[kernel] Use sgl_kernel rope

[ByronHsu]

Motivation

Depends on https://github.com/sgl-project/sglang/actions/runs/12984127304

Modifications

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling.

[zhyncs]

Please update this version limit.

sglang/python/pyproject.toml

Line 30 in 741fccd

"sgl-kernel>=0.0.2.post18", "torch", "vllm==0.6.4.post1",

[ByronHsu]

needs #3173

[zhyncs]

ref
https://github.com/sgl-project/sglang/blob/main/test/srt/test_session_control.py
https://github.com/sgl-project/sglang/actions/runs/12988314124

[zhyncs]

Everything is going well except for an accuracy issue with test_session_control.

[ByronHsu]

I suspect the test is flaky by nature. I switched from 1B to 8B model on the main and test_session_control failed with different output: #3184.

Here is my finding so far

Kernel	1B Model	8B Model
vLLM	Pass	Fail
flashinfer	Fail	Fail

Given other tests for accuracy all pass. I think the correctness looks ok

[zhyncs]

Byron is cooking!! Great work.

[ByronHsu]

"Byron" XD