From 63d66b0572d11cedd5574b38da720299599539b3 Mon Sep 17 00:00:00 2001 From: duzx16 Date: Sat, 29 Apr 2023 10:34:45 +0800 Subject: [PATCH] Remove assert in load_cpu_kernel --- quantization.py | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/quantization.py b/quantization.py index 49b53de..d739c90 100644 --- a/quantization.py +++ b/quantization.py @@ -442,7 +442,6 @@ class QuantizedEmbedding(Embedding): # TODO: backward, check empty_init def load_cpu_kernel(**kwargs): global cpu_kernels cpu_kernels = CPUKernel(**kwargs) - assert cpu_kernels.load def quantize(model, weight_bit_width, use_quantization_cache=False, empty_init=False, **kwargs): @@ -453,9 +452,8 @@ def quantize(model, weight_bit_width, use_quantization_cache=False, empty_init=F dense_h_to_4h_quantization_cache = None dense_4h_to_h_quantization_cache = None - try: - load_cpu_kernel(**kwargs) - except: + load_cpu_kernel(**kwargs) + if not cpu_kernels.load: if kernels is None: # CUDA kernels failed print("Cannot load cpu or cuda kernel, quantization failed:") assert kernels is not None