From 0f732c06b3aa4e9a86a06354a7cf30b47949db29 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Sun, 2 Feb 2025 16:23:31 -0800 Subject: [PATCH] Update utils.py --- unsloth/kernels/utils.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/unsloth/kernels/utils.py b/unsloth/kernels/utils.py index c5df015ca0..753eda5b3b 100644 --- a/unsloth/kernels/utils.py +++ b/unsloth/kernels/utils.py @@ -404,7 +404,7 @@ def fast_linear_forward(proj, X, temp_lora = None, out = None): elif bsz == 1 and q_len == 1: out = fast_gemv(X, W, W_quant, out = out) else: - W = fast_dequantize(W.t(), W_quant) + W = fast_dequantize(W.t(), W_quant, use_global_buffer = True) out = torch.matmul(X, W, out = out) pass @@ -438,7 +438,7 @@ pass def matmul_lora(X, W, W_quant, A, B, s, out = None): dtype = X.dtype - W = fast_dequantize(W.t(), W_quant) + W = fast_dequantize(W.t(), W_quant, use_global_buffer = True) if X.dim() == 3: batch, seq_len, d = X.shape