stream : disable temperature fallback

For real-time processing, we always want a single decoder running at T=0

examples/main/main.cpp

@@ -654,6 +654,7 @@ int main(int argc, char ** argv) {
 
             wparams.greedy.best_of        = params.best_of;
             wparams.beam_search.beam_size = params.beam_size;
+            wparams.temperature_inc = -1;
 
             wparams.prompt_tokens     = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
             wparams.prompt_n_tokens   = prompt_tokens.empty() ? 0       : prompt_tokens.size();

examples/stream.wasm/emscripten.cpp

@@ -49,6 +49,9 @@ void stream_main(size_t index) {
     wparams.max_tokens       = 32;
     wparams.audio_ctx        = 768; // partial encoder context for better performance
 
+    // disable temperature fallback
+    wparams.temperature_inc  = -1.0f;
+
     wparams.language         = "en";
 
     printf("stream: using %d threads\n", wparams.n_threads);

examples/stream/stream.cpp

@@ -615,9 +615,8 @@ int main(int argc, char ** argv) {
             wparams.audio_ctx        = params.audio_ctx;
             wparams.speed_up         = params.speed_up;
 
-            // disable best_of fallback
+            // disable temperature fallback
             wparams.temperature_inc  = -1.0f;
-            wparams.greedy.best_of   = -1;
 
             wparams.prompt_tokens    = params.no_context ? nullptr : prompt_tokens.data();
             wparams.prompt_n_tokens  = params.no_context ? 0       : prompt_tokens.size();