请求CastNumpy2Scalar支持fp16、bf16等常见数据类型

[ZHUI]

需求描述 Feature Description

Paddle/paddle/fluid/pybind/eager_utils.cc

Lines 1318 to 1359 in bcf7513

	paddle::experimental::Scalar CastNumpy2Scalar(PyObject* obj,
	const std::string& op_type,
	ssize_t arg_pos) {
	PyTypeObject* type = obj->ob_type;
	auto type_name = std::string(type->tp_name);
	VLOG(4) << "type_name: " << type_name;
	if (type_name == "numpy.ndarray" && PySequence_Check(obj)) {
	PyObject* item = nullptr;
	item = PySequence_GetItem(obj, 0);
	if (PyObject_CheckFloatOrToFloat(&item)) {
	float value = static_cast<float>(PyFloat_AsDouble(item));
	return paddle::experimental::Scalar(value);
	} else {
	PADDLE_THROW(platform::errors::InvalidArgument(
	"%s(): argument (position %d) is numpy.ndarry, the inner elements "
	"must be "
	"numpy.float32/float64 now, but got %s",
	op_type,
	arg_pos + 1,
	type_name)); // NOLINT
	}
	} else if (type_name == "numpy.float64") {
	double value = CastPyArg2Double(obj, op_type, arg_pos);
	return paddle::experimental::Scalar(value);
	} else if (type_name == "numpy.float32") {
	float value = CastPyArg2Float(obj, op_type, arg_pos);
	return paddle::experimental::Scalar(value);
	} else if (type_name == "numpy.int64") {
	int64_t value = CastPyArg2Long(obj, op_type, arg_pos);
	return paddle::experimental::Scalar(value);
	} else if (type_name == "numpy.int32" || type_name == "numpy.intc") {
	int value = CastPyArg2Int(obj, op_type, arg_pos);
	return paddle::experimental::Scalar(value);
	} else {
	PADDLE_THROW(platform::errors::InvalidArgument(
	"%s(): argument (position %d) must be "
	"numpy.float32/float64, numpy.int32/int64, but got %s",
	op_type,
	arg_pos + 1,
	type_name)); // NOLINT
	}
	}

测试样例:

>>> paddle.to_tensor([1], dtype="float16") + numpy.array([1], dtype="float16")[0]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: (InvalidArgument) __add__(): argument (position 1) must be numpy.float32/float64, numpy.int32/int64, but got numpy.float16 (at /paddle/paddle/fluid/pybind/eager_utils.cc:1280)

替代实现 Alternatives

No response

[paddle-bot]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[cuicheng01]

请问您现在的paddle版本是多少呢，最新的paddle已支持。

[ZHUI]

我就是拿的develop，你试试？
报错的代码大概位置也给了。

[cuicheng01]

我这里使用2.3.2暂时不能复现你的问题，你切换到2.3.2试一下呢？

[ZHUI]

最新dev，commit id 9218e742f8d5f74c5392c8c129fe024e2e1b6719

>>> import paddle
>>> import numpy
>>> paddle.to_tensor([1], dtype="float16") + numpy.array([1], dtype="float16")[0]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: (InvalidArgument) __add__(): argument (position 1) must be numpy.float32/float64, numpy.int32/int64, but got numpy.float16 (at /paddle/paddle/fluid/pybind/eager_utils.cc:1357)

>>> paddle.version.commit
'9218e742f8d5f74c5392c8c129fe024e2e1b6719'