smart-sun-31 / config.json
quik-models's picture
Upload autoresearch experiment (val_bpb=0.778052)
ed064c1 verified
Raw
History Blame Contribute Delete
266 Bytes
{
"attention_backend": "sdpa",
"compute_dtype": "torch.bfloat16",
"n_embd": 1408,
"n_head": 11,
"n_kv_head": 11,
"n_layer": 22,
"sequence_len": 2048,
"use_activation_checkpointing": true,
"vocab_size": 16384,
"window_pattern": "SSSL"
}