@@ -14,11 +14,11 @@
struct gpt_params {
int32_t seed = -1; // RNG seed
- int32_t n_threads;
+ int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
int32_t n_predict = 128; // new tokens to predict
int32_t repeat_last_n = 64; // last n tokens to penalize
int32_t n_ctx = 512; //context size
-
+
// sampling parameters
int32_t top_k = 40;
float top_p = 0.95f;