matryoshka_othello/trainer_12/config.json

{
    "trainer": {
        "trainer_class": "MatryoshkaBatchTopKTrainer",
        "dict_class": "MatryoshkaBatchTopKSAE",
        "lr": 0.0004,
        "steps": 36621,
        "auxk_alpha": 0.03125,
        "warmup_steps": 1000,
        "decay_start": null,
        "threshold_beta": 0.999,
        "threshold_start_step": 1000,
        "top_k_aux": 256,
        "seed": 42,
        "activation_dim": 512,
        "dict_size": 4096,
        "group_fractions": [
            0.03125,
            0.0625,
            0.125,
            0.25,
            0.53125
        ],
        "group_weights": [
            0.2,
            0.2,
            0.2,
            0.2,
            0.2
        ],
        "group_sizes": [
            128,
            256,
            512,
            1024,
            2176
        ],
        "k": 256,
        "device": "cuda:0",
        "layer": 5,
        "lm_name": "Baidicoot/Othello-GPT-Transformer-Lens",
        "wandb_name": "MatryoshkaBatchTopKTrainer-othello-12_trainer_12",
        "submodule_name": null
    },
    "buffer": {
        "d_submodule": 512,
        "io": "out",
        "n_ctxs": 1000.0,
        "ctx_len": 59,
        "refresh_batch_size": 64,
        "out_batch_size": 8192,
        "device": "cuda:0"
    }
}