Update llama.py

This commit is contained in:
Daniel Han 2025-03-21 16:44:10 -07:00
commit 8720678ba6

View file

@ -929,9 +929,9 @@ def LlamaModel_fast_forward_inference(
temp_mlp = torch.empty((2, bsz, 1, mlp_size), dtype = X.dtype, device = "cuda:0")
temp_gate, temp_up = temp_mlp[0], temp_mlp[1]
print(type(past_key_values), len(past_key_values))
seq_len = past_key_values[0][0].shape[-2]
print(type(past_key_values), len(past_key_values), seq_len)
if bsz != 1:
attention_mask = _prepare_4d_causal_attention_mask_for_sdpa(
attention_mask,