Upload folder using huggingface_hub

attention_kindself_n_heads2_seed1338/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_2/attention_kindself_n_heads2_seed1338", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 2, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 100, "warmup_steps": 200, "group": "wider_is_better_2", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1338, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "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": 80, "total_batch_size": 20480, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": true, "disable_selection": true, "mup_enable_coord_check_logging": false, "max_lr": 0.00015, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "1.5e-4_20480_2_1338", "n_embd": 128}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_2/attention_kindself_n_heads2_seed1338", "resume_checkpoint": null, "resume_optimizer": false, "add_a_head": false, "add_head_to_start": true, "new_head_init": "normal", "n_heads": 2, "protect_bos_token": true, "prevent_from_masking_myself": true, "max_steps": 1000, "warmup_steps": 200, "group": "wider_is_better_2", "use_wandb": true, "kill_self_after_run": false, "random_seed": 1338, "memory_penalty_epsilon": 0.1, "selection_head_linear_combo": "none", "selection_head_linear_combo_scale": 1.0, "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": 80, "total_batch_size": 20480, "protection_head_scaling_factor": 1.0, "protection_head_bias": 0.0, "n_sliced_masks": null, "n_latent_masks": null, "mask_layernorm": false, "residual_attention_masks": false, "compute_base_shapes": false, "base_shapes_savefile": null, "mup": true, "disable_selection": true, "mup_enable_coord_check_logging": false, "max_lr": 0.00015, "decay_lr": true, "readout_zero_init": false, "query_zero_init": false, "l1_loss": false, "debugpy": false, "key": "1.5e-4_20480_2_1338", "n_embd": 128}

attention_kindself_n_heads2_seed1338/dataloader_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8513131d2bf732f8f6d97ff82d32c505b2a2f3f9f2247bac315bb09ce0e1f97
+size 964

attention_kindself_n_heads2_seed1338/log2.txt

@@ -1,8 +1,53 @@
-max_steps: 100
+max_steps: 1000
 0 val loss 11.7388
 0 val perplexity 125337.7812
-0 train 11.765342 (lr=7.5000e-07) (hash(x)=99006591)
-50 val loss 10.4093
-50 val perplexity 33165.7539
-99 val loss 9.7923
-99 val perplexity 17895.2012
+0 train 11.765713 (lr=7.5000e-07) (hash(x)=99006591)
+50 val loss 10.4110
+50 val perplexity 33221.8164
+100 val loss 9.7642
+100 val perplexity 17399.4941
+100 train 9.716583 (lr=7.5750e-05) (hash(x)=92897581)
+150 val loss 8.7169
+150 val perplexity 6105.4497
+200 val loss 8.0353
+200 val perplexity 3088.0925
+200 train 7.986736 (lr=1.5000e-04) (hash(x)=102397744)
+250 val loss 7.7502
+250 val perplexity 2322.1006
+300 val loss 7.6778
+300 val perplexity 2159.9612
+300 train 7.528432 (lr=1.4486e-04) (hash(x)=96438273)
+350 val loss 7.6504
+350 val perplexity 2101.4463
+400 val loss 7.6091
+400 val perplexity 2016.5544
+400 train 7.386041 (lr=1.3023e-04) (hash(x)=96121843)
+450 val loss 7.5694
+450 val perplexity 1937.9167
+500 val loss 7.5382
+500 val perplexity 1878.3588
+500 train 7.485167 (lr=1.0833e-04) (hash(x)=102538951)
+550 val loss 7.5006
+550 val perplexity 1809.1721
+600 val loss 7.4816
+600 val perplexity 1775.0961
+600 train 7.258781 (lr=8.2500e-05) (hash(x)=98779605)
+650 val loss 7.4589
+650 val perplexity 1735.1646
+700 val loss 7.4300
+700 val perplexity 1685.8812
+700 train 7.137525 (lr=5.6669e-05) (hash(x)=95523522)
+750 val loss 7.4134
+750 val perplexity 1657.9785
+800 val loss 7.3962
+800 val perplexity 1629.7191
+800 train 7.294775 (lr=3.4770e-05) (hash(x)=110845816)
+850 val loss 7.3761
+850 val perplexity 1597.4272
+900 val loss 7.3491
+900 val perplexity 1554.7263
+900 train 7.459410 (lr=2.0138e-05) (hash(x)=108356411)
+950 val loss 7.3407
+950 val perplexity 1541.7485
+999 val loss 7.3253
+999 val perplexity 1518.2883

attention_kindself_n_heads2_seed1338/model_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ccf9cd6c291705d92caa7c5e3d82636245985ecf2f619ec076909c2042809ec
+size 38587970

attention_kindself_n_heads2_seed1338/optimizer_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d90a4f03006480f2ea1f9a2fa705578bf9ee93f3ace6925b9f903ed8819fe55
+size 70895430