Upload folder using huggingface_hub

scaling_sdpa_granularity_2/hd_128_seed_1340/args.json

@@ -0,0 +1 @@
+{"hellaswag": true, "attention_kind": "self", "log_dir": "scaling_sdpa_granularity_2/hd_128_seed_1340", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 2, "n_embd": 256, "head_dim": 128, "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": 1340, "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_128", "latent_mask_precision": "bfloat16", "att_conv_precision": "bfloat16", "attn_precision": "bfloat16", "profile_kind": "none", "dense_attention_kind": "mha", "head_dim_value": 128, "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_128_seed_1340/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_128_seed_1340/log2.txt

@@ -0,0 +1,529 @@
+max_steps: 4375
+0 val loss 11.3290
+0 val perplexity 83198.6328
+0 train 11.329158 (lr=4.1958e-06) (hash(x)=173692205)
+10 train 9.912161 (lr=4.6154e-05) (hash(x)=166272335)
+20 train 9.313446 (lr=8.8112e-05) (hash(x)=152988153)
+30 train 8.541751 (lr=1.3007e-04) (hash(x)=165375545)
+40 train 7.748933 (lr=1.7203e-04) (hash(x)=160434435)
+50 train 7.374897 (lr=2.1399e-04) (hash(x)=155536673)
+60 train 7.184825 (lr=2.5594e-04) (hash(x)=152324251)
+70 train 7.145854 (lr=2.9790e-04) (hash(x)=152719350)
+80 train 7.033312 (lr=3.3986e-04) (hash(x)=164461998)
+90 train 6.842875 (lr=3.8182e-04) (hash(x)=160609072)
+100 val loss 6.7795
+100 val perplexity 879.6347
+100 train 6.766354 (lr=4.2378e-04) (hash(x)=166364183)
+110 train 6.704045 (lr=4.6573e-04) (hash(x)=163817364)
+120 train 6.638437 (lr=5.0769e-04) (hash(x)=165919909)
+130 train 6.530449 (lr=5.4965e-04) (hash(x)=168185952)
+140 train 6.438025 (lr=5.9161e-04) (hash(x)=157062392)
+150 train 6.425519 (lr=6.3357e-04) (hash(x)=162070558)
+160 train 6.326694 (lr=6.7552e-04) (hash(x)=167448687)
+170 train 6.417840 (lr=7.1748e-04) (hash(x)=163268047)
+180 train 6.442662 (lr=7.5944e-04) (hash(x)=162958229)
+190 train 6.193923 (lr=8.0140e-04) (hash(x)=158951412)
+200 val loss 6.1875
+200 val perplexity 486.6269
+200 train 6.151926 (lr=8.4336e-04) (hash(x)=160898816)
+210 train 6.108888 (lr=8.8531e-04) (hash(x)=153803852)
+220 train 6.053051 (lr=9.2727e-04) (hash(x)=150534761)
+230 train 6.067735 (lr=9.6923e-04) (hash(x)=167269298)
+240 train 6.122449 (lr=1.0112e-03) (hash(x)=158873716)
+250 train 6.083160 (lr=1.0531e-03) (hash(x)=163730102)
+260 train 6.027517 (lr=1.0951e-03) (hash(x)=173556083)
+270 train 5.929714 (lr=1.1371e-03) (hash(x)=153796179)
+280 train 5.919212 (lr=1.1790e-03) (hash(x)=168523171)
+290 train 5.836184 (lr=1.2210e-03) (hash(x)=145409987)
+300 val loss 5.8198
+300 val perplexity 336.8960
+300 train 5.839667 (lr=1.2629e-03) (hash(x)=157831091)
+310 train 5.717012 (lr=1.3049e-03) (hash(x)=161940559)
+320 train 5.770606 (lr=1.3469e-03) (hash(x)=151097783)
+330 train 5.704552 (lr=1.3888e-03) (hash(x)=149560294)
+340 train 5.686882 (lr=1.4308e-03) (hash(x)=164151118)
+350 train 5.615056 (lr=1.4727e-03) (hash(x)=154545714)
+360 train 5.567285 (lr=1.5147e-03) (hash(x)=168490983)
+370 train 5.586521 (lr=1.5566e-03) (hash(x)=157243384)
+380 train 5.584722 (lr=1.5986e-03) (hash(x)=161439594)
+390 train 5.523410 (lr=1.6406e-03) (hash(x)=158011549)
+400 val loss 5.5078
+400 val perplexity 246.6054
+400 train 5.594202 (lr=1.6825e-03) (hash(x)=161977759)
+410 train 5.539558 (lr=1.7245e-03) (hash(x)=169755458)
+420 train 5.514872 (lr=1.7664e-03) (hash(x)=165992049)
+430 train 5.504977 (lr=1.8084e-03) (hash(x)=167325398)
+440 train 5.470846 (lr=1.8503e-03) (hash(x)=138405929)
+450 train 5.434680 (lr=1.8923e-03) (hash(x)=148529257)
+460 train 5.506111 (lr=1.9343e-03) (hash(x)=187877362)
+470 train 5.385541 (lr=1.9762e-03) (hash(x)=174097227)
+480 train 5.284330 (lr=2.0182e-03) (hash(x)=163811577)
+490 train 5.372972 (lr=2.0601e-03) (hash(x)=147406221)
+500 val loss 5.3368
+500 val perplexity 207.8365
+500 train 5.262817 (lr=2.1021e-03) (hash(x)=159337928)
+510 train 5.269812 (lr=2.1441e-03) (hash(x)=155002527)
+520 train 5.327226 (lr=2.1860e-03) (hash(x)=163685257)
+530 train 5.303315 (lr=2.2280e-03) (hash(x)=139648586)
+540 train 5.335370 (lr=2.2699e-03) (hash(x)=158641796)
+550 train 5.195616 (lr=2.3119e-03) (hash(x)=163537724)
+560 train 5.250512 (lr=2.3538e-03) (hash(x)=168609033)
+570 train 5.288931 (lr=2.3958e-03) (hash(x)=155826061)
+580 train 5.286271 (lr=2.4378e-03) (hash(x)=166604533)
+590 train 5.255665 (lr=2.4797e-03) (hash(x)=150461749)
+600 val loss 5.1676
+600 val perplexity 175.4953
+600 train 5.205902 (lr=2.5217e-03) (hash(x)=158810876)
+610 train 5.545979 (lr=2.5636e-03) (hash(x)=161506635)
+620 train 5.079613 (lr=2.6056e-03) (hash(x)=153757394)
+630 train 5.160909 (lr=2.6476e-03) (hash(x)=163276041)
+640 train 5.095736 (lr=2.6895e-03) (hash(x)=157085458)
+650 train 5.082491 (lr=2.7315e-03) (hash(x)=156529604)
+660 train 5.130531 (lr=2.7734e-03) (hash(x)=161460943)
+670 train 5.124577 (lr=2.8154e-03) (hash(x)=163730633)
+680 train 5.221930 (lr=2.8573e-03) (hash(x)=172158062)
+690 train 5.035507 (lr=2.8993e-03) (hash(x)=168091722)
+700 val loss 5.0177
+700 val perplexity 151.0651
+700 train 5.035420 (lr=2.9413e-03) (hash(x)=163165614)
+710 train 5.202437 (lr=2.9832e-03) (hash(x)=170735476)
+720 train 4.994598 (lr=3.0000e-03) (hash(x)=166422327)
+730 train 4.938062 (lr=2.9999e-03) (hash(x)=151626971)
+740 train 4.902167 (lr=2.9997e-03) (hash(x)=158266349)
+750 train 4.925980 (lr=2.9994e-03) (hash(x)=165279256)
+760 train 4.835217 (lr=2.9990e-03) (hash(x)=164452783)
+770 train 4.843714 (lr=2.9985e-03) (hash(x)=165056209)
+780 train 4.680177 (lr=2.9979e-03) (hash(x)=153309114)
+790 train 4.636685 (lr=2.9972e-03) (hash(x)=159931795)
+800 val loss 4.8751
+800 val perplexity 130.9846
+800 train 4.680568 (lr=2.9964e-03) (hash(x)=170348996)
+810 train 4.905035 (lr=2.9955e-03) (hash(x)=168036306)
+820 train 4.815428 (lr=2.9945e-03) (hash(x)=160083807)
+830 train 4.970596 (lr=2.9934e-03) (hash(x)=155107640)
+840 train 4.787980 (lr=2.9922e-03) (hash(x)=164587669)
+850 train 4.835510 (lr=2.9909e-03) (hash(x)=154395552)
+860 train 4.791926 (lr=2.9896e-03) (hash(x)=156970280)
+870 train 4.765543 (lr=2.9881e-03) (hash(x)=154303917)
+880 train 4.768934 (lr=2.9865e-03) (hash(x)=156001085)
+890 train 4.697968 (lr=2.9848e-03) (hash(x)=154724999)
+900 val loss 4.7388
+900 val perplexity 114.2991
+900 train 4.695426 (lr=2.9830e-03) (hash(x)=149318422)
+910 train 4.713167 (lr=2.9811e-03) (hash(x)=174182960)
+920 train 4.744485 (lr=2.9792e-03) (hash(x)=160259547)
+930 train 4.574017 (lr=2.9771e-03) (hash(x)=156727936)
+940 train 4.709822 (lr=2.9749e-03) (hash(x)=147186760)
+950 train 4.590715 (lr=2.9726e-03) (hash(x)=156729166)
+960 train 4.619873 (lr=2.9703e-03) (hash(x)=163690830)
+970 train 4.613582 (lr=2.9678e-03) (hash(x)=164750190)
+980 train 4.532709 (lr=2.9652e-03) (hash(x)=165946603)
+990 train 4.591125 (lr=2.9626e-03) (hash(x)=156682992)
+1000 val loss 4.6656
+1000 val perplexity 106.2297
+1000 train 4.844721 (lr=2.9598e-03) (hash(x)=152147013)
+1010 train 4.804036 (lr=2.9570e-03) (hash(x)=171157521)
+1020 train 4.688347 (lr=2.9540e-03) (hash(x)=191295677)
+1030 train 4.554823 (lr=2.9510e-03) (hash(x)=159879474)
+1040 train 4.693347 (lr=2.9478e-03) (hash(x)=166566331)
+1050 train 4.582646 (lr=2.9446e-03) (hash(x)=157178542)
+1060 train 4.603128 (lr=2.9412e-03) (hash(x)=152806413)
+1070 train 4.618773 (lr=2.9378e-03) (hash(x)=148403650)
+1080 train 4.597904 (lr=2.9343e-03) (hash(x)=150234604)
+1090 train 4.622439 (lr=2.9307e-03) (hash(x)=153101997)
+1100 val loss 4.6008
+1100 val perplexity 99.5604
+1100 train 4.641544 (lr=2.9270e-03) (hash(x)=154889828)
+1110 train 4.614079 (lr=2.9231e-03) (hash(x)=177021958)
+1120 train 4.615329 (lr=2.9192e-03) (hash(x)=147360351)
+1130 train 4.494336 (lr=2.9152e-03) (hash(x)=149540375)
+1140 train 4.550661 (lr=2.9112e-03) (hash(x)=152068875)
+1150 train 4.414666 (lr=2.9070e-03) (hash(x)=166896658)
+1160 train 4.357806 (lr=2.9027e-03) (hash(x)=161056451)
+1170 train 4.256883 (lr=2.8983e-03) (hash(x)=154114105)
+1180 train 4.389136 (lr=2.8939e-03) (hash(x)=165060550)
+1190 train 4.552098 (lr=2.8893e-03) (hash(x)=166969373)
+1200 val loss 4.5441
+1200 val perplexity 94.0775
+1200 train 4.552005 (lr=2.8847e-03) (hash(x)=160075492)
+1210 train 4.627972 (lr=2.8800e-03) (hash(x)=158231512)
+1220 train 4.496707 (lr=2.8751e-03) (hash(x)=168549643)
+1230 train 4.489129 (lr=2.8702e-03) (hash(x)=155189035)
+1240 train 4.530328 (lr=2.8652e-03) (hash(x)=178512760)
+1250 train 4.415970 (lr=2.8601e-03) (hash(x)=146425022)
+1260 train 4.509857 (lr=2.8550e-03) (hash(x)=159136514)
+1270 train 4.492183 (lr=2.8497e-03) (hash(x)=156001968)
+1280 train 4.466566 (lr=2.8443e-03) (hash(x)=164259999)
+1290 train 4.412362 (lr=2.8389e-03) (hash(x)=161354928)
+1300 val loss 4.4786
+1300 val perplexity 88.1114
+1300 train 4.449751 (lr=2.8333e-03) (hash(x)=158995886)
+1310 train 4.468017 (lr=2.8277e-03) (hash(x)=155383293)
+1320 train 4.442424 (lr=2.8220e-03) (hash(x)=157798110)
+1330 train 4.293063 (lr=2.8162e-03) (hash(x)=150111713)
+1340 train 4.288443 (lr=2.8103e-03) (hash(x)=164386787)
+1350 train 4.436598 (lr=2.8044e-03) (hash(x)=160484704)
+1360 train 4.257019 (lr=2.7983e-03) (hash(x)=169124186)
+1370 train 4.232933 (lr=2.7922e-03) (hash(x)=162460921)
+1380 train 4.455724 (lr=2.7860e-03) (hash(x)=176469537)
+1390 train 4.460512 (lr=2.7797e-03) (hash(x)=171186702)
+1400 val loss 4.4265
+1400 val perplexity 83.6372
+1400 train 4.541006 (lr=2.7733e-03) (hash(x)=166459649)
+1410 train 4.499881 (lr=2.7668e-03) (hash(x)=173618054)
+1420 train 4.461231 (lr=2.7603e-03) (hash(x)=166595245)
+1430 train 4.451481 (lr=2.7536e-03) (hash(x)=161021911)
+1440 train 4.374422 (lr=2.7469e-03) (hash(x)=155839867)
+1450 train 4.508657 (lr=2.7401e-03) (hash(x)=160040811)
+1460 train 4.393574 (lr=2.7333e-03) (hash(x)=160394504)
+1470 train 4.366079 (lr=2.7263e-03) (hash(x)=154986331)
+1480 train 4.434709 (lr=2.7193e-03) (hash(x)=159568067)
+1490 train 4.427073 (lr=2.7121e-03) (hash(x)=170775607)
+1500 val loss 4.4109
+1500 val perplexity 82.3470
+1500 train 4.356575 (lr=2.7050e-03) (hash(x)=152612395)
+1510 train 4.342467 (lr=2.6977e-03) (hash(x)=151245168)
+1520 train 4.249704 (lr=2.6903e-03) (hash(x)=163463453)
+1530 train 4.404288 (lr=2.6829e-03) (hash(x)=168631026)
+1540 train 4.460400 (lr=2.6754e-03) (hash(x)=160083779)
+1550 train 4.404381 (lr=2.6678e-03) (hash(x)=176913716)
+1560 train 4.313577 (lr=2.6602e-03) (hash(x)=149889155)
+1570 train 4.388822 (lr=2.6525e-03) (hash(x)=155207555)
+1580 train 4.262300 (lr=2.6447e-03) (hash(x)=154467728)
+1590 train 4.398449 (lr=2.6368e-03) (hash(x)=163338122)
+1600 val loss 4.3719
+1600 val perplexity 79.1935
+1600 train 4.298754 (lr=2.6289e-03) (hash(x)=153840208)
+1610 train 4.236393 (lr=2.6208e-03) (hash(x)=157648881)
+1620 train 4.278807 (lr=2.6128e-03) (hash(x)=153418765)
+1630 train 4.228697 (lr=2.6046e-03) (hash(x)=156530827)
+1640 train 4.494641 (lr=2.5964e-03) (hash(x)=169919274)
+1650 train 4.349800 (lr=2.5881e-03) (hash(x)=158235114)
+1660 train 4.582599 (lr=2.5797e-03) (hash(x)=163545476)
+1670 train 4.405477 (lr=2.5713e-03) (hash(x)=183148505)
+1680 train 4.306026 (lr=2.5628e-03) (hash(x)=151198383)
+1690 train 4.311630 (lr=2.5542e-03) (hash(x)=159798979)
+1700 val loss 4.3461
+1700 val perplexity 77.1786
+1700 train 4.202085 (lr=2.5455e-03) (hash(x)=153937979)
+1710 train 4.256691 (lr=2.5368e-03) (hash(x)=163663673)
+1720 train 4.307277 (lr=2.5281e-03) (hash(x)=168209417)
+1730 train 4.403978 (lr=2.5192e-03) (hash(x)=169627748)
+1740 train 4.183916 (lr=2.5103e-03) (hash(x)=156391292)
+1750 train 4.195385 (lr=2.5014e-03) (hash(x)=152353279)
+1760 train 4.273163 (lr=2.4924e-03) (hash(x)=169569102)
+1770 train 4.158600 (lr=2.4833e-03) (hash(x)=153444218)
+1780 train 4.237548 (lr=2.4741e-03) (hash(x)=152591697)
+1790 train 4.368547 (lr=2.4649e-03) (hash(x)=176923774)
+1800 val loss 4.3225
+1800 val perplexity 75.3739
+1800 train 4.385180 (lr=2.4556e-03) (hash(x)=160268754)
+1810 train 4.310806 (lr=2.4463e-03) (hash(x)=161145587)
+1820 train 4.322252 (lr=2.4369e-03) (hash(x)=162293952)
+1830 train 4.367172 (lr=2.4275e-03) (hash(x)=179707686)
+1840 train 4.228599 (lr=2.4180e-03) (hash(x)=176247518)
+1850 train 4.274719 (lr=2.4084e-03) (hash(x)=166603610)
+1860 train 4.276209 (lr=2.3988e-03) (hash(x)=165957414)
+1870 train 4.307077 (lr=2.3891e-03) (hash(x)=159922566)
+1880 train 4.184916 (lr=2.3794e-03) (hash(x)=154743265)
+1890 train 4.281101 (lr=2.3696e-03) (hash(x)=148872048)
+1900 val loss 4.3204
+1900 val perplexity 75.2206
+1900 train 4.288528 (lr=2.3598e-03) (hash(x)=167836353)
+1910 train 4.237945 (lr=2.3499e-03) (hash(x)=153969641)
+1920 train 4.360941 (lr=2.3400e-03) (hash(x)=159154377)
+1930 train 4.308533 (lr=2.3300e-03) (hash(x)=166351013)
+1940 train 4.430557 (lr=2.3200e-03) (hash(x)=172131477)
+1950 train 4.269465 (lr=2.3099e-03) (hash(x)=190945203)
+1960 train 4.359749 (lr=2.2998e-03) (hash(x)=161296279)
+1970 train 4.311443 (lr=2.2896e-03) (hash(x)=165088578)
+1980 train 4.319698 (lr=2.2793e-03) (hash(x)=153534765)
+1990 train 4.285650 (lr=2.2691e-03) (hash(x)=155548622)
+2000 val loss 4.2851
+2000 val perplexity 72.6119
+2000 train 4.215850 (lr=2.2588e-03) (hash(x)=163314857)
+2010 train 4.239890 (lr=2.2484e-03) (hash(x)=163661705)
+2020 train 4.106477 (lr=2.2380e-03) (hash(x)=168141011)
+2030 train 4.203107 (lr=2.2275e-03) (hash(x)=153626022)
+2040 train 4.136258 (lr=2.2170e-03) (hash(x)=146318026)
+2050 train 4.217386 (lr=2.2065e-03) (hash(x)=167739815)
+2060 train 4.172077 (lr=2.1959e-03) (hash(x)=154999799)
+2070 train 4.242785 (lr=2.1853e-03) (hash(x)=181501630)
+2080 train 4.259363 (lr=2.1746e-03) (hash(x)=155805721)
+2090 train 4.297694 (lr=2.1639e-03) (hash(x)=158244510)
+2100 val loss 4.2762
+2100 val perplexity 71.9643
+2100 train 4.345075 (lr=2.1532e-03) (hash(x)=147877788)
+2110 train 4.264530 (lr=2.1424e-03) (hash(x)=156688909)
+2120 train 4.268808 (lr=2.1316e-03) (hash(x)=163897683)
+2130 train 4.258202 (lr=2.1208e-03) (hash(x)=166670021)
+2140 train 4.201597 (lr=2.1099e-03) (hash(x)=166034067)
+2150 train 4.219452 (lr=2.0990e-03) (hash(x)=149358136)
+2160 train 4.116647 (lr=2.0881e-03) (hash(x)=156501573)
+2170 train 4.165063 (lr=2.0771e-03) (hash(x)=155673523)
+2180 train 4.153607 (lr=2.0661e-03) (hash(x)=150360430)
+2190 train 4.224645 (lr=2.0550e-03) (hash(x)=176159016)
+2200 val loss 4.2497
+2200 val perplexity 70.0818
+2200 train 4.251291 (lr=2.0440e-03) (hash(x)=165442718)
+2210 train 4.298554 (lr=2.0329e-03) (hash(x)=177529267)
+2220 train 4.209774 (lr=2.0217e-03) (hash(x)=197393885)
+2230 train 4.261082 (lr=2.0106e-03) (hash(x)=169437808)
+2240 train 4.320828 (lr=1.9994e-03) (hash(x)=154335840)
+2250 train 4.269605 (lr=1.9882e-03) (hash(x)=154514509)
+2260 train 4.126194 (lr=1.9770e-03) (hash(x)=158439895)
+2270 train 4.190021 (lr=1.9657e-03) (hash(x)=150326591)
+2280 train 4.207006 (lr=1.9544e-03) (hash(x)=163533419)
+2290 train 4.013283 (lr=1.9431e-03) (hash(x)=150576123)
+2300 val loss 4.2273
+2300 val perplexity 68.5312
+2300 train 4.010305 (lr=1.9318e-03) (hash(x)=149109715)
+2310 train 4.113476 (lr=1.9205e-03) (hash(x)=161192834)
+2320 train 4.081414 (lr=1.9091e-03) (hash(x)=152309013)
+2330 train 4.247547 (lr=1.8977e-03) (hash(x)=144666293)
+2340 train 4.385841 (lr=1.8863e-03) (hash(x)=176885772)
+2350 train 4.373675 (lr=1.8749e-03) (hash(x)=175195635)
+2360 train 4.439493 (lr=1.8635e-03) (hash(x)=153998404)
+2370 train 4.408747 (lr=1.8520e-03) (hash(x)=159393178)
+2380 train 4.194473 (lr=1.8406e-03) (hash(x)=172339628)
+2390 train 4.239025 (lr=1.8291e-03) (hash(x)=146795138)
+2400 val loss 4.2137
+2400 val perplexity 67.6096
+2400 train 4.144070 (lr=1.8176e-03) (hash(x)=162512827)
+2410 train 4.166721 (lr=1.8061e-03) (hash(x)=163561481)
+2420 train 4.190928 (lr=1.7946e-03) (hash(x)=155454305)
+2430 train 4.134076 (lr=1.7830e-03) (hash(x)=166346010)
+2440 train 4.124585 (lr=1.7715e-03) (hash(x)=160028108)
+2450 train 4.075200 (lr=1.7600e-03) (hash(x)=166851234)
+2460 train 4.159366 (lr=1.7484e-03) (hash(x)=161312139)
+2470 train 4.195799 (lr=1.7368e-03) (hash(x)=143069188)
+2480 train 4.278462 (lr=1.7253e-03) (hash(x)=158251395)
+2490 train 4.342030 (lr=1.7137e-03) (hash(x)=158119693)
+2500 val loss 4.1923
+2500 val perplexity 66.1742
+2500 train 4.194202 (lr=1.7021e-03) (hash(x)=148144913)
+2510 train 4.309975 (lr=1.6906e-03) (hash(x)=159474538)
+2520 train 4.160221 (lr=1.6790e-03) (hash(x)=150836192)
+2530 train 4.117862 (lr=1.6674e-03) (hash(x)=156823868)
+2540 train 4.278078 (lr=1.6558e-03) (hash(x)=155702319)
+2550 train 4.089338 (lr=1.6442e-03) (hash(x)=160471487)
+2560 train 4.020177 (lr=1.6326e-03) (hash(x)=160276214)
+2570 train 4.144533 (lr=1.6210e-03) (hash(x)=167637000)
+2580 train 4.077360 (lr=1.6094e-03) (hash(x)=149819695)
+2590 train 4.112842 (lr=1.5979e-03) (hash(x)=160749905)
+2600 val loss 4.1872
+2600 val perplexity 65.8381
+2600 train 4.059939 (lr=1.5863e-03) (hash(x)=155617603)
+2610 train 4.389883 (lr=1.5747e-03) (hash(x)=162181794)
+2620 train 4.238740 (lr=1.5632e-03) (hash(x)=171300086)
+2630 train 4.230862 (lr=1.5516e-03) (hash(x)=159707436)
+2640 train 4.330113 (lr=1.5400e-03) (hash(x)=137869312)
+2650 train 4.244238 (lr=1.5285e-03) (hash(x)=115611274)
+2660 train 4.129931 (lr=1.5170e-03) (hash(x)=151140502)
+2670 train 4.198371 (lr=1.5054e-03) (hash(x)=162821182)
+2680 train 4.208638 (lr=1.4939e-03) (hash(x)=174840522)
+2690 train 4.182845 (lr=1.4824e-03) (hash(x)=161323382)
+2700 val loss 4.1596
+2700 val perplexity 64.0478
+2700 train 4.060623 (lr=1.4709e-03) (hash(x)=159630183)
+2710 train 4.191040 (lr=1.4594e-03) (hash(x)=169557486)
+2720 train 4.102142 (lr=1.4480e-03) (hash(x)=154059878)
+2730 train 4.108820 (lr=1.4365e-03) (hash(x)=153014579)
+2740 train 4.080665 (lr=1.4251e-03) (hash(x)=152040333)
+2750 train 4.264620 (lr=1.4137e-03) (hash(x)=164487951)
+2760 train 4.202245 (lr=1.4023e-03) (hash(x)=189559478)
+2770 train 4.239017 (lr=1.3909e-03) (hash(x)=152017273)
+2780 train 4.212034 (lr=1.3795e-03) (hash(x)=168478683)
+2790 train 4.203107 (lr=1.3682e-03) (hash(x)=167047337)
+2800 val loss 4.1428
+2800 val perplexity 62.9782
+2800 train 4.212909 (lr=1.3569e-03) (hash(x)=153213767)
+2810 train 4.065434 (lr=1.3456e-03) (hash(x)=141332160)
+2820 train 4.086657 (lr=1.3343e-03) (hash(x)=158788151)
+2830 train 4.098769 (lr=1.3230e-03) (hash(x)=161375220)
+2840 train 4.010297 (lr=1.3118e-03) (hash(x)=162935851)
+2850 train 3.959760 (lr=1.3006e-03) (hash(x)=159464909)
+2860 train 4.018600 (lr=1.2894e-03) (hash(x)=156126362)
+2870 train 3.995692 (lr=1.2783e-03) (hash(x)=161380583)
+2880 train 4.089808 (lr=1.2671e-03) (hash(x)=156964606)
+2890 train 4.159009 (lr=1.2560e-03) (hash(x)=169458983)
+2900 val loss 4.1383
+2900 val perplexity 62.6992
+2900 train 4.252225 (lr=1.2450e-03) (hash(x)=176573157)
+2910 train 4.147831 (lr=1.2339e-03) (hash(x)=153552099)
+2920 train 4.087148 (lr=1.2229e-03) (hash(x)=165236334)
+2930 train 4.178411 (lr=1.2119e-03) (hash(x)=165262998)
+2940 train 4.043644 (lr=1.2010e-03) (hash(x)=155086138)
+2950 train 4.113328 (lr=1.1901e-03) (hash(x)=150891340)
+2960 train 4.104153 (lr=1.1792e-03) (hash(x)=155191944)
+2970 train 4.117688 (lr=1.1684e-03) (hash(x)=163561216)
+2980 train 4.071639 (lr=1.1576e-03) (hash(x)=161288249)
+2990 train 3.980731 (lr=1.1468e-03) (hash(x)=164593955)
+3000 val loss 4.1153
+3000 val perplexity 61.2722
+3000 train 4.045059 (lr=1.1361e-03) (hash(x)=161887650)
+3010 train 4.067009 (lr=1.1254e-03) (hash(x)=158376770)
+3020 train 4.017281 (lr=1.1147e-03) (hash(x)=158653093)
+3030 train 4.075210 (lr=1.1041e-03) (hash(x)=157601570)
+3040 train 4.109300 (lr=1.0935e-03) (hash(x)=162074203)
+3050 train 4.156050 (lr=1.0830e-03) (hash(x)=171875870)
+3060 train 4.153726 (lr=1.0725e-03) (hash(x)=152843048)
+3070 train 4.173660 (lr=1.0620e-03) (hash(x)=165808171)
+3080 train 4.153944 (lr=1.0516e-03) (hash(x)=158272669)
+3090 train 4.178918 (lr=1.0412e-03) (hash(x)=157688554)
+3100 val loss 4.1072
+3100 val perplexity 60.7747
+3100 train 4.169457 (lr=1.0309e-03) (hash(x)=166005015)
+3110 train 4.098954 (lr=1.0207e-03) (hash(x)=158763736)
+3120 train 4.144180 (lr=1.0104e-03) (hash(x)=167439328)
+3130 train 4.267995 (lr=1.0002e-03) (hash(x)=148048338)
+3140 train 4.194165 (lr=9.9011e-04) (hash(x)=164522898)
+3150 train 4.171941 (lr=9.8002e-04) (hash(x)=157242118)
+3160 train 4.104424 (lr=9.6999e-04) (hash(x)=149556058)
+3170 train 4.142094 (lr=9.6000e-04) (hash(x)=148563572)
+3180 train 4.170441 (lr=9.5007e-04) (hash(x)=174824363)
+3190 train 4.313228 (lr=9.4019e-04) (hash(x)=159427153)
+3200 val loss 4.0829
+3200 val perplexity 59.3162
+3200 train 4.204639 (lr=9.3036e-04) (hash(x)=161394675)
+3210 train 4.181211 (lr=9.2058e-04) (hash(x)=166134016)
+3220 train 4.240358 (lr=9.1085e-04) (hash(x)=154731855)
+3230 train 4.074627 (lr=9.0118e-04) (hash(x)=156158458)
+3240 train 3.989538 (lr=8.9157e-04) (hash(x)=173128783)
+3250 train 4.067182 (lr=8.8201e-04) (hash(x)=154070645)
+3260 train 4.008566 (lr=8.7251e-04) (hash(x)=153348463)
+3270 train 4.081103 (lr=8.6307e-04) (hash(x)=156976700)
+3280 train 4.073939 (lr=8.5368e-04) (hash(x)=155420483)
+3290 train 3.990471 (lr=8.4435e-04) (hash(x)=149021726)
+3300 val loss 4.0744
+3300 val perplexity 58.8180
+3300 train 4.036510 (lr=8.3508e-04) (hash(x)=160700347)
+3310 train 4.076362 (lr=8.2588e-04) (hash(x)=165468658)
+3320 train 4.130893 (lr=8.1673e-04) (hash(x)=163266950)
+3330 train 4.154663 (lr=8.0764e-04) (hash(x)=170100446)
+3340 train 4.144356 (lr=7.9862e-04) (hash(x)=151366602)
+3350 train 4.228275 (lr=7.8966e-04) (hash(x)=150644566)
+3360 train 4.051836 (lr=7.8076e-04) (hash(x)=164233827)
+3370 train 4.121541 (lr=7.7192e-04) (hash(x)=164188603)
+3380 train 4.044434 (lr=7.6315e-04) (hash(x)=168215736)
+3390 train 3.963793 (lr=7.5445e-04) (hash(x)=162692228)
+3400 val loss 4.0643
+3400 val perplexity 58.2231
+3400 train 4.118140 (lr=7.4581e-04) (hash(x)=148500491)
+3410 train 3.952079 (lr=7.3724e-04) (hash(x)=139154138)
+3420 train 3.948520 (lr=7.2874e-04) (hash(x)=181460455)
+3430 train 4.048703 (lr=7.2030e-04) (hash(x)=178896304)
+3440 train 4.054691 (lr=7.1193e-04) (hash(x)=149701574)
+3450 train 4.040005 (lr=7.0363e-04) (hash(x)=178339553)
+3460 train 4.060825 (lr=6.9541e-04) (hash(x)=163112813)
+3470 train 4.126402 (lr=6.8725e-04) (hash(x)=158012268)
+3480 train 4.108067 (lr=6.7916e-04) (hash(x)=157067275)
+3490 train 4.109801 (lr=6.7114e-04) (hash(x)=163370975)
+3500 val loss 4.0477
+3500 val perplexity 57.2672
+3500 train 4.074748 (lr=6.6320e-04) (hash(x)=160383657)
+3510 train 4.005992 (lr=6.5533e-04) (hash(x)=156523307)
+3520 train 4.052684 (lr=6.4753e-04) (hash(x)=167614080)
+3530 train 4.122105 (lr=6.3981e-04) (hash(x)=163104143)
+3540 train 4.020593 (lr=6.3216e-04) (hash(x)=158510849)
+3550 train 4.060208 (lr=6.2458e-04) (hash(x)=151149010)
+3560 train 4.146153 (lr=6.1708e-04) (hash(x)=166504920)
+3570 train 4.101566 (lr=6.0966e-04) (hash(x)=177141678)
+3580 train 3.943745 (lr=6.0231e-04) (hash(x)=157572718)
+3590 train 4.301079 (lr=5.9504e-04) (hash(x)=155138495)
+3600 val loss 4.0360
+3600 val perplexity 56.5982
+3600 train 4.128091 (lr=5.8785e-04) (hash(x)=165538372)
+3610 train 4.094797 (lr=5.8074e-04) (hash(x)=153930045)
+3620 train 4.181409 (lr=5.7370e-04) (hash(x)=148944287)
+3630 train 4.103976 (lr=5.6675e-04) (hash(x)=152951123)
+3640 train 4.078807 (lr=5.5987e-04) (hash(x)=152974720)
+3650 train 4.008730 (lr=5.5308e-04) (hash(x)=155442773)
+3660 train 3.991292 (lr=5.4636e-04) (hash(x)=163439074)
+3670 train 4.094393 (lr=5.3973e-04) (hash(x)=162484510)
+3680 train 4.057738 (lr=5.3318e-04) (hash(x)=171702529)
+3690 train 3.945738 (lr=5.2671e-04) (hash(x)=156564909)
+3700 val loss 4.0327
+3700 val perplexity 56.4110
+3700 train 3.893731 (lr=5.2033e-04) (hash(x)=140880241)
+3710 train 3.991026 (lr=5.1402e-04) (hash(x)=154078620)
+3720 train 3.993203 (lr=5.0780e-04) (hash(x)=157029930)
+3730 train 4.097836 (lr=5.0167e-04) (hash(x)=159656742)
+3740 train 4.067860 (lr=4.9562e-04) (hash(x)=166856858)
+3750 train 4.001625 (lr=4.8965e-04) (hash(x)=163228730)
+3760 train 4.046811 (lr=4.8377e-04) (hash(x)=172270929)
+3770 train 3.992501 (lr=4.7798e-04) (hash(x)=158075162)
+3780 train 4.054654 (lr=4.7227e-04) (hash(x)=151894909)
+3790 train 4.047204 (lr=4.6665e-04) (hash(x)=166661644)
+3800 val loss 4.0143
+3800 val perplexity 55.3849
+3800 train 3.981267 (lr=4.6112e-04) (hash(x)=169003916)
+3810 train 4.078616 (lr=4.5567e-04) (hash(x)=158320575)
+3820 train 3.984195 (lr=4.5031e-04) (hash(x)=156723804)
+3830 train 3.941811 (lr=4.4504e-04) (hash(x)=151045721)
+3840 train 4.053659 (lr=4.3986e-04) (hash(x)=161384988)
+3850 train 3.755682 (lr=4.3477e-04) (hash(x)=153163470)
+3860 train 3.780734 (lr=4.2977e-04) (hash(x)=159294553)
+3870 train 3.869576 (lr=4.2486e-04) (hash(x)=149254649)
+3880 train 3.852206 (lr=4.2004e-04) (hash(x)=169983073)
+3890 train 3.711509 (lr=4.1530e-04) (hash(x)=166230303)
+3900 val loss 4.0204
+3900 val perplexity 55.7224
+3900 train 4.064318 (lr=4.1066e-04) (hash(x)=150193462)
+3910 train 4.144692 (lr=4.0611e-04) (hash(x)=165783034)
+3920 train 4.032724 (lr=4.0166e-04) (hash(x)=170963531)
+3930 train 4.107780 (lr=3.9729e-04) (hash(x)=170794255)
+3940 train 4.061764 (lr=3.9302e-04) (hash(x)=173233324)
+3950 train 4.107297 (lr=3.8884e-04) (hash(x)=155491235)
+3960 train 4.191706 (lr=3.8475e-04) (hash(x)=151130973)
+3970 train 4.034113 (lr=3.8076e-04) (hash(x)=160788059)
+3980 train 4.079150 (lr=3.7685e-04) (hash(x)=154935745)
+3990 train 3.916988 (lr=3.7305e-04) (hash(x)=167636525)
+4000 val loss 4.0012
+4000 val perplexity 54.6617
+4000 train 3.991260 (lr=3.6933e-04) (hash(x)=181909918)
+4010 train 3.959821 (lr=3.6572e-04) (hash(x)=156869134)
+4020 train 3.902684 (lr=3.6219e-04) (hash(x)=161833430)
+4030 train 4.110707 (lr=3.5876e-04) (hash(x)=160284977)
+4040 train 3.812346 (lr=3.5543e-04) (hash(x)=147725752)
+4050 train 3.790784 (lr=3.5219e-04) (hash(x)=154595084)
+4060 train 3.849233 (lr=3.4905e-04) (hash(x)=175075232)
+4070 train 3.858919 (lr=3.4600e-04) (hash(x)=151811717)
+4080 train 3.968429 (lr=3.4305e-04) (hash(x)=157540169)
+4090 train 4.093104 (lr=3.4019e-04) (hash(x)=165996843)
+4100 val loss 3.9981
+4100 val perplexity 54.4943
+4100 train 4.030262 (lr=3.3744e-04) (hash(x)=160351198)
+4110 train 4.054459 (lr=3.3477e-04) (hash(x)=168096977)
+4120 train 3.993429 (lr=3.3221e-04) (hash(x)=162609202)
+4130 train 4.001949 (lr=3.2974e-04) (hash(x)=156667217)
+4140 train 4.026743 (lr=3.2737e-04) (hash(x)=152750290)
+4150 train 3.992709 (lr=3.2510e-04) (hash(x)=169619396)
+4160 train 4.048236 (lr=3.2292e-04) (hash(x)=159107550)
+4170 train 3.999295 (lr=3.2085e-04) (hash(x)=155951176)
+4180 train 3.949090 (lr=3.1887e-04) (hash(x)=166284453)
+4190 train 4.076835 (lr=3.1699e-04) (hash(x)=152671857)
+4200 val loss 3.9917
+4200 val perplexity 54.1493
+4200 train 4.041010 (lr=3.1520e-04) (hash(x)=146023049)
+4210 train 3.958335 (lr=3.1352e-04) (hash(x)=172064951)
+4220 train 3.529423 (lr=3.1193e-04) (hash(x)=149621476)
+4230 train 3.736262 (lr=3.1044e-04) (hash(x)=146485058)
+4240 train 3.796316 (lr=3.0905e-04) (hash(x)=164185738)
+4250 train 3.705650 (lr=3.0776e-04) (hash(x)=158666667)
+4260 train 4.052936 (lr=3.0657e-04) (hash(x)=167422908)
+4270 train 4.019994 (lr=3.0548e-04) (hash(x)=162494922)
+4280 train 4.046692 (lr=3.0449e-04) (hash(x)=160033749)
+4290 train 4.011508 (lr=3.0359e-04) (hash(x)=154077913)
+4300 val loss 3.9888
+4300 val perplexity 53.9885
+4300 train 4.057397 (lr=3.0280e-04) (hash(x)=170440016)
+4310 train 4.041889 (lr=3.0210e-04) (hash(x)=159947398)
+4320 train 4.053513 (lr=3.0150e-04) (hash(x)=162316772)
+4330 train 4.149829 (lr=3.0101e-04) (hash(x)=153263344)
+4340 train 3.910232 (lr=3.0061e-04) (hash(x)=155575816)
+4350 train 3.930528 (lr=3.0031e-04) (hash(x)=156835929)
+4360 train 4.005615 (lr=3.0011e-04) (hash(x)=152507945)
+4370 train 3.949054 (lr=3.0001e-04) (hash(x)=164438795)
+4374 val loss 3.9859
+4374 val perplexity 53.8337

scaling_sdpa_granularity_2/hd_128_seed_1340/model_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9ecdfb171b0fcab6eac66dda24b58f9c1b65607764ac2031849be0e3f34adc0
+size 89692138

scaling_sdpa_granularity_2/hd_128_seed_1340/optimizer_04374.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32e2ce53dafb95565714b5850985f172b91ad634918b3dd9117deb0bd14959e0
+size 179406342