diff --git a/vllm/v1/spec_decode/ngram_proposer.py b/vllm/v1/spec_decode/ngram_proposer.py index fd8e0a6fd1d2..aed050a3540c 100644 --- a/vllm/v1/spec_decode/ngram_proposer.py +++ b/vllm/v1/spec_decode/ngram_proposer.py @@ -71,8 +71,8 @@ class NgramProposer: Args: valid_ngram_requests: Set of indices of requests that need ngram proposals. - num_tokens_no_spec: - Numpy array of shape (batch_size,) representing the number + num_tokens_no_spec: + Numpy array of shape (batch_size,) representing the number of tokens without speculative tokens for each request. token_ids_cpu: Numpy array of shape (batch_size, max_model_len)