| {"sae": {"expansion_factor": 64, "normalize_decoder": true, "num_latents": 0, "k": 32, "multi_topk": false}, "batch_size": 4, "grad_acc_steps": 8, "micro_acc_steps": 2, "lr": null, "lr_warmup_steps": 1000, "auxk_alpha": 0.0, "dead_feature_threshold": 10000000, "hookpoints": ["gpt_neox.layers.0", "gpt_neox.layers.1", "gpt_neox.layers.2", "gpt_neox.layers.3", "gpt_neox.layers.4", "gpt_neox.layers.5"], "layers": [0, 1, 2, 3, 4, 5], "layer_stride": 1, "distribute_modules": false, "save_every": 1000, "max_tokens": 1000000000, "log_to_wandb": false, "run_name": "/home/random_sae/experiments/saved_models/pythia-70m-deduped_64_k32/redpajama-data-1t-sample_plain_text_100M_rerandomised", "wandb_log_frequency": 1} |