add ucclep integration

[CalebZ9909]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

BEFORE SUBMITTING, PLEASE READ https://docs.vllm.ai/en/latest/contributing (anything written below this line will be removed by GitHub Actions)

[CalebZ9909]

I ran into a weird error: I triggered the CUDA Triton Error: [invalid argument] with the following messege:

[1;36m(EngineCore_DP6 pid=3065384)[0;0m   File "/fsx/ubuntu/uccl_yihan/vllm/vllm/model_executor/layers/fused_moe/fused_batched_moe.py", line 447, in invoke_moe_batched_triton_kernel
[1;36m(EngineCore_DP5 pid=3065380)[0;0m   File "/fsx/ubuntu/miniconda3/lib/python3.13/site-packages/torch/nn/modules/module.py", line 1784, in _call_impl
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     batched_triton_kernel[grid](
[1;36m(EngineCore_DP5 pid=3065380)[0;0m     return forward_call(*args, **kwargs)
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     ~~~~~~~~~~~~~~~~~~~~~~~~~~~^
[1;36m(EngineCore_DP5 pid=3065380)[0;0m   File "/fsx/ubuntu/uccl_yihan/vllm/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 1160, in forward
[1;36m(EngineCore_DP6 pid=3065384)[0;0m         A,
[1;36m(EngineCore_DP5 pid=3065380)[0;0m     fused_out = self._fused_experts(
[1;36m(EngineCore_DP6 pid=3065384)[0;0m         ^^
[1;36m(EngineCore_DP5 pid=3065380)[0;0m         in_dtype=hidden_states.dtype,
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     ...<38 lines>...
[1;36m(EngineCore_DP5 pid=3065380)[0;0m     ...<11 lines>...
[1;36m(EngineCore_DP6 pid=3065384)[0;0m         BLOCK_K=BLOCK_K,
[1;36m(EngineCore_DP5 pid=3065380)[0;0m         expert_tokens_meta=expert_tokens_meta,
[1;36m(EngineCore_DP6 pid=3065384)[0;0m         ^^^^^^^^^^^^^^^^
[1;36m(EngineCore_DP5 pid=3065380)[0;0m     )
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     )
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     ^
[1;36m(EngineCore_DP5 pid=3065380)[0;0m   File "/fsx/ubuntu/uccl_yihan/vllm/vllm/model_executor/layers/fused_moe/modular_kernel.py", line 1013, in _fused_experts
[1;36m(EngineCore_DP6 pid=3065384)[0;0m   File "/fsx/ubuntu/miniconda3/lib/python3.13/site-packages/triton/runtime/jit.py", line 390, in <lambda>
[1;36m(EngineCore_DP5 pid=3065380)[0;0m     self.fused_experts.apply(
[1;36m(EngineCore_DP6 pid=3065384)[0;0m     return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)

this continues going deep into the driver file that I did not paste here. I don't know why this happened (maybe some parameter issue?).
https://docs.vllm.ai/en/stable/serving/expert_parallel_deployment.html#configuration According to this, I set all of that, I am not sure if export VLLM_MOE_DP_CHUNK_SIZE = matters the memory size that triton will allocate, right now I just set it to be a number that is large enough,i am checking if it can cause some memory issue