PaddleCheckError: Expected t->IsInitialized() == true, but received t->IsInitialized():0 != true
The Tensor in the coalesce_tensor Op's Input Variable Input(conv2d_65.w_0.quantized.dequantized) is not initialized. at [/home/omni/Paddle/paddle/fluid/framework/operator.cc:1155]

paddle是两周前拉取的develop最新的分支编译的，量化模型用的是paddle slim底层api

请问是使用paddleslim中哪个量化api啊？能给出更多你调用的核心代码吗？

PaddleCheckError: Expected t->IsInitialized() == true, but received t->IsInitialized():0 != true
The Tensor in the coalesce_tensor Op's Input Variable Input(conv2d_65.w_0.quantized.dequantized) is not initialized. at [/home/omni/Paddle/paddle/fluid/framework/operator.cc:1155]
paddle是两周前拉取的develop最新的分支编译的，量化模型用的是paddle slim底层api

请问是使用paddleslim中哪个量化api啊？能给出更多你调用的核心代码吗？

`

transform_pass = QuantizationTransformPass(
    scope=fluid.global_scope(),
    place=place,
    activation_quantize_type="range_abs_max",
    weight_quantize_type="abs_max")

transform_pass.apply(main_graph)
transform_pass.apply(test_graph)
build_strategy = fluid.BuildStrategy()
build_strategy.memory_optimize = False
build_strategy.enable_inplace = False
binary = fluid.CompiledProgram(main_graph.graph).with_data_parallel(
    loss_name=train_cost.name, build_strategy=build_strategy)
test_prog = test_graph.to_program()`

PaddleCheckError: Expected t->IsInitialized() == true, but received t->IsInitialized():0 != true
The Tensor in the coalesce_tensor Op's Input Variable Input(conv2d_65.w_0.quantized.dequantized) is not initialized. at [/home/omni/Paddle/paddle/fluid/framework/operator.cc:1155]
paddle是两周前拉取的develop最新的分支编译的，量化模型用的是paddle slim底层api

请问是使用paddleslim中哪个量化api啊？能给出更多你调用的核心代码吗？

`

transform_pass = QuantizationTransformPass(
    scope=fluid.global_scope(),
    place=place,
    activation_quantize_type="range_abs_max",
    weight_quantize_type="abs_max")

transform_pass.apply(main_graph)
transform_pass.apply(test_graph)
build_strategy = fluid.BuildStrategy()
build_strategy.memory_optimize = False
build_strategy.enable_inplace = False
binary = fluid.CompiledProgram(main_graph.graph).with_data_parallel(
    loss_name=train_cost.name, build_strategy=build_strategy)
test_prog = test_graph.to_program()`

是不是使用多卡训练会出现这个问题？可以尝试添加build_strategy.fuse_all_reduce_ops = False试下。

PaddleCheckError: Expected t->IsInitialized() == true, but received t->IsInitialized():0 != true
The Tensor in the coalesce_tensor Op's Input Variable Input(conv2d_65.w_0.quantized.dequantized) is not initialized. at [/home/omni/Paddle/paddle/fluid/framework/operator.cc:1155]
paddle是两周前拉取的develop最新的分支编译的，量化模型用的是paddle slim底层api

请问是使用paddleslim中哪个量化api啊？能给出更多你调用的核心代码吗？

`

transform_pass = QuantizationTransformPass(
    scope=fluid.global_scope(),
    place=place,
    activation_quantize_type="range_abs_max",
    weight_quantize_type="abs_max")

transform_pass.apply(main_graph)
transform_pass.apply(test_graph)
build_strategy = fluid.BuildStrategy()
build_strategy.memory_optimize = False
build_strategy.enable_inplace = False
binary = fluid.CompiledProgram(main_graph.graph).with_data_parallel(
    loss_name=train_cost.name, build_strategy=build_strategy)
test_prog = test_graph.to_program()`

是不是使用多卡训练会出现这个问题？可以尝试添加build_strategy.fuse_all_reduce_ops = False试下。

谢谢，是多卡训练，添加这个属性后已经可以正常训练了