[Question] Why configs/qwen3-30B-A3B-eagle3.json draft_vocab_size=32000

[huang3eng]

Why is the draft_vocab_size in configs/qwen3-30B-A3B-eagle3.json 32,000 instead of the vocab_size 151,936 in qwen3-30B-A3B?

[FrankLeeeee]

The draft model will count the most frequent vocab and using a small vocab is much faster and memory efficient. We followed the original implementation of EAGLE3.

[huang3eng]

Alright, I understand. When I configure eagle3 in the sglang of the RL framework slime, if the draft vocab size is not equal to the vocab size, an error will occur in the first rollout stage.

[FrankLeeeee]

What is the error like?