forme / original /params.json
sarkoB6273's picture
Duplicate from NousResearch/Meta-Llama-3.1-405B-FP8
92f0462
Raw
History Blame Contribute Delete
202 Bytes
{"dim": 16384, "ffn_dim_multiplier": 1.2, "multiple_of": 4096, "n_heads": 128, "n_kv_heads": 8, "n_layers": 126, "norm_eps": 1e-05, "rope_theta": 500000.0, "use_scaled_rope": true, "vocab_size": 128256}