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 2024-01-21 18:52:15 +11:00
commit ac99a47a45
1 changed files with 3 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

7
unsloth/models/llama.py
View file

@ -189,7 +189,7 @@ def LlamaAttention_fast_forward(
bsz, q_len, _ = hidden_states.size()
# Check for inference
if use_cache and past_key_value is not None and q_len == 1:
if past_key_value is not None and q_len == 1:
A, past_key_value = LlamaAttention_fast_forward_inference(
self,
hidden_states,
@ -305,8 +305,7 @@ def LlamaDecoderLayer_fast_forward(
"""
bsz, q_len, hd = hidden_states.size()
if (not self.training and q_len == 1):
print(1)
if (past_key_value is not None and q_len == 1):
# Self Attention
residual = hidden_states
hidden_states = fast_rms_layernorm_inference(self.input_layernorm, hidden_states)
@ -525,7 +524,7 @@ def LlamaModel_fast_forward(
pass
bsz, q_len, hd = hidden_states.size()
if (not self.training and q_len == 1):
if (past_key_value is not None and q_len == 1):
hidden_states = fast_rms_layernorm_inference(self.norm, hidden_states)
else:
hidden_states = fast_rms_layernorm(self.norm, hidden_states)