fixup! lookup: evaluation tools, use corpus/previous gens

Author: JohannesGaessler

common/ngram-cache.cpp

@@ -8,8 +8,8 @@ void llama_ngram_cache_update(llama_ngram_cache & ngram_cache, int ngram_min, in
     const int64_t t_start_ms = ggml_time_ms();
     const int64_t inp_size = inp.size();
 
-    const int64_t n_todo = inp_size * (ngram_max - ngram_min + 1);
-    int64_t n_done = 0;
+    const long n_todo = inp_size * (ngram_max - ngram_min + 1);
+    long n_done = 0;
 
     for (int64_t ngram_size = ngram_min; ngram_size <= ngram_max; ++ngram_size) {
         const int64_t i_start = std::max(inp_size - nnew, ngram_size);
@@ -34,10 +34,10 @@ void llama_ngram_cache_update(llama_ngram_cache & ngram_cache, int ngram_min, in
             ++n_done;
 
             if (print_progress && n_done % 10000000 == 0) {
-                const int64_t t_now_ms = ggml_time_ms();
-                const int64_t eta_ms   = (inp_size*(ngram_max-ngram_min+1) - n_done) * (t_now_ms - t_start_ms) / n_done;
-                const int64_t eta_min  = eta_ms / (60*1000);
-                const int64_t eta_s    = (eta_ms - 60*1000*eta_min) / 1000;
+                const long t_now_ms = ggml_time_ms();
+                const long eta_ms   = (inp_size*(ngram_max-ngram_min+1) - n_done) * (t_now_ms - t_start_ms) / n_done;
+                const long eta_min  = eta_ms / (60*1000);
+                const long eta_s    = (eta_ms - 60*1000*eta_min) / 1000;
 
                 fprintf(stderr, "%s: %ld/%ld done, ETA: %02ld:%02ld\n", __func__, n_done, n_todo, eta_min, eta_s);
             }