Views
No views yet
(ln1): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
(attn): MLA(
(compress_kv): Linear(in_features=256, out_features=64, bias=False)
(kv_proj): Linear(in_features=64, out_features=512, bias=False)
(compress_q): Linear(in_features=256, out_features=64, bias=False)
(q_proj): Linear(in_features=64, out_features=256, bias=False)
(q_rope): Linear(in_features=64, out_features=256, bias=False)
(k_rope): Linear(in_features=64, out_features=32, bias=False)
(out_proj): Linear(in_features=256, out_features=256, bias=False)
(attn_dropout): Dropout(p=0.1, inplace=False)
)
(ln2): LayerNorm((256,), eps=1e-05, elementwise_affine=True)
(ffn): MOELayer(
(experts): ModuleList(
(0-7): 8 x FFN(
(gate_proj): Linear(in_features=256, out_features=512, bias=False)
(up_proj): Linear(in_features=256, out_features=512, bias=False)
(down_proj): Linear(in_features=512, out_features=256, bias=False)
(ffn_dropout): Dropout(p=0.1, inplace=False)
)
)
)
)