tejas-generalforecast's picture
Convert medium/classifier.pt from nvidia/Kumo-Tabular
f7c3ce9 verified
Raw History Blame Contribute Delete
375 Bytes
{
"architecture_name": "kumo_tabular",
"config": {
"cell_channels": 256,
"num_embedding_layers": 6,
"num_embedding_heads": 4,
"num_inducing_points": 256,
"group_size": 3,
"num_frequencies": 32,
"num_readout_tokens": 4,
"icl_channels": 512,
"num_icl_layers": 24,
"num_icl_heads": 8,
"num_icl_key_value_heads_for_query": 2
}
}