Open zt706 opened 1 year ago
这还是个挺核心的错误,用ProgramEntrance_1 不会有这个错误吗?好像都会调用到这个地方的
这还是个挺核心的错误,用ProgramEntrance_1 不会有这个错误吗?好像都会调用到这个地方的
ProgramEntrance_1 不会报错
让我们偷偷发一下模型?
你好,都是官方代码,模型默认是 resnet18
唯一改动是:https://github.com/openppl-public/ppq/blob/master/ProgramEntrance_2.py#L234 这里用 calibrationdataloader = [torch.zeros(size=[1,3,224,224]) for in range(32)]
这个问题有解决了吗?我也遇到这个问题了
您好,这个bug解决了,我在QAT中遇到了这个bug
环境使用 stephen222/ppq:ubuntu18.04_cuda11.4_cudnn8.4_trt8.4.1.5 容器, 代码使用最新master 分支
[Warning] Compling Kernels... Please wait (It will take a few minutes). Traceback (most recent call last): File "ProgramEntrance_2.py", line 293, in
dispatching = PFL.Dispatcher(graph=graph).dispatch(
File "/workspace/ppq/ppq/scheduler/dispatchers.py", line 202, in dispatch
quant_operations = search_engine.opset_matching(
File "/workspace/ppq/ppq/IR/search.py", line 610, in opset_matching
partial_matchings = self._opset_matching(
File "/workspace/ppq/ppq/IR/search.py", line 484, in _opset_matching
if ep_expr is not None and ep_expr(op):
File "/workspace/ppq/ppq/scheduler/dispatchers.py", line 205, in
ep_expr = lambda x: (x.type not in quant_types) or x.is_boundary,
TypeError: argument of type 'method' is not iterable