From e500b785f76306800dbef5d367a84b064a4c0d2e Mon Sep 17 00:00:00 2001 From: Daniel Han-Chen Date: Sat, 3 Feb 2024 23:12:16 +1100 Subject: [PATCH] Update llama.py --- unsloth/models/llama.py | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py index bc401dbbe4..07301bcb7d 100644 --- a/unsloth/models/llama.py +++ b/unsloth/models/llama.py @@ -680,9 +680,10 @@ def LlamaModel_fast_forward_inference( decoder_layer.self_attn, hidden_states, past_key_values[idx], - temp_QA, - temp_KV, - RH_Q, + dp_prefill = False, + temp_QA = temp_QA, + temp_KV = temp_KV, + RH_Q = RH_Q, ) hidden_states += residual