Don't crash when prompt is too large.

llm.cpp

@@ -252,6 +252,13 @@ bool LLMObject::handleResponse(int32_t token, const std::string &response)
     fflush(stdout);
 #endif
 
+    // check for error
+    if (token < 0) {
+        m_response.append(response);
+        emit responseChanged();
+        return false;
+    }
+
     // Save the token to our prompt ctxt
     if (s_ctx.tokens.size() == s_ctx.n_ctx)
         s_ctx.tokens.erase(s_ctx.tokens.begin());

llmodel/gptj.cpp

@@ -707,6 +707,13 @@ void GPTJ::prompt(const std::string &prompt,
     // save the context size
     promptCtx.n_ctx = d_ptr->model.hparams.n_ctx;
 
+    if ((int) embd_inp.size() > promptCtx.n_ctx - 4) {
+        response(-1, "ERROR: The prompt size exceeds the context window size and cannot be processed.");
+        std::cerr << "GPT-J ERROR: The prompt is" << embd_inp.size() <<
+            "tokens and the context window is" << promptCtx.n_ctx << "!\n";
+        return;
+    }
+
     promptCtx.n_predict = std::min(promptCtx.n_predict, promptCtx.n_ctx - (int) embd_inp.size());
     promptCtx.n_past = std::min(promptCtx.n_past, promptCtx.n_ctx);
 

llmodel/llamamodel.cpp

@@ -102,7 +102,9 @@ void LLamaModel::prompt(const std::string &prompt,
     promptCtx.n_ctx = llama_n_ctx(d_ptr->ctx);
 
     if ((int) embd_inp.size() > promptCtx.n_ctx - 4) {
-        std::cerr << "LLAMA ERROR: prompt is too long\n";
+        response(-1, "The prompt size exceeds the context window size and cannot be processed.");
+        std::cerr << "LLAMA ERROR: The prompt is" << embd_inp.size() <<
+            "tokens and the context window is" << promptCtx.n_ctx << "!\n";
         return;
     }