From ee1a0ec9cb367ba41d138134795cbbbe93d2bf1c Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sat, 28 Oct 2023 14:23:11 +0300 Subject: llama : add option for greedy sampling with probs (#3813) * llama : add option for greedy sampling with probs * llama : add comment about llama_sample_token_greedy() missing probs * sampling : temp == 0.0 -> no probs, temp < 0.0 -> probs --- examples/speculative/speculative.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'examples') diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp index f921b784..323c7465 100644 --- a/examples/speculative/speculative.cpp +++ b/examples/speculative/speculative.cpp @@ -148,7 +148,7 @@ int main(int argc, char ** argv) { std::vector drafts(n_seq_dft); params.sparams.grammar.clear(); // the draft samplers will copy the target sampler's grammar - params.sparams.temp = std::max(0.01f, params.sparams.temp); + params.sparams.temp = -1.0f; // force greedy sampling with probs for the draft model for (int s = 0; s < n_seq_dft; ++s) { drafts[s].ctx_sampling = llama_sampling_init(params.sparams); -- cgit v1.2.3