Use different language for prompt size too large. (#3004)

Signed-off-by: Adam Treat <treat.adam@gmail.com>
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
Co-authored-by: Jared Van Bortel <jared@nomic.ai>

gpt4all-backend/src/llmodel_shared.cpp

@@ -161,7 +161,9 @@ bool LLModel::decodePrompt(std::function<bool(int32_t)> promptCallback,
                            std::vector<Token> embd_inp,
                            bool isResponse) {
     if ((int) embd_inp.size() > promptCtx.n_ctx - 4) {
-        responseCallback(-1, "ERROR: The prompt size exceeds the context window size and cannot be processed.");
+        // FIXME: (Adam) We should find a way to bubble these strings to the UI level to allow for
+        // translation
+        responseCallback(-1, "Your message was too long and could not be processed. Please try again with something shorter.");
         std::cerr << implementation().modelType() << " ERROR: The prompt is " << embd_inp.size() <<
             " tokens and the context window is " << promptCtx.n_ctx << "!\n";
         return false;

gpt4all-bindings/python/CHANGELOG.md

@@ -11,6 +11,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Changed
 - Rebase llama.cpp on latest upstream as of September 26th ([#2998](https://github.com/nomic-ai/gpt4all/pull/2998))
+- Change the error message when a message is too long ([#3004](https://github.com/nomic-ai/gpt4all/pull/3004))
 
 ## [2.8.2] - 2024-08-14
 

gpt4all-chat/CHANGELOG.md

@@ -11,6 +11,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Changed
 - Rebase llama.cpp on latest upstream as of September 26th ([#2998](https://github.com/nomic-ai/gpt4all/pull/2998))
+- Change the error message when a message is too long ([#3004](https://github.com/nomic-ai/gpt4all/pull/3004))
 
 ### Fixed
 - Fix a crash when attempting to continue a chat loaded from disk ([#2995](https://github.com/nomic-ai/gpt4all/pull/2995))

gpt4all-chat/src/chatllm.cpp

@@ -706,6 +706,9 @@ bool ChatLLM::handleResponse(int32_t token, const std::string &response)
 #endif
 
     // check for error
+    // FIXME (Adam) The error messages should not be treated as a model response or part of the
+    // normal conversation. They should be serialized along with the conversation, but the strings
+    // are separate and we should preserve info that these are error messages and not actual model responses.
     if (token < 0) {
         m_response.append(response);
         m_trimmedResponse = remove_leading_whitespace(m_response);