.2484e-03) (hash(x)=160371335)
+2020 train 4.293603 (lr=2.2380e-03) (hash(x)=167890767)
+2030 train 4.373238 (lr=2.2275e-03) (hash(x)=165127715)
+2040 train 4.323822 (lr=2.2170e-03) (hash(x)=151587302)
+2050 train 4.338520 (lr=2.2065e-03) (hash(x)=172347989)
+2060 train 4.408688 (lr=2.1959e-03) (hash(x)=161301919)
+2070 train 4.338322 (lr=2.1853e-03) (hash(x)=161472934)
+2080 train 4.433645 (lr=2.1746e-03) (hash(x)=159777322)
+2090 train 4.337436 (lr=2.1639e-03) (hash(x)=151149770)
+2100 val loss 4.3008
+2100 val perplexity 73.7623
+2100 train 4.175828 (lr=2.1532e-03) (hash(x)=160283520)
+2110 train 4.343955 (lr=2.1424e-03) (hash(x)=154222025)
+2120 train 4.266326 (lr=2.1316e-03) (hash(x)=155638178)
+2130 train 4.199375 (lr=2.1208e-03) (hash(x)=149125384)
+2140 train 4.263052 (lr=2.1099e-03) (hash(x)=159496074)
+2150 train 4.351649 (lr=2.0990e-03) (hash(x)=162019154)
+2160 train 4.494893 (lr=2.0881e-03) (hash(x)=155466928)
+2170 train 4.395245 (lr=2.0771e-03) (hash(x)=173469994)
+2180 train 4.495804 (lr=2.0661e-03) (hash(x)=163448753)
+2190 train 4.328511 (lr=2.0550e-03) (hash(x)=166595123)
+2200 val loss 4.2783
+2200 val perplexity 72.1208
+2200 train 4.189005 (lr=2.0440e-03) (hash(x)=165958875)
+2210 train 4.318968 (lr=2.0329e-03) (hash(x)=165019039)
+2220 train 4.313242 (lr=2.0217e-03) (hash(x)=167325225)
+2230 train 4.146289 (lr=2.0106e-03) (hash(x)=154287439)
+2240 train 4.230020 (lr=1.9994e-03) (hash(x)=158565817)
+2250 train 4.182448 (lr=1.9882e-03) (hash(x)=160592835)
+2260 train 4.207808 (lr=1.9770e-03) (hash(x)=144591648)
+2270 train 4.106730 (lr=1.9657e-03) (hash(x)=150486876)
+2280 train 4.294670 (lr=1.9544e-03) (hash(x)=159194788)
+2290 train 4.387677 (lr=1.9431e-03) (hash(x)=164315345)
+2300 val loss 4.2554
+2300 val perplexity 70.4870
+2300 train 4.291744 (lr=1.9318e-03) (hash(x)=170470665)
+2310 train 4.247319 (lr=1.9205e-03) (hash(x)=176539068)
+2320 train 4.230088 (lr=1.9091e-03) (hash(x)=161008474)
+2330 train 4.321466 (lr=1.8977e-03) (hash(x)=168474380)
+2340 train 4.307979 (lr=1.8863e-03) (hash(x)=156072131)
+2350 train 4.429787 (lr=1.8749e-03) (hash(x)=165127087)
+2360 train 4.383093 (lr=1.8635e-03) (hash(x)=159462801)
+2370 train 4.228158 (lr=1.8520e-03) (hash(x)=161698394)
+2380 train 4.153435 (lr=1.8406e-03) (hash(x)=165588078)
+2390 train 4.159166 (lr=1.8291e-03) (hash(x)=171673672)
+2400 val loss 4.2494
+2400 val perplexity 70.0638
+2400 train 4.223001 (lr=1.8176e-03) (hash(x)=157381516)
+2410 train 4.225038 (lr=1.8061e-03) (hash(x)=163238532)
+2420 train 4.187691 (lr=1.7946e-03) (hash(x)=156178237)
+2430 train 4.071740 (lr=1.7830e-03) (hash(x)=155080617)
+2440 train 4.169430 (lr=1.7715e-03) (hash(x)=167490538)
+2450 train 4.136563 (lr=1.7600e-03) (hash(x)=163757352)
+2460 train 4.096606 (lr=1.7484e-03) (hash(x)=156734811)
+2470 train 4.085884 (lr=1.7368e-03) (hash(x)=161522341)
+2480 train 4.361667 (lr=1.7253e-03) (hash(x)=174032060)
+2490 train 4.316449 (lr=1.7137e-03) (hash(x)=163065818)
+2500 val loss 4.2175
+2500 val perplexity 67.8635
+2500 train 4.376933 (lr=1.7021e-03) (hash(x)=126959780)
+2510 train 4.301071 (lr=1.6906e-03) (hash(x)=153844939)
+2520 train 4.283594 (lr=1.6790e-03) (hash(x)=174018174)
+2530 train 4.256296 (lr=1.6674e-03) (hash(x)=150568427)
+2540 train 4.244142 (lr=1.6558e-03) (hash(x)=161725206)
+2550 train 4.223493 (lr=1.6442e-03) (hash(x)=181346417)
+2560 train 4.210876 (lr=1.6326e-03) (hash(x)=168730011)
+2570 train 4.092896 (lr=1.6210e-03) (hash(x)=155900214)
+2580 train 4.140552 (lr=1.6094e-03) (hash(x)=160014621)
+2590 train 4.115013 (lr=1.5979e-03) (hash(x)=148996689)
+2600 val loss 4.1986
+2600 val perplexity 66.5909
+2600 train 4.199552 (lr=1.5863e-03) (hash(x)=145183968)
+2610 train 3.948023 (lr=1.5747e-03) (hash(x)=159420240)
+2620 train 3.920271 (lr=1.5632e-03) (hash(x)=151169903)
+2630 train 3.951603 (lr=1.5516e-03) (hash(x)=149454811)
+2640 train 3.999276 (lr=1.5400e-03) (hash(x)=155603914)
+2650 train 4.003675 (lr=1.5285e-03) (hash(x)=169545195)
+2660 train 4.359704 (lr=1.5170e-03) (hash(x)=168454079)
+2670 train 4.248343 (lr=1.5054e-03) (hash(x)=163110417)
+2680 train 4.207772 (lr=1.4939e-03) (hash(x)=166075603)
+2690 train 4.235386 (lr=1.4824e-03) (hash(x)=174132258)
+2700 val loss 4.1921
+2700 val perplexity 66.1594
+2700 train 4.259754 (lr=1.4709e-03) (hash(x)=169846940)
+2710 train 4.127174 (lr=1.4594e-03) (hash(x)=151553731)
+2720 train 4.228018 (lr=1.4480e-03) (hash(x)=164024255)
+2730 train 4.240393 (lr=1.4365e-03) (hash(x)=156458097)
+2740 train 4.177577 (lr=1.4251e-03) (hash(x)=154663748)
+2750 train 4.060587 (lr=1.4137e-03) (hash(x)=157548181)
+2760 train 4.166643 (lr=1.4023e-03) (hash(x)=151584783)
+2770 train 4.190203 (lr=1.3909e-03) (hash(x)=152692594)
+2780 train 4.156138 (lr=1.3795e-03) (hash(x)=165952919)
+2790 train 4.046065 (lr=1.3682e-03) (hash(x)=157484183)
+2800 val loss 4.1741
+2800 val perplexity 64.9824
+2800 train 4.114857 (lr=1.3569e-03) (hash(x)=154790034)
+2810 train 3.979152 (lr=1.3456e-03) (hash(x)=157856097)
+2820 train 4.006119 (lr=1.3343e-03) (hash(x)=168903939)
+2830 train 4.082151 (lr=1.3230e-03) (hash(x)=167001475)
+2840 train 3.914632 (lr=1.3118e-03) (hash(x)=158987150)
+2850 train 4.167495 (lr=1.3006e-03) (hash(x)=168695407)
+2860 train 4.283469 (lr=1.2894e-03) (hash(x)=164999034)
+2870 train 4.133539 (lr=1.2783e-03) (hash(x)=191197316)
+2880 train 4.089507 (lr=1.2671e-03) (hash(x)=159625200)
+2890 train 4.249399 (lr=1.2560e-03) (hash(x)=157756126)
+2900 val loss 4.1431
+2900 val perplexity 62.9971
+2900 train 4.112794 (lr=1.2450e-03) (hash(x)=164721377)
+2910 train 4.142870 (lr=1.2339e-03) (hash(x)=155873181)
+2920 train 4.099512 (lr=1.2229e-03) (hash(x)=156477082)
+2930 train 4.102098 (lr=1.2119e-03) (hash(x)=151261633)
+2940 train 4.039104 (lr=1.2010e-03) (hash(x)=160090385)
+2950 train 4.195669 (lr=1.1901e-03) (hash(x)=157300618)
+2960 train 4.093276 (lr=1.1792e-03) (hash(x)=163095555)
+2970 train 4.145304 (lr=1.1684e-03) (hash(x)=149649212)
+2980 train 3.934758 (lr=1.1576e-03) (hash(x)=174517675)
+2990 train 3.918721 (lr=1.1468e-03) (hash(x)=161344491)
+3000 val loss 4.1479
+3000 val perplexity 63.2980
+3000 train 3.939673 (lr=1.1361e-03) (hash(x)=163816905)
+3010 train 3.844280 (lr=1.1254e-03) (hash(x)=157420680)
+3020 train 3.969043 (lr=1.1147e-03) (hash(x)=153843035)
+3030 train 4.138878 (lr=1.1041e-03) (hash(x)=149770373)
+3040 train 4.206076 (lr=1.0935e-03) (hash(x)=165320741)
+3050 train 4.191844 (lr=1.0830e-03) (hash(x)=170181151)
+3060 train 4.111987 (lr=1.0725e-03) (hash(x)=163837230)
+3070 train 4.107593 (lr=1.0620e-03) (hash(x)=156812797)
+3080 train 4.062653 (lr=1.0516e-03) (hash(x)=143663627)
+3090 train 4.090146 (lr=1.0412e-03) (hash(x)=165052478)
+3100 val loss 4.1221
+3100 val perplexity 61.6864
+3100 train 4.031918 (lr=1.0309e-03) (hash(x)=151458705)
+3110 train 4.086414 (lr=1.0207e-03) (hash(x)=158228317)
+3120 train 4.250598 (lr=1.0104e-03) (hash(x)=149856801)
+3130 train 3.933855 (lr=1.0002e-03) (hash(x)=146297847)
+3140 train 4.030526 (lr=9.9011e-04) (hash(x)=172795167)
+3150 train 4.081095 (lr=9.8002e-04) (hash(x)=145936074)
+3160 train 4.166273 (lr=9.6999e-04) (hash(x)=157564303)
+3170 train 4.276204 (lr=9.6000e-04) (hash(x)=157451362)
+3180 train 4.190989 (lr=9.5007e-04) (hash(x)=156366043)
+3190 train 4.138556 (lr=9.4019e-04) (hash(x)=162344697)
+3200 val loss 4.1010
+3200 val perplexity 60.4022
+3200 train 4.160634 (lr=9.3036e-04) (hash(x)=164206298)
+3210 train 4.248073 (lr=9.2058e-04) (hash(x)=149468251)
+3220 train 4.168515 (lr=9.1085e-04) (hash(x)=157808606)
+3230 train 4.317824 (lr=9.0118e-04) (hash(x)=157078690)
+3240 train 4.114352 (lr=8.9157e-04) (hash(x)=165699696)
+3250 train 4.067305 (lr=8.8201e-04) (hash(x)=181352842)
+3260 train 4.162502 (lr=8.7251e-04) (hash(x)=142534383)
+3270 train 4.106884 (lr=8.6307e-04) (hash(x)=165297359)
+3280 train 4.152468 (lr=8.5368e-04) (hash(x)=160326815)
+3290 train 4.075771 (lr=8.4435e-04) (hash(x)=150043063)
+3300 val loss 4.0943
+3300 val perplexity 59.9964
+3300 train 4.038572 (lr=8.3508e-04) (hash(x)=164133091)
+3310 train 3.973501 (lr=8.2588e-04) (hash(x)=145803863)
+3320 train 3.997127 (lr=8.1673e-04) (hash(x)=157282009)
+3330 train 3.938763 (lr=8.0764e-04) (hash(x)=159126415)
+3340 train 4.150199 (lr=7.9862e-04) (hash(x)=164426675)
+3350 train 4.177926 (lr=7.8966e-04) (hash(x)=170424924)
+3360 train 4.128692 (lr=7.8076e-04) (hash(x)=163161621)
+3370 train 4.187483 (lr=7.7192e-04) (hash(x)=160786731)
+3380 train 4.141851 (lr=7.6315e-04) (hash(x)=165954688)
+3390 train 4.177793 (lr=7.5445e-04) (hash(x)=147371107)
+3400 val loss 4.0713
+3400 val perplexity 58.6333
+3400 train 4.123873 (lr=7.4581e-04) (hash(x)=153067604)
+3410 train 4.120017 (lr=7.3724e-04) (hash(x)=161748196)
+3420 train 4.130265 (lr=7.2874e-04) (hash(x)=168297963)
+3430 train 4.132401 (lr=7.2030e-04) (hash(x)=145086892)
+3440 train 4.035774 (lr=7.1193e-04) (hash(x)=173125711)
+3450 train 4.116905 (lr=7.0363e-04) (hash(x)=158080216)
+3460 train 3.949184 (lr=6.9541e-04) (hash(x)=158055750)
+3470 train 4.053069 (lr=6.8725e-04) (hash(x)=151381400)
+3480 train 3.990652 (lr=6.7916e-04) (hash(x)=147262346)
+3490 train 3.976129 (lr=6.7114e-04) (hash(x)=154740086)
+3500 val loss 4.0706
+3500 val perplexity 58.5925
+3500 train 3.892829 (lr=6.6320e-04) (hash(x)=145730968)
+3510 train 4.044770 (lr=6.5533e-04) (hash(x)=151355929)
+3520 train 4.023386 (lr=6.4753e-04) (hash(x)=165136835)
+3530 train 4.109654 (lr=6.3981e-04) (hash(x)=163869056)
+3540 train 4.047315 (lr=6.3216e-04) (hash(x)=161580134)
+3550 train 4.076477 (lr=6.2458e-04) (hash(x)=168595690)
+3560 train 4.086512 (lr=6.1708e-04) (hash(x)=158576466)
+3570 train 4.065926 (lr=6.0966e-04) (hash(x)=157298619)
+3580 train 4.132777 (lr=6.0231e-04) (hash(x)=160892091)
+3590 train 4.003926 (lr=5.9504e-04) (hash(x)=165168008)
+3600 val loss 4.0492
+3600 val perplexity 57.3495
+3600 train 4.107705 (lr=5.8785e-04) (hash(x)=144911638)
+3610 train 4.073857 (lr=5.8074e-04) (hash(x)=150614513)
+3620 train 4.061858 (lr=5.7370e-04) (hash(x)=167695016)
+3630 train 3.977583 (lr=5.6675e-04) (hash(x)=182708233)
+3640 train 4.076303 (lr=5.5987e-04) (hash(x)=154146274)
+3650 train 4.048207 (lr=5.5308e-04) (hash(x)=159129968)
+3660 train 3.950872 (lr=5.4636e-04) (hash(x)=146209497)
+3670 train 3.988142 (lr=5.3973e-04) (hash(x)=145578104)
+3680 train 3.964446 (lr=5.3318e-04) (hash(x)=150243521)
+3690 train 3.926083 (lr=5.2671e-04) (hash(x)=154328313)
+3700 val loss 4.0503
+3700 val perplexity 57.4168
+3700 train 3.908203 (lr=5.2033e-04) (hash(x)=159904582)
+3710 train 4.271736 (lr=5.1402e-04) (hash(x)=175986417)
+3720 train 4.170239 (lr=5.0780e-04) (hash(x)=158509479)
+3730 train 4.171577 (lr=5.0167e-04) (hash(x)=156087760)
+3740 train 4.183320 (lr=4.9562e-04) (hash(x)=177667170)
+3750 train 4.141140 (lr=4.8965e-04) (hash(x)=145256842)
+3760 train 4.104769 (lr=4.8377e-04) (hash(x)=149855635)
+3770 train 4.154613 (lr=4.7798e-04) (hash(x)=168798886)
+3780 train 4.053381 (lr=4.7227e-04) (hash(x)=156277179)
+3790 train 4.022933 (lr=4.6665e-04) (hash(x)=149698945)
+3800 val loss 4.0319
+3800 val perplexity 56.3655
+3800 train 3.990012 (lr=4.6112e-04) (hash(x)=168039654)
+3810 train 4.104480 (lr=4.5567e-04) (hash(x)=154595574)
+3820 train 4.150745 (lr=4.5031e-04) (hash(x)=158303284)
+3830 train 4.167000 (lr=4.4504e-04) (hash(x)=164128644)
+3840 train 4.102428 (lr=4.3986e-04) (hash(x)=166842997)
+3850 train 4.044752 (lr=4.3477e-04) (hash(x)=177681163)
+3860 train 4.009187 (lr=4.2977e-04) (hash(x)=177062575)
+3870 train 4.043082 (lr=4.2486e-04) (hash(x)=158957159)
+3880 train 3.976121 (lr=4.2004e-04) (hash(x)=151567740)
+3890 train 4.063010 (lr=4.1530e-04) (hash(x)=153261675)
+3900 val loss 4.0225
+3900 val perplexity 55.8399
+3900 train 4.309438 (lr=4.1066e-04) (hash(x)=160268683)
+3910 train 4.042603 (lr=4.0611e-04) (hash(x)=153745240)
+3920 train 4.099527 (lr=4.0166e-04) (hash(x)=164537824)
+3930 train 4.049195 (lr=3.9729e-04) (hash(x)=162062116)
+3940 train 4.006866 (lr=3.9302e-04) (hash(x)=168626742)
+3950 train 4.079290 (lr=3.8884e-04) (hash(x)=148753561)
+3960 train 4.005887 (lr=3.8475e-04) (hash(x)=163526242)
+3970 train 4.219161 (lr=3.8076e-04) (hash(x)=159329323)
+3980 train 4.132071 (lr=3.7685e-04) (hash(x)=146502339)
+3990 train 3.830221 (lr=3.7305e-04) (hash(x)=158409708)
+4000 val loss 4.0159
+4000 val perplexity 55.4718
+4000 train 4.023207 (lr=3.6933e-04) (hash(x)=154645137)
+4010 train 4.010023 (lr=3.6572e-04) (hash(x)=160044026)
+4020 train 3.996684 (lr=3.6219e-04) (hash(x)=155544686)
+4030 train 3.975216 (lr=3.5876e-04) (hash(x)=147589421)
+4040 train 4.084553 (lr=3.5543e-04) (hash(x)=158607990)
+4050 train 3.996593 (lr=3.5219e-04) (hash(x)=154108872)
+4060 train 4.081810 (lr=3.4905e-04) (hash(x)=167101465)
+4070 train 4.047534 (lr=3.4600e-04) (hash(x)=160323363)
+4080 train 4.036805 (lr=3.4305e-04) (hash(x)=150423692)
+4090 train 4.006117 (lr=3.4019e-04) (hash(x)=167461885)
+4100 val loss 4.0058
+4100 val perplexity 54.9137
+4100 train 4.049407 (lr=3.3744e-04) (hash(x)=151783377)
+4110 train 4.036817 (lr=3.3477e-04) (hash(x)=165679649)
+4120 train 3.952070 (lr=3.3221e-04) (hash(x)=153643039)
+4130 train 3.995200 (lr=3.2974e-04) (hash(x)=149648749)
+4140 train 4.024305 (lr=3.2737e-04) (hash(x)=162751418)
+4150 train 4.027511 (lr=3.2510e-04) (hash(x)=157047478)
+4160 train 3.963090 (lr=3.2292e-04) (hash(x)=156391812)
+4170 train 4.030575 (lr=3.2085e-04) (hash(x)=180926985)
+4180 train 4.080810 (lr=3.1887e-04) (hash(x)=168699059)
+4190 train 3.987277 (lr=3.1699e-04) (hash(x)=155410372)
+4200 val loss 4.0014
+4200 val perplexity 54.6759
+4200 train 3.997289 (lr=3.1520e-04) (hash(x)=152615252)
+4210 train 3.965501 (lr=3.1352e-04) (hash(x)=155679453)
+4220 train 4.081465 (lr=3.1193e-04) (hash(x)=152954399)
+4230 train 4.089381 (lr=3.1044e-04) (hash(x)=164608084)
+4240 train 4.103578 (lr=3.0905e-04) (hash(x)=162463867)
+4250 train 4.022344 (lr=3.0776e-04) (hash(x)=168832088)
+4260 train 4.021495 (lr=3.0657e-04) (hash(x)=162698882)
+4270 train 4.037497 (lr=3.0548e-04) (hash(x)=180861492)
+4280 train 3.918041 (lr=3.0449e-04) (hash(x)=160824300)
+4290 train 3.980855 (lr=3.0359e-04) (hash(x)=157750938)
+4300 val loss 4.0012
+4300 val perplexity 54.6632
+4300 train 3.926705 (lr=3.0280e-04) (hash(x)=157216833)
+4310 train 3.929052 (lr=3.0210e-04) (hash(x)=176184661)
+4320 train 4.093684 (lr=3.0150e-04) (hash(x)=161537117)
+4330 train 4.044873 (lr=3.0101e-04) (hash(x)=160700216)
+4340 train 4.013275 (lr=3.0061e-04) (hash(x)=175519903)
+4350 train 4.002595 (lr=3.0031e-04) (hash(x)=167988701)
+4360 train 3.949073 (lr=3.0011e-04) (hash(x)=162619084)
+4370 train 4.336299 (lr=3.0001e-04) (hash(x)=166292684)
+4374 val loss 3.9946
+4374 val perplexity 54.3021

scaling_sdpa_granularity_2/hd_4_seed_1339/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12d8b9b24f423f420485ed805844b7e96da38d6cafd483135f96e4aafd40938b
+size 89692138

scaling_sdpa_granularity_2/hd_4_seed_1339/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7937aaa23bf1e8299d42dc0c1023750d297c6da2f23f5efef91f49fa665fa53b
+size 179406342