【PIR API adaptor No.248、249】 Migrate CTCLoss/RNNTLossinto pir

[DrRyanHuang]

PR types

Others

PR changes

Others

Description

PIR API 推全升级

• 新IR Python API适配升级 #58067

CTCLoss 单测通过率 5/7
test_warpctc_op TestWarpCTCOpError.test_errors 未开启
TestWarpCTCOp 未开启由于：

self.inputs = {
    "Logits": (logits, self.logits_lod),
    "Label": (labels, self.labels_lod),
}

有些继承自OpTest的单测，inputs配置类似上面的代码，比如Logits输入是一个tuple，第二个参数是lod参数。这类的单测在pir下还不支持，请先排查下这类单测，先不要打开pir测试，然后我们再看下是否还会有其他问题~

RNNTLoss 单测通过率 4/5: TestWarpRNNTOpError.test_errors 未开启

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[paddle-ci-bot]

Sorry to inform you that cc75327's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually.

[0x45f]

self.inputs = {
    "Logits": (logits, self.logits_lod),
    "Label": (labels, self.labels_lod),
}

有些继承自OpTest的单测，inputs配置类似上面的代码，比如Logits输入是一个tuple，第二个参数是lod参数。这类的单测在pir下还不支持，请先排查下这类单测，先不要打开pir测试，然后我们再看下是否还会有其他问题~

[0x45f]

看CI报错是test_check_grad会有精度问题，先把test_check_grad的check_pir设置为False吧，我找相关同学看下~

[DrRyanHuang]

FAIL: test_eager_ctcloss (test_warpctc_op.TestCTCLossAPICase)

这个应该和 pir 关系不大，也出现了数值问题，可以帮忙看一下嘛🙏 @0x45f

2023-11-15 15:51:21 ======================================================================
2023-11-15 15:51:21 FAIL: test_eager_ctcloss (test_warpctc_op.TestCTCLossAPICase)
2023-11-15 15:51:21 ----------------------------------------------------------------------
2023-11-15 15:51:21 Traceback (most recent call last):
2023-11-15 15:51:21   File "C:\home\workspace\Paddle\build\test\legacy_test\test_warpctc_op.py", line 741, in test_eager_ctcloss
2023-11-15 15:51:21     test_functional_api()
2023-11-15 15:51:21   File "C:\home\workspace\Paddle\build\test\legacy_test\test_warpctc_op.py", line 737, in test_functional_api
2023-11-15 15:51:21     np.testing.assert_allclose(
2023-11-15 15:51:21   File "C:\home\workspace\cache\python_venv\lib\site-packages\numpy\testing\_private\utils.py", line 1504, in assert_allclose
2023-11-15 15:51:21     assert_array_compare(compare, actual, desired, err_msg=str(err_msg),
2023-11-15 15:51:21   File "C:\Python39\lib\contextlib.py", line 79, in inner
2023-11-15 15:51:21     return func(*args, **kwds)
2023-11-15 15:51:21   File "C:\home\workspace\cache\python_venv\lib\site-packages\numpy\testing\_private\utils.py", line 797, in assert_array_compare
2023-11-15 15:51:21     raise AssertionError(msg)
2023-11-15 15:51:21 AssertionError: 
2023-11-15 15:51:21 Not equal to tolerance rtol=1e-05, atol=1
2023-11-15 15:51:21 Mismatched elements: 1 / 1 (100%)
2023-11-15 15:51:21 Max absolute difference: 1.0417404
2023-11-15 15:51:21 Max relative difference: 0.06375698
2023-11-15 15:51:21  x: array(17.380978, dtype=float32)
2023-11-15 15:51:21  y: array(16.339237, dtype=float32)
2023-11-15 15:51:21 ----------------------------------------------------------------------

[xiaoguoguo626807]

#59189 反向问题修复

[MarioLulab]

LGTM