Tiny refactor computation of shared expert fusion weights

[fzyzcjy]

Motivation

This PR depends on #5188

It can be useful in #5143 (review) and #5101

python -m sglang.launch_server    --model-path deepseek-ai/DeepSeek-V3-0324 --trust-remote-code    --tp 16 --disable-radix-cache    --host 0.0.0.0 --port 20000 --dist-init-addr 10.10.37.16:37000 --nnodes 2 --node-rank 0

(cd /host_home/primary_synced/sglang && python3 benchmark/gsm8k/bench_sglang.py --port 20000 --parallel 1400 --num-questions 1400)

gsm8k: 93.8

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 and Accuracy Results.
• For reviewers: If you haven't made any contributions to this PR and are only assisting with merging the main branch, please remove yourself as a co-author when merging the PR.
• Please feel free to join our Slack channel at https://slack.sglang.ai to discuss your PR.

[fzyzcjy]

Ping me when this PR is ready for merge / needs modification, then I will resolve conflicts

[lambert0312]

Ping me when this PR is ready for merge / needs modification, then I will resolve conflicts

@fzyzcjy Could you please remove the changes in deepseek_v2.py in this PR? I have resolved the conflicts and updated this part of the changes in #5143, and also attached the changes of nextn, thank you!
This PR is to extract the logic of repeated shared expert fusion weights calculations.

[fzyzcjy]

Could you please remove the changes in deepseek_v2.py in this PR?

Hmm, to double check: do you mean making deepseek_v2.py have no diff, and only add a new function (that is not called yet) in the weight_utils?

[lambert0312]

Hmm, to double check: do you mean making deepseek_v2.py have no diff, and only add a new function (that is not called yet) in the weight_utils?

yes

[fzyzcjy]

OK I will do that when this PR is going to be merged

[lambert0312]

I think this PR can be merged. Please complete the modification as soon as possible. Thank you!

[fzyzcjy]

Yes I think so, I will have time tomorrow for it hopefully

[lambert0312]

Yes I think so, I will have time tomorrow for it hopefully

A simpler MTP loading method has been refactored, and this PR does not seem to be needed @fzyzcjy