diff --git a/unsloth/kernels/utils.py b/unsloth/kernels/utils.py index 4439a47f23..7cd51e9ff0 100644 --- a/unsloth/kernels/utils.py +++ b/unsloth/kernels/utils.py @@ -139,6 +139,7 @@ pass if HAS_CUDA_STREAM: @torch.inference_mode def fast_dequantize(W, quant_state = None, out = None, use_global_buffer = False): + use_global_buffer = False if quant_state is None: return W if type(quant_state) is not list: # New quant_state as a class