mindone adapted ms2.6 && ms2.7

[QYuFong]

What does this PR do?

Fixes # (issue)

Adds # (feature)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes? E.g. record bug fixes or new features in What's New. Here are the
  documentation guidelines
• Did you build and run the code without any errors?
• Did you report the running environment (NPU type/MS version) and performance in the doc? (better record it for data loading, model inference, or training tasks)
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@xxx

[hadipash]

Update here too?

mindone/mindone/utils/env.py

Line 33 in 5a76add

jit_level: Optional[Literal["O0", "O1", "O2"]] = None,

[SamitHuang]

/gemini review

[gemini-code-assist]

Code Review

This pull request adapts the code for MindSpore versions 2.6 and 2.7. The changes mainly involve updating the JIT optimization level and replacing deprecated functions like chunk_ext with mint.chunk.

I've found a critical issue in examples/opensora_hpcai/scripts/inference.py where the performance metric for sampling speed is calculated incorrectly. This leads to misleading logs. I've provided a suggestion to fix the formula and use the correct variable for batch size. The other changes look good and are consistent with the goal of adapting to a new MindSpore version.

[gemini-code-assist]

The formula for calculating sampling speed in s/step is incorrect. This will result in misleading performance metrics being logged.

There are two issues with the current calculation:

1. Incorrect formula: To calculate seconds per step (s/step), you should divide the total time by the total number of steps. The current formula batch_time / args.sampling_steps * ns is evaluated as (batch_time / args.sampling_steps) * ns due to operator precedence, which is not mathematically correct for s/step. It should be batch_time / (args.sampling_steps * num_samples).

2. Incorrect number of samples: The calculation uses ns, which is the total number of prompts for the entire run. For a per-batch performance metric, it should use the number of samples in the current batch. You can get this from samples.shape[0], as samples is available in this scope.

I've suggested a fix below that addresses both points.

Suggested change

	f"Batch time cost: {batch_time:.3f}s, sampling speed: {batch_time / args.sampling_steps * ns:.4f} s/step"
	f"Batch time cost: {batch_time:.3f}s, sampling speed: {batch_time / (args.sampling_steps * samples.shape[0]):.4f} s/step"

[SamitHuang]

@QYuFong Pls fix

[QYuFong]

Done

[QYuFong]

Update here too?

mindone/mindone/utils/env.py

Line 33 in 5a76add

jit_level: Optional[Literal["O0", "O1", "O2"]] = None,

Just sd3 need to change O2 -> O1, because SD3 has discontinuous input, the GE backend no longer supports it. Other models are not affected for the time being.

[hadipash]

Just sd3 need to change O2 -> O1, because SD3 has discontinuous input, the GE backend no longer supports it. Other models are not affected for the time being.

I mean update typing and docs: jit_level: Optional[Literal["O0", "O1", "O2"]] = None -> jit_level: Optional[Literal["O0", "O1"]] = None