[update] update model ctx len

Author: LittleMouse

projects/llm_framework/main_cosy_voice/models/mode_CosyVoice2-0.5B-ax650.json

@@ -54,7 +54,7 @@
         "n_timesteps": 10,
         "b_use_mmap_load_embed": true,
         "b_dynamic_load_axmodel_layer": false,
-        "precompute_len": 100,
+        "precompute_len": 200,
         "cmm_size": 1227676,
         "ext_scripts": [
             "tokenizer_CosyVoice2-0.5B-ax650.py",

projects/llm_framework/main_cosy_voice/src/main.cpp

@@ -15,6 +15,7 @@
 #include <unistd.h>
 #include <base64.h>
 #include <fstream>
+#include <future>
 #include <stdexcept>
 #include <samplerate.h>
 #include <semaphore.h>
@@ -407,22 +408,36 @@ class llm_task {
         timer time_total;
         time_total.start();
         try {
-            int llm_ret          = 0;
-            auto llm_thread_func = [this, &text, &prompt_text_embeds, &prompt_speech_embeds, &llm_ret]() {
+            int llm_ret = 0;
+            std::promise<int> prom;
+            std::future<int> fut = prom.get_future();
+            auto llm_thread_func = [this, &text, &prompt_text_embeds, &prompt_speech_embeds, &llm_ret, &prom]() {
                 llm_ret = lLaMa_->Run(text, prompt_text_embeds, prompt_speech_embeds, g_token_buffer, g_buffer_mutex,
                                       g_buffer_cv, g_llm_finished);
+                prom.set_value(llm_ret);
             };
             std::thread llm_thread(llm_thread_func);
             llm_thread.detach();
+
+            if (fut.wait_for(std::chrono::milliseconds(10)) == std::future_status::ready) {
+                int llm_ret = fut.get();
+                if (llm_ret == -1) {
+                    SLOGE("Error, Generate failed");
+                    if (out_callback_) out_callback_("Error, Generate failed", true);
+                    return llm_ret;
+                }
+            }
+
             int prompt_token_len = prompt_speech_embeds_flow.size() / lToken2Wav._attr.flow_embed_size;
             if (prompt_token_len < 75) {
                 SLOGE("Error, prompt speech token len %d < 75", prompt_token_len);
                 if (llm_thread.joinable()) llm_thread.join();
+                if (out_callback_) {
+                    out_callback_("Error, prompt speech token len %d < 75", true);
+                }
                 return -1;
             }
-            if (llm_ret == -1) {
-                return llm_ret;
-            }
+
             int prompt_token_align_len = 75;
             std::vector<float> prompt_speech_embeds_flow1;
             prompt_speech_embeds_flow1.insert(prompt_speech_embeds_flow1.begin(), prompt_speech_embeds_flow.begin(),