Upload folder using huggingface_hub

attention_kindselective_n_heads2_seed1339/args.json

@@ -1 +1 @@
-{"hellaswag": true, "attention_kind": "selective", "log_dir": "wider_is_better_2/attention_kindselective_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": false, "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_kindselective_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": false, "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_kindselective_n_heads2_seed1339/dataloader_00999.pt

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

attention_kindselective_n_heads2_seed1339/log2.txt

@@ -1,8 +1,53 @@
-max_steps: 100
+max_steps: 1000
 0 val loss 11.7471
 0 val perplexity 126386.8516
 0 train 11.738260 (lr=7.5000e-07) (hash(x)=107123132)
-50 val loss 10.5025
-50 val perplexity 36404.6875
-99 val loss 9.5862
-99 val perplexity 14562.5625
+50 val loss 10.5008
+50 val perplexity 36345.9922
+100 val loss 9.5989
+100 val perplexity 14748.7441
+100 train 9.610984 (lr=7.5750e-05) (hash(x)=101193890)
+150 val loss 8.2620
+150 val perplexity 3873.8713
+200 val loss 7.7626
+200 val perplexity 2350.9575
+200 train 7.601669 (lr=1.5000e-04) (hash(x)=95079403)
+250 val loss 7.6888
+250 val perplexity 2183.7668
+300 val loss 7.6583
+300 val perplexity 2118.1362
+300 train 7.509525 (lr=1.4486e-04) (hash(x)=96917853)
+350 val loss 7.6596
+350 val perplexity 2120.8447
+400 val loss 7.6557
+400 val perplexity 2112.6025
+400 train 7.508607 (lr=1.3023e-04) (hash(x)=93672417)
+450 val loss 7.6534
+450 val perplexity 2107.7827
+500 val loss 7.6489
+500 val perplexity 2098.4023
+500 train 7.561789 (lr=1.0833e-04) (hash(x)=98036365)
+550 val loss 7.6362
+550 val perplexity 2071.9031
+600 val loss 7.6217
+600 val perplexity 2042.1052
+600 train 7.674348 (lr=8.2500e-05) (hash(x)=100769188)
+650 val loss 7.6124
+650 val perplexity 2023.1761
+700 val loss 7.6120
+700 val perplexity 2022.3361
+700 train 7.566431 (lr=5.6669e-05) (hash(x)=102854282)
+750 val loss 7.6004
+750 val perplexity 1999.0525
+800 val loss 7.5911
+800 val perplexity 1980.5055
+800 train 7.497270 (lr=3.4770e-05) (hash(x)=99815715)
+850 val loss 7.5890
+850 val perplexity 1976.2594
+900 val loss 7.5804
+900 val perplexity 1959.4106
+900 train 7.345047 (lr=2.0138e-05) (hash(x)=96846928)
+950 val loss 7.5746
+950 val perplexity 1948.0265
+999 val loss 7.5653
+999 val perplexity 1929.9879

attention_kindselective_n_heads2_seed1339/model_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5b9427e25cd8c794fcd6493afec02b0ed8412464d8210605219577d79a8b8c17
+size 38587970

attention_kindselective_n_heads2_seed1339/optimizer_00999.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f956661af1400c1252eb28b994c5ebea1c46edd35433ef521a03303b0bb600d
+size 70895430