change outputs and grads from fp16-fp16-comparision to fp16-fp32 #50700
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
|
❌ The PR is not created using PR's template. You can refer to this Demo. |
ZzSean
reviewed
Feb 22, 2023
| @@ -559,6 +559,56 @@ def feed_var(self, input_vars, place): | |||
|
|
|||
| return feed_map | |||
|
|
|||
| # only called by the fp16 | |||
| def feed_ref_var(self, input_vars, place): | |||
ZzSean
reviewed
Feb 22, 2023
| @@ -816,6 +1073,208 @@ def _check_api_outs_by_dygraph_outs(self, api_outs, dygraph_outs, place): | |||
| + self.__class__.__name__, | |||
| ) | |||
|
|
|||
| def _calc_ref_python_api_output(self, place, egr_inps=None, egr_oups=None): | |||
There was a problem hiding this comment.
唯一的区别是函数中调用的append_input_output_for_dygraph 变为append_ref_input_output_for_dygraph，可以写成一个，减少重复代码
ZzSean
force-pushed
the
fp16_compare_with_paddle_fp32
branch
from
c15ff33 to
7916071
Compare
…smile/Paddle into fp16_compare_with_paddle_fp32
ZzSean
reviewed
Feb 23, 2023
| @@ -1509,10 +1664,18 @@ def _compare_list(self, name, actual, expect): | |||
|
|
|||
| def compare_single_output_with_expect(self, name, expect): | |||
| actual, actual_np = self.find_actual_value(name) | |||
| expect_np = expect[0] if isinstance(expect, tuple) else expect | |||
| # expect, expect_np = self.find_expect_value(name) | |||
| + " max_tol: " | ||
| + str(max_tol) | ||
| ), | ||
| ) |
| def _compare_numpy(self, name, actual_np, expect_np): | ||
| if self.op_test.dtype == np.float16: |
…to fp16_compare_with_paddle_fp32
…smile/Paddle into fp16_compare_with_paddle_fp32
Xreki
approved these changes
Feb 24, 2023
| @@ -579,10 +642,10 @@ def _append_ops(self, block): | |||
| else: | |||
| self.infer_dtype_from_inputs_outputs(self.inputs, self.outputs) | |||
| inputs = append_input_output( | |||
| block, op_proto, self.inputs, True, self.dtype | |||
| block, op_proto, self.inputs, True, self.dtype, self.is_calc_ref | |||
There was a problem hiding this comment.
is_calc_ref这个变量名从命名看，表意不太明确，建议下个PR改下名。
| @@ -713,7 +782,12 @@ def create_var(np_value, name, is_input, if_return_inputs_grad_dict): | |||
| lod_temp = np_value[1] | |||
|
|
|||
| if is_input: | |||
| v = self._create_var_from_numpy(np_value_temp) | |||
| if is_calc_ref and np_value_temp.dtype == np.float16: | |||
| stop_gradient=False, | ||
| ) | ||
| if is_calc_ref and np_value_temp.dtype == np.float16: | ||
| v = block.create_var( |
There was a problem hiding this comment.
不太理解，这里构造的Program是哪样的呢？打印出来贴在PR描述里面吧。另外加些注释说明下。
zhangting2020
approved these changes
Feb 24, 2023
| NO_FP16_COMPARED_WITH_FP32_OP_LIST = [ | ||
| 'fake_quantize_moving_average_abs_max', | ||
| 'p_norm', | ||
| ] |
There was a problem hiding this comment.
这个列表建议加注释说明下，目的是什么，什么情况下可以添加op。可以下个PR修改下
Vvsmile
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
New features
PR changes
Others
Describe
change outputs and grads from fp16-fp16-comparision and fp16-fp32-comparision