Kismit
Kismit
加了=还是报一样的错 $ paddle_lite_opt --model_file=qat_model.pdmodel --param_file=qat_model.pdiparams --optimize_out_type=naive_buffer --optimize_out=trans_qat --valid_targets=arm Error: This model is not supported, because 1 ops are not supported on 'arm'. These unsupported ops are: 'moving_average_abs_max_scale'. Aborted (core dumped)
qat量化的模型是transformer,怎么查看具体是哪些op不支持
paddle_lite_opt --model_file qat_model.pdmodel --param_file qat_model.pdiparams --optimize_out_type=naive_buffer --optimize_out=trans_qat --valid_targets=arm --quant_model=true --quant_type=QUANT_INT8 --print_model_ops=true OPs in the input model include: OP_name Host X86 CUDA ARM OpenCL FPGA NPU XPU RKNPU APUHuaweiAscendNPUImaginationNNA IntelFPGA Any...
怎样看到具体是哪个Layer或者function不支持,现在是不知道哪个用到了moving_average_abs_max_scale导致, 还有这个 not_equal 是啥意思
我应该没有主动使用moving_average_abs_max_scale这种方式,下面是config配置: quant_config = { # weight preprocess type, default is None and no preprocessing is performed. 'weight_preprocess_type': 'PACT', # activation preprocess type, default is None and no preprocessing is performed....
更正一下: 'activation_quantize_type': 'range_abs_max', --》 'activation_quantize_type': ‘moving_average_abs_max’
import argparse import os from pprint import pprint import paddle import paddle.distributed as dist import yaml from attrdict import AttrDict from examples import reader from pptrtrans.transformers import TransformerModel, CrossEntropyCriterion from...
pptrtrans是哪个github repo?还是非公开的代码? 这里pptrtrans是创建的项目的名字,然后Transformer用的是paddlenlp里面的transformer,只是不想加太多包,直接拷贝到本项目中了,paddlepaddle和paddleslim用的都是v2.2最新版本
``` import sys sys.path.append('../..') import argparse import os from pprint import pprint import paddle import paddle.distributed as dist import yaml from attrdict import AttrDict from examples import reader from paddlenlp.transformers...
transformer.small_quant.yaml内容: ``` # The frequency to save trained models when training. save_step: 10000 # The frequency to fetch and print output when training. print_step: 100 # Path of the checkpoint,...