CLEF-Screen / head /joint_head_config.json
macmacmacmac's picture
CLEF-Screen: JointSchemaHead redo FT (seed1, 3 epochs) from jev-retrain MI355X run 2026-11-04
2b605e8 verified
Raw History Blame Contribute Delete
119 Bytes
{
"hidden_size": 4096,
"width": 1024,
"routing_layers": 2,
"layers": 4,
"heads": 16,
"feedforward": 4096
}