sampling: fix top_k <= 0

Author: JohannesGaessler

llama.cpp

@@ -8371,6 +8371,10 @@ void llama_sample_top_k(struct llama_context * ctx, llama_token_data_array * can
     //     return;
     // }
 
+    if (k <= 0) {
+        k = candidates->size;
+    }
+
     const int64_t t_start_sample_us = ggml_time_us();
 
     k = std::max(k, (int) min_keep);

tests/test-sampling.cpp

@@ -235,6 +235,8 @@ int main(void) {
 
     test_top_k({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f}, 1);
     test_top_k({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f, 0.2f}, 3);
+    test_top_k({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f, 0.2f, 0.1f}, 4);
+    test_top_k({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f, 0.2f, 0.1f}, 0);
 
     test_top_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f}, 0);
     test_top_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f, 0.3f}, 0.7f);