ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update llama.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-02-03 18:20:06 +11:00
commit dd03abedd6
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
unsloth/models/llama.py
View file

@ -124,7 +124,7 @@ def LlamaAttention_fast_forward_inference(
# Prefill phase
# if not hasattr(self, "paged_attention"):
# if do_prefill:
if do_prefill:
# self.paged_attention = torch.empty((KV_CACHE_INCREMENT+1, 2, bsz, n_kv_heads, head_dim), dtype = dtype, device = "cuda")
# self.paged_attention_K = self.paged_attention[:,0]
# self.paged_attention_V = self.paged_attention[:,1]
@ -134,7 +134,7 @@ def LlamaAttention_fast_forward_inference(
# self.temp_KV = torch.empty((2, bsz, 1, n_kv_heads*head_dim), dtype = dtype, device = "cuda")
# self.RH_Q = torch.empty((bsz, n_heads, 1, head_dim), dtype = dtype, device = "cuda")
# self.attention = torch.empty((bsz, n_heads, 1, KV_CACHE_INCREMENT), dtype = dtype, device = "cuda")
# self.scalar = 1.0 / math_sqrt(self.head_dim)
self.scalar = 1.0 / math_sqrt(self.head_dim)
# elif kv_seq_len >= self.paged_attention.shape[0]:
# self.paged_attention.resize_((self.paged_attention.shape[0]+KV_CACHE_INCREMENT, 2, bsz, n_kv_heads, head_dim))
# self.paged_attention_K = self.paged_attention[:,0]