Add Ascend NPU support for verl

[Yikun]

trafficstars

This RFC proposed to provide initial support for verl on Ascend NPU.

Pre work

• Enable Ascend CI: https://github.com/volcengine/verl/pull/465

PR1：Support FSDP worker and vLLM Ascend https://github.com/volcengine/verl/pull/332

A basic PR based on FSDP and Ascend NPU to make GRPO workflow work on Ascend NPU (based on https://github.com/volcengine/verl/pull/198 which is proposed by @Chendong98 )

Include Single controller, FSDP worker Ascend NPU support, vLLM Ascend support (>=0.7.1)

Co-author:

• @as12138 (FSDP worker / vLLM Ascend)
• @Chendong98(Single controller)
• @celestialli (vLLM Ascend)

PR2：Support Megatron 0.6 in veRL https://github.com/volcengine/verl/pull/85

Owner: @Chendong98

PR3: Support MindSpeed worker in veRL TBD

Based on PR2 Owner: @Chendong98

[Yikun]

We had a internal meeting about Ascend NPU support for veRL, above is the initial info to make things clear for several PRs recent submitted (https://github.com/volcengine/verl/pull/332 https://github.com/volcengine/verl/pull/198 https://github.com/volcengine/verl/pull/85), hopes no more misunderstanding.

[takagi97]

@Yikun Thanks for your efforts! Can PR https://github.com/volcengine/verl/pull/332 be directly implemented on 910b2c 64GB?

[Yikun]

Yes, the initial support will based on community CI which are running on Altlas A2 series, for more info pls see CI info: https://github.com/volcengine/verl/actions/runs/13649493801/job/38154761144

[takagi97]

Yes, the initial support will based on community CI which are running on Altlas A2 series, for more info pls see CI info: https://github.com/volcengine/verl/actions/runs/13649493801/job/38154761144

Thank you for your quick response! I will try it.

[Syno8]

Thank you very much, I will try it later.