Results
2
comments of
大帆
I have a question that Qwen3-30B-A3B eagle head_dim is 64, but head_dim of Qwen3-30B-A3B is 128. They are different.
Hello. I have run this on vLLM with num_spec_tokens=1(draft token=1). When testing with the GSM8K dataset, the accept ratio came out to be 60%. Would you please tell me the...