resid_post_layer_19/trainer_1/config.json

{
    "trainer": {
        "trainer_class": "BatchTopKTrainer",
        "dict_class": "BatchTopKSAE",
        "lr": 0.0001,
        "steps": 244140,
        "auxk_alpha": 0.03125,
        "warmup_steps": 1000,
        "decay_start": 195312,
        "threshold_beta": 0.999,
        "threshold_start_step": 1000,
        "top_k_aux": 2048,
        "seed": 0,
        "activation_dim": 4096,
        "dict_size": 131072,
        "k": 64,
        "device": "cuda",
        "layer": 19,
        "lm_name": "meta-llama/Llama-3.1-8B-Instruct",
        "wandb_name": "3_l19-BatchTopKTrainer-meta-llama/Llama-3.1-8B-Instruct-resid_post_layer_19_trainer_1",
        "submodule_name": "resid_post_layer_19"
    },
    "buffer": {
        "d_submodule": 4096,
        "io": "out",
        "n_ctxs": 244,
        "ctx_len": 1024,
        "refresh_batch_size": 16,
        "out_batch_size": 2048,
        "device": "cuda",
        "internal_device": "cuda"
    }
}