maseek-mini / config.json
voperl's picture
Maseek mini: MLA + SSM + MoE, pretrained on FineWeb-Edu
ff00cce verified
Raw History Blame Contribute Delete
252 Bytes
{
"model_type": "maseek",
"architectures": [
"Maseek"
],
"vocab_size": 50257,
"n_embd": 512,
"n_layer": 6,
"block_size": 1024,
"dropout": 0.1,
"ssm_d_state": 16,
"moe_n_experts": 8,
"moe_k_top": 2,
"torch_dtype": "float32"
}