Table 3 Wall time of the model’s inference on the USPTO MIT test set in reaction product prediction with beam search (BS) and speculative beam search (SBS). The number of generated sequences (denoted by “K”) is 5. “B” stands for batch size. The number of drafts and the draft length in the speculative beam search is different for every batch size. The average time and the standard deviation are estimated based on five runs

From: Accelerating the inference of string generation-based chemical reaction models for industrial applications