[hardware] add flops count support for A3 device

[codemayq]

What does this PR do?

Add concise overview of what this PR aims to achieve or accomplish. Reference related GitHub issues and PRs that help with the review.

add flops count support for A3 device

Checklist Before Starting

• Search for similar PRs. Paste at least one query link here: ...
• Format the PR title as [{modules}] {type}: {description} (This will be checked by the CI)
  • {modules} include fsdp, megatron, sglang, vllm, rollout, trainer, ci, training_utils, recipe, hardware, deployment, ray, worker, single_controller, misc, perf, model, algo, env, tool, ckpt, doc, data
  • If this PR involves multiple modules, separate them with , like [megatron, fsdp, doc]
  • {type} is in feat, fix, refactor, chore, test
  • If this PR breaks any API (CLI arguments, config, function signature, etc.), add [BREAKING] to the beginning of the title.
  • Example: [BREAKING][fsdp, megatron] feat: dynamic batching

Test

Not related.

API and Usage Example

Not related.

Design & Code Changes

Not related.

Checklist Before Submitting

Important

Please check all the following items before requesting a review, otherwise the reviewer might deprioritize this PR for review.

• Read the Contribute Guide.
• Apply pre-commit checks: pre-commit install && pre-commit run --all-files --show-diff-on-failure --color=always
• Add / Update the documentation.
• Add unit or end-to-end test(s) to the CI workflow to cover all the code. If not feasible, explain why: ...
• Once your PR is ready for CI, send a message in the ci-request channel in the verl Slack workspace. (If not accessible, please try the Feishu group (飞书群).)

[gemini-code-assist]

Code Review

This pull request adds FLOPS counting support for the 'A3 device', which is identified as 'Ascend910'. My review found a potential correctness issue. The change groups 'Ascend910' with '910B', assigning them the same FLOPS value. However, these devices have different performance characteristics. Using the same value can lead to inaccurate Model FLOPS Utilization (MFU) calculations. I've recommended handling them separately with their correct FLOPS values.

[gemini-code-assist]

The FLOPS value for Ascend 910 and 910B devices seems to be incorrect, and they should not be grouped together as they have different performance characteristics. According to public documentation, Ascend 910 provides 256 TFLOPS (FP16) and Ascend 910B provides 320 TFLOPS (FP16). Using a single, potentially incorrect, value of 354 TFLOPS for both could lead to inaccurate MFU calculations. Please consider using separate, accurate values for each device.

Suggested change

	elif "910B" in device_name or "Ascend910" in device_name:
	flops = 354e12
	elif "910B" in device_name:
	flops = 320e12
	elif "Ascend910" in device_name:
	flops = 256e12