Upload folder using huggingface_hub

attention_kindself_n_heads2_seed1339/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_2/attention_kindself_n_heads2_seed1339", "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": 1339, "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_1339", "n_embd": 128}
+{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_2/attention_kindself_n_heads2_seed1339", "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": 1339, "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_1339", "n_embd": 128}

attention_kindself_n_heads2_seed1339/dataloader_00999.pt

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

attention_kindself_n_heads2_seed1339/log2.txt

@@ -1,8 +1,53 @@
-max_steps: 100
-0 val loss 11.7348
-0 val perplexity 124842.5859
+max_steps: 1000
+0 val loss 11.7346
+0 val perplexity 124815.0859
 0 train 11.725669 (lr=7.5000e-07) (hash(x)=107123132)
-50 val loss 10.5239
-50 val perplexity 37194.7500
-99 val loss 9.8733
-99 val perplexity 19404.9863
+50 val loss 10.5248
+50 val perplexity 37227.2578
+100 val loss 9.9258
+100 val perplexity 20450.8223
+100 train 9.942008 (lr=7.5750e-05) (hash(x)=101193890)
+150 val loss 8.8761
+150 val perplexity 7159.0596
+200 val loss 8.0385
+200 val perplexity 3097.8174
+200 train 7.907970 (lr=1.5000e-04) (hash(x)=95079403)
+250 val loss 7.7222
+250 val perplexity 2257.9138
+300 val loss 7.6657
+300 val perplexity 2133.9622
+300 train 7.533260 (lr=1.4486e-04) (hash(x)=96917853)
+350 val loss 7.6427
+350 val perplexity 2085.2805
+400 val loss 7.6271
+400 val perplexity 2053.1021
+400 train 7.480326 (lr=1.3023e-04) (hash(x)=93672417)
+450 val loss 7.6040
+450 val perplexity 2006.2183
+500 val loss 7.5738
+500 val perplexity 1946.4379
+500 train 7.482606 (lr=1.0833e-04) (hash(x)=98036365)
+550 val loss 7.5469
+550 val perplexity 1894.9224
+600 val loss 7.5130
+600 val perplexity 1831.6515
+600 train 7.561394 (lr=8.2500e-05) (hash(x)=100769188)
+650 val loss 7.4800
+650 val perplexity 1772.1656
+700 val loss 7.4568
+700 val perplexity 1731.6459
+700 train 7.405756 (lr=5.6669e-05) (hash(x)=102854282)
+750 val loss 7.4347
+750 val perplexity 1693.7745
+800 val loss 7.4204
+800 val perplexity 1669.6633
+800 train 7.337776 (lr=3.4770e-05) (hash(x)=99815715)
+850 val loss 7.4062
+850 val perplexity 1646.2092
+900 val loss 7.3941
+900 val perplexity 1626.3074
+900 train 7.146854 (lr=2.0138e-05) (hash(x)=96846928)
+950 val loss 7.3838
+950 val perplexity 1609.6306
+999 val loss 7.3658
+999 val perplexity 1581.0527

attention_kindself_n_heads2_seed1339/model_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b063150f8139bc0b5fab6e173ef0a5c9dd780f83ea4fac6cee844f85205f0615
+size 38587970

attention_kindself_n_heads2_seed1339/optimizer_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4529eb3bfacb30030865096703ae5766e7edd1bd8b151c714defd6bace0deb70
+size 70895430