[BugFix][CI] Fix GLM5.1-W8A8 MTP load weight error when vLLM is updated to v0.21.0 (#10317)
What this PR does / why we need it?
After vllm PR #41706,
GlmMoeDsaForCausalLM.load_weights uses AutoWeightsLoader which does
not skip rot.weight of the mtp layer, and causes ValueError while
loading weights.
Does this PR introduce any user-facing change?
No.
How was this patch tested?
The model weights can be correctly loaded, and the acceptance rate of
draft model is also right.

(APIServer pid=774401) INFO 06-11 11:07:59 [loggers.py:271] Engine
000: Avg prompt throughput: 1.0 tokens/s, Avg generation throughput: 7.1
tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.1%,
Prefix cache hit rate: 0.0%
(APIServer pid=774401) INFO 06-11 11:07:59 [metrics.py:101] SpecDecoding
metrics: Mean acceptance length: 3.50, Accepted throughput: 0.94
tokens/s, Drafted throughput: 1.13 tokens/s, Accepted: 50 tokens,
Drafted: 60 tokens, Per-position acceptance rate: 1.000, 0.900, 0.600,
Avg Draft acceptance rate: 83.3%
(APIServer pid=774401) INFO 06-11 11:08:09 [loggers.py:271] Engine 000:
Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 8.5
tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.3%,
Prefix cache hit rate: 0.0%
(APIServer pid=774401) INFO 06-11 11:08:09 [metrics.py:101] SpecDecoding
metrics: Mean acceptance length: 2.58, Accepted throughput: 5.20
tokens/s, Drafted throughput: 9.90 tokens/s, Accepted: 52 tokens,
Drafted: 99 tokens, Per-position acceptance rate: 0.818, 0.606, 0.152,
Avg Draft acceptance rate: 52.5%
(APIServer pid=774401) INFO 06-11 11:08:19 [loggers.py:271] Engine 000:
Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 10.0
tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.4%,
Prefix cache hit rate: 0.0%
(APIServer pid=774401) INFO 06-11 11:08:19 [metrics.py:101] SpecDecoding
metrics: Mean acceptance length: 3.03, Accepted throughput: 6.70
tokens/s, Drafted throughput: 9.90 tokens/s, Accepted: 67 tokens,
Drafted: 99 tokens, Per-position acceptance rate: 0.909, 0.697, 0.424,
Avg Draft acceptance rate: 67.7%
(APIServer pid=774401) INFO 06-11 11:08:29 [loggers.py:271] Engine 000:
Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 9.5
tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.4%,
Prefix cache hit rate: 0.0%
(APIServer pid=774401) INFO 06-11 11:08:29 [metrics.py:101] SpecDecoding
metrics: Mean acceptance length: 2.79, Accepted throughput: 6.10
tokens/s, Drafted throughput: 10.20 tokens/s, Accepted: 61 tokens,
Drafted: 102 tokens, Per-position acceptance rate: 0.824, 0.588, 0.382,
Avg Draft acceptance rate: 59.8%


vLLM version: v0.21.0
vLLM main:
https://github.com/vllm-project/vllm/commit/9090368b650896bf5fc990c921df7eb4c20355a5

Signed-off-by: Wangbingjie wangbj1207@126.com