Super tiny enable draft-weights-cpu-backup to avoid MTP acc len issue (#971)

Author: fzyzcjy

docs/en/advanced/speculative-decoding.md

@@ -11,6 +11,7 @@ For models with MTP layers (e.g., GLM-4.6, DeepSeek-V3/R1), simply add:
 --sglang-speculative-num-steps 3
 --sglang-speculative-eagle-topk 1
 --sglang-speculative-num-draft-tokens 4
+--sglang-enable-draft-weights-cpu-backup
 ```
 
 If you want to use a separately trained draft model (e.g., one trained with [SpecForge](https://docs.sglang.ai/SpecForge/)), also set:

scripts/run-glm4.5-355B-A32B.sh

@@ -122,6 +122,7 @@ SGLANG_ARGS=(
    --sglang-speculative-num-steps 1
    --sglang-speculative-eagle-topk 1
    --sglang-speculative-num-draft-tokens 2
+   --sglang-enable-draft-weights-cpu-backup
 
 )
 

scripts/run-mimo-7B-rl-eagle.sh

@@ -113,6 +113,7 @@ SGLANG_ARGS=(
    --sglang-speculative-num-steps 3
    --sglang-speculative-eagle-topk 1
    --sglang-speculative-num-draft-tokens 4
+   --sglang-enable-draft-weights-cpu-backup
 )
 
 MISC_ARGS=(

scripts/run-qwen3-next-80B-A3B.sh

@@ -129,6 +129,7 @@ SGLANG_ARGS=(
    --sglang-speculative-num-steps 2
    --sglang-speculative-eagle-topk 1
    --sglang-speculative-num-draft-tokens 3
+   --sglang-enable-draft-weights-cpu-backup
 
    --sglang-max-running-requests 512
 )