{"sae": {"expansion_factor": 16, "normalize_decoder": true, "num_latents": 0, "k": -1, "multi_topk": false, "jumprelu": true, "jumprelu_init_threshold": 0.001, "jumprelu_bandwidth": 0.001, "jumprelu_target_l0": null, "init_enc_as_dec_transpose": true}, "batch_size": 4, "max_seq_len": 1024, "num_training_tokens": 1000000000, "cycle_iterator": true, "grad_acc_steps": 1, "micro_acc_steps": 1, "adam_8bit": false, "adam_epsilon": 1e-08, "adam_betas": [0.0, 0.999], "lr": 0.0007, "lr_init": 7e-05, "lr_end": 7e-05, "lr_scheduler_name": "constant", "lr_warmup_steps": 0.01, "lr_decay_steps": 0.2, "l1_coefficient": 0.0003, "l1_warmup_steps": 0.05, "use_l2_loss": true, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["layers.0", "layers.1", "layers.2", "layers.3", "layers.4", "layers.5", "layers.6", "layers.7", "layers.8", "layers.9", "layers.10"], "layers": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10], "layer_stride": 1, "distribute_modules": false, "save_every": 50000, "normalize_activations": 1, "num_norm_estimation_tokens": 2000000, "clusters": null, "cluster_hookpoints": null, "hook": null, "log_to_wandb": true, "run_name": "checkpoints/pythia-160m-deduped-1024-lambda-0.0003-target-L0-None-lr-0.0007", "wandb_log_frequency": 1}