From 23e6ebf14dc7b2d9bcba737b29abf6f1922682fc Mon Sep 17 00:00:00 2001 From: Daniel Han-Chen Date: Sat, 24 Feb 2024 17:13:40 +1100 Subject: [PATCH] Update gemma.py --- unsloth/models/gemma.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/unsloth/models/gemma.py b/unsloth/models/gemma.py index f8e8f11d0b..a8825352b5 100644 --- a/unsloth/models/gemma.py +++ b/unsloth/models/gemma.py @@ -67,7 +67,8 @@ def GemmaDecoderLayer_fast_forward( # padding_mask: Optional[torch.LongTensor] = None, *args, **kwargs, ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]: - if past_key_value is not None: + print(past_key_value) + if False:#past_key_value is not None: do_prefill = not hasattr(self.self_attn, "paged_attention") # Self Attention