【PaddlePaddle Hackathon 4】No.56 : add fp16 test and bf16 for poisson #51662
Conversation
|
你的PR提交成功，感谢你对开源项目的贡献! |
| or not core.is_bfloat16_supported(core.CUDAPlace(0)), | ||
| "core is not complied with CUDA and not support the bfloat16", | ||
| ) | ||
| class TestSearchSortedBF16(OpTest): |
There was a problem hiding this comment.
TestSearchSortedBF16改为TestSearchSortedBF16Op
| or not core.is_bfloat16_supported(core.CUDAPlace(0)), | ||
| "core is not complied with CUDA and not support the bfloat16", | ||
| ) | ||
| class TestPoissonBF16(OpTest): |
There was a problem hiding this comment.
TestPoissonBF16改为TestPoissonBF16Op
|
Almost LGTM |
fix function name
fix function name
|
LGTM，Rerun一下Build相关的测试 |
|
@Vvsmile 已经rerun了，但还是报错，应该如何修复这个报错呢？
|
longranger2
changed the title
|
Sorry to inform you that 2f9ffc0's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
|
很抱歉，经过我们的反复讨论，你的PR暂未达到合入标准，请阅读飞桨原生算子开发规范，你可以重新提交新的PR，我们先将此PR关闭，感谢你的贡献。 |
|
@zhangting2020 辛苦帮忙review下～， |
test/legacy_test/test_poisson_op.py
Outdated
| x = np.full([2048, 1024], self.lam, dtype="float32") | ||
| out = np.ones([2048, 1024], dtype="float32") | ||
| x = convert_uint16_to_float(convert_float_to_uint16(x)) | ||
| out = convert_uint16_to_float(convert_float_to_uint16(out)) |
There was a problem hiding this comment.
397和398不用加，因为400~401行，你要把float转uint16
test/legacy_test/test_poisson_op.py
Outdated
| self.dtype = np.uint16 | ||
|
|
||
| def verify_output(self, outs): | ||
| hist, prob = output_hist(np.array(outs[0]), self.lam, self.a, self.b) |
There was a problem hiding this comment.
这一行错误原因是outs[0]是op的输出，由于输出是bf16类型，np.array后变成了uint16。所以为了得到正确的计算结果，这里要将转为uint16的numpy数据，转回float32
hist, prob = output_hist(convert_uint16_to_float(np.array(outs[0])), self.lam, self.a, self.b)
|
@zhangting2020 修改好了，辛苦review下～ |
|
LGTM，记得更新文档 |
|
@zhangting2020 文档已经更新，辛苦reveiw～ |
PR types
Others
PR changes
APIs
Description
相关链接：
英文文档：
中文文档：