Function get_token

smallthinker/common/ngram-cache.cpp:55–57

Helper function to get a token from the combined, speculative sequence of inp and draft.

// Helper function to get a token from the combined, speculative sequence of inp and draft.
static llama_token get_token(const std::vector<llama_token> & inp, const std::vector<llama_token> & draft, const size_t i) {
    return i < inp.size() ? inp[i] : draft[1 + i - inp.size()];
}

// If sample size or percentage are below these thresholds the draft is aborted early:
constexpr int    draft_min_sample_size_lax[LLAMA_NGRAM_MAX] = { 2,  2,  1,  1};