Windows test

Author: ivanstepanovftw

tests/test-sampling.cpp

@@ -1,6 +1,6 @@
 #include "llama.h"
-#include <assert.h>
-#include <math.h>
+#include <cassert>
+#include <cmath>
 #include <numeric>
 #include <cassert>
 #include <iostream>
@@ -20,15 +20,19 @@ void dump(const llama_token_data_array * candidates) {
 void test_top_k(const std::vector<float> & probs,
                 const std::vector<float> & expected_probs,
                 int k) {
+    printf("%s:%d (%s)\n", __FILE__, __LINE__, __func__);
     size_t n_vocab = probs.size();
     std::vector<llama_token_data> candidates;
     candidates.reserve(n_vocab);
     for (llama_token token_id = 0; token_id < (llama_token)n_vocab; token_id++) {
-        float logit = log(probs[token_id]);
+        printf("%s:%d (%s) token_id: %d\n", __FILE__, __LINE__, __func__, token_id);
+        float logit = std::log(probs[token_id]);
         candidates.emplace_back(llama_token_data{token_id, logit, 0.0f});
     }
+    printf("%s:%d (%s)\n", __FILE__, __LINE__, __func__);
 
     llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
+    printf("%s:%d (%s)\n", __FILE__, __LINE__, __func__);
     llama_sample_softmax(nullptr, &candidates_p);
     DUMP(&candidates_p);
     llama_sample_top_k(nullptr, &candidates_p, k);