| {"model_name": "gpt2-small", "model_class_name": "HookedTransformer", "hook_point": "blocks.10.hook_resid_pre", "hook_point_eval": "blocks.{layer}.attn.pattern", "hook_point_layer": 10, "hook_point_head_index": null, "dataset_path": "apollo-research/Skylion007-openwebtext-tokenizer-gpt2", "is_dataset_tokenized": true, "context_size": 128, "use_cached_activations": false, "cached_activations_path": null, "d_in": 768, "d_sae": 46080, "b_dec_init_method": "geometric_median", "expansion_factor": 60, "activation_fn": "relu", "normalize_sae_decoder": true, "noise_scale": 0.0, "from_pretrained_path": null, "apply_b_dec_to_input": false, "decoder_orthogonal_init": true, "n_batches_in_buffer": 8, "training_tokens": 2000000, "finetuning_tokens": 0, "store_batch_size": 16, "train_batch_size": 4096, "device": "mps", "seed": 42, "dtype": "torch.float32", "prepend_bos": false, "adam_beta1": 0, "adam_beta2": 0.999, "mse_loss_normalization": null, "l1_coefficient": 0.001, "lp_norm": 1, "lr": 0.0003, "lr_scheduler_name": "constant", "lr_warm_up_steps": 500, "lr_end": 2.9999999999999997e-05, "lr_decay_steps": 0, "n_restart_cycles": 1, "finetuning_method": null, "use_ghost_grads": false, "feature_sampling_window": 2000, "dead_feature_window": 1000, "dead_feature_threshold": 1e-08, "log_to_wandb": true, "wandb_project": "mats_sae_training_language_model", "run_name": "46080-L1-0.001-LR-0.0003-Tokens-2.000e+06", "wandb_entity": null, "wandb_log_frequency": 10, "n_checkpoints": 0, "checkpoint_path": "8crnit9h/lrkvecno", "verbose": false, "model_kwargs": {}, "tokens_per_buffer": 4194304} |