| { |
| "n_embd": 1792, |
| "n_inner": 3584, |
| "n_head": 16, |
| "n_layer": 36, |
|
|
| "mlp_fc1_bias": false, |
| "mlp_fc2_bias": false, |
| "out_proj_bias": false, |
| "qkv_proj_bias": false, |
| "reorder_and_upcast_attn": false, |
| "scale_attn_by_inverse_layer_idx": false, |
|
|
| "activation_function": "swiglu", |
| "resid_pdrop": 0.0, |
| "rms_norm": true, |
| "residual_in_fp32": true, |
| "pad_vocab_size_multiple": 8, |
| "use_flash_attn": true, |
| "special_initializer": true, |
| "rotary_emb_fraction": 1, |
| "max_position_embeddings": 0, |
|
|
| "alt_mixer_layers": [1, 6, 11, 16, 21, 27, 33], |
| "alt_mixer_2_layers": [2, 7, 12, 17, 22, 28, 34], |
| "mixer": { |
| "_target_": "based.models.mixers.base_conv.BaseConvWithSiLU4", |
| "expand_proj": 4, |
| "l_max": 2048, |
| "kernel_sizes": [3], |
| "use_bias": true |
| }, |
| "alt_mixer": { |
| "_target_": "based.models.mixers.linear_attn.LinearAttention", |
| "feature_dim": 16, |
| "feature_name": "taylor_exp", |
| "l_max": 2048, |
| "num_heads": 16, |
| "num_key_value_heads": 16, |
| "train_view": "linear" |
| }, |
| "alt_mixer_2": { |
| "_target_": "based.models.mixers.slide_fa2.SlidingsMHA", |
| "causal": true, |
| "num_heads": 16, |
| "window_size": 128 |
| } |
| } |
