wandb-config.yaml

_wandb:
    value:
        cli_version: 0.19.11
        m:
            - "1": gpu/memory_allocated_gb
              "6":
                - 3
              "7": []
            - "1": gpu/max_memory_allocated_gb
              "6":
                - 3
              "7": []
            - "1": gpu/memory_reserved_gb
              "6":
                - 3
              "7": []
        python_version: 3.11.10
        t:
            "1":
                - 1
                - 11
                - 49
                - 51
                - 55
                - 71
            "2":
                - 1
                - 11
                - 49
                - 51
                - 55
                - 71
            "3":
                - 2
                - 7
                - 13
                - 16
                - 23
                - 55
                - 61
            "4": 3.11.10
            "5": 0.19.11
            "6": 4.52.4
            "8":
                - 5
            "12": 0.19.11
            "13": linux-x86_64
act_fn:
    value: relu
batch_size:
    value: 8192
before_ln:
    value: false
c_coeff:
    value: 4
cooldown_start_frac:
    value: 0.8
d_feature:
    value: 163840
d_model:
    value: 5120
device:
    value: cuda:0
initial_lr:
    value: 0.0002
layer_idx:
    value: 0
lr:
    value: 0.0002
min_lr_ratio:
    value: 0
model_name:
    value: Qwen/Qwen3-14B
model_type:
    value: qwen
n_batches:
    value: 277
n_grad_steps:
    value: 4
n_steps:
    value: 122070
preact_coeff:
    value: 6e-05
shuffle_buffer_batches:
    value: 32
skip_connections:
    value: false
sparsity_coeff_final:
    value: 8
x_scale:
    value: 1
y_scale:
    value: 1