[Feat] Refactor for parallel_config in FusedMoEModularKernel

[yewentao256]

Purpose

Intend to fix #29999 (comment)

and #29999 (comment)

But after discussion with @ProExpertProg and @bnellnm , we found that require a must-pass param needs to update everything especially for tests, we decide to instead using a moe_parallel_config

This PR could also fix current issue in main

(Worker_TP2_EP2 pid=248418) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP2_EP2 pid=248418) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2048.
(Worker_TP4_EP4 pid=248420) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP4_EP4 pid=248420) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2048.
(Worker_TP3_EP3 pid=248419) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP3_EP3 pid=248419) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2048.
(Worker_TP5_EP5 pid=248421) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP5_EP5 pid=248421) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2048.
(Worker_TP0_EP0 pid=248416) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP0_EP0 pid=248416) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2048.
(Worker_TP6_EP6 pid=248422) WARNING 12-10 13:35:58 [vllm.py:1394] Current vLLM config is not set.
(Worker_TP6_EP6 pid=248422) INFO 12-10 13:35:58 [scheduler.py:228] Chunked prefill is enabled with max_num_batched_tokens=2

[gemini-code-assist]

Code Review

This pull request adds a helpful comment to vllm/model_executor/layers/fused_moe/modular_kernel.py clarifying the logic for handling the parallel_config parameter in FusedMoEModularKernel. The comment explains that while explicit passing is preferred, a fallback to the current vLLM config exists for testing purposes. This improves code clarity and maintainability. The change is correct and I have no further suggestions.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.

[yewentao256]

@bnellnm @ProExpertProg CC