trainer_0/config.json

{
    "trainer": {
        "trainer_class": "TopKTrainer",
        "dict_class": "AutoEncoderTopK",
        "lr": 8.944271909999159e-05,
        "steps": 12207,
        "auxk_alpha": 0.03125,
        "warmup_steps": 0,
        "decay_start": null,
        "threshold_beta": 0.999,
        "threshold_start_step": 1000,
        "seed": null,
        "activation_dim": 10240,
        "dict_size": 81920,
        "k": 128,
        "device": "cuda:2",
        "layer": 40,
        "lm_name": "blah",
        "wandb_name": "AutoEncoderTopK",
        "submodule_name": null
    },
    "buffer": {
        "n_models": 2,
        "d_submodule": 5120,
        "io": "out",
        "n_ctxs": 256,
        "ctx_len": 512,
        "refresh_batch_size": 64,
        "out_batch_size": 4096,
        "device": "cuda:2",
        "rescale_acts": false
    }
}