speculative : print draft token pieces

ggerganov · ggerganov · commit 57c111f47fc3 · 2023-09-04T12:54:38.000+03:00
diff --git a/examples/speculative/speculative.cpp b/examples/speculative/speculative.cpp
@@ -241,7 +241,7 @@ int main(int argc, char ** argv) {
             llama_sample_softmax(ctx_dft, &cur_p);
 
             for (int i = 0; i < 3; ++i) {
-                LOG(" - draft candidate %d: %d (%.3f)\n", i, cur_p.data[i].id, cur_p.data[i].p);
+                LOG(" - draft candidate %3d: %6d (%8.3f) '%s'\n", i, cur_p.data[i].id, cur_p.data[i].p, llama_token_to_piece(ctx_dft, cur_p.data[i].id).c_str());
             }
 
             // too low probability, stop drafting

Original file line number	Diff line number	Diff line change
`@@ -241,7 +241,7 @@ int main(int argc, char ** argv) {`
`241`	`241`	`llama_sample_softmax(ctx_dft, &cur_p);`
`242`	`242`
`243`	`243`	`for (int i = 0; i < 3; ++i) {`
`244`		`- LOG(" - draft candidate %d: %d (%.3f)\n", i, cur_p.data[i].id, cur_p.data[i].p);`
	`244`	`+ LOG(" - draft candidate %3d: %6d (%8.3f) '%s'\n", i, cur_p.data[i].id, cur_p.data[i].p, llama_token_to_piece(ctx_dft, cur_p.data[i].id).c_str());`
`245`	`245`	`}`
`246`	`246`
`247`	`247`	`// too low probability, stop drafting`