rerun softmax0/1 15M logging kurtosis over trainstep

out/softmax1-15m-2023_08_25_11_40_38/config.json

@@ -1 +0,0 @@
-{"out_dir": "out/softmax1-15m-2023_08_25_11_40_38", "eval_interval": 1000, "log_interval": 1, "eval_iters": 50, "eval_only": false, "always_save_checkpoint": true, "init_from": "scratch", "wandb_log": true, "wandb_project": "softmax1-tinystories", "wandb_run_name": "softmax1-15m-2023_08_25_11_40_38", "batch_size": 96, "max_seq_len": 256, "vocab_source": "llama2", "vocab_size": 32000, "dim": 288, "n_layers": 6, "n_heads": 6, "n_kv_heads": 6, "multiple_of": 32, "dropout": 0.0, "gradient_accumulation_steps": 4, "learning_rate": 0.0005, "max_iters": 100000, "weight_decay": 0.1, "beta1": 0.9, "beta2": 0.95, "grad_clip": 1.0, "decay_lr": true, "warmup_iters": 1000, "device": "cuda", "dtype": "float16", "compile": true, "softmax1": true}

out/softmax1-15m-2023_08_25_11_42_26/config.json

@@ -1 +0,0 @@
-{"out_dir": "out/softmax1-15m-2023_08_25_11_42_26", "eval_interval": 1000, "log_interval": 1, "eval_iters": 50, "eval_only": false, "always_save_checkpoint": true, "init_from": "scratch", "wandb_log": true, "wandb_project": "softmax1-tinystories", "wandb_run_name": "softmax1-15m-2023_08_25_11_42_26", "batch_size": 96, "max_seq_len": 256, "vocab_source": "llama2", "vocab_size": 32000, "dim": 288, "n_layers": 6, "n_heads": 6, "n_kv_heads": 6, "multiple_of": 32, "dropout": 0.0, "gradient_accumulation_steps": 4, "learning_rate": 0.0005, "max_iters": 100000, "weight_decay": 0.1, "beta1": 0.9, "beta2": 0.95, "grad_clip": 1.0, "decay_lr": true, "warmup_iters": 1000, "device": "cuda", "dtype": "float16", "compile": true, "softmax1": true}

out/softmax1-15m-2023_08_25_11_45_19/config.json

@@ -1 +0,0 @@
-{"out_dir": "out/softmax1-15m-2023_08_25_11_45_19", "eval_interval": 1000, "log_interval": 1, "eval_iters": 50, "eval_only": false, "always_save_checkpoint": true, "init_from": "scratch", "wandb_log": true, "wandb_project": "softmax1-tinystories", "wandb_run_name": "softmax1-15m-2023_08_25_11_45_19", "batch_size": 96, "max_seq_len": 256, "vocab_source": "llama2", "vocab_size": 32000, "dim": 288, "n_layers": 6, "n_heads": 6, "n_kv_heads": 6, "multiple_of": 32, "dropout": 0.0, "gradient_accumulation_steps": 4, "learning_rate": 0.0005, "max_iters": 100000, "weight_decay": 0.1, "beta1": 0.9, "beta2": 0.95, "grad_clip": 1.0, "decay_lr": true, "warmup_iters": 1000, "device": "cuda", "dtype": "float16", "compile": false, "softmax1": true}