From 88bbe3095226cb0b5f5ad25110753ee05a15bbc2 Mon Sep 17 00:00:00 2001
From: Adam Treat <treat.adam@gmail.com>
Date: Tue, 11 Jul 2023 12:09:33 -0400
Subject: [PATCH] Provide a guardrail for OOM errors.

---
 gpt4all-chat/chatllm.cpp    | 11 +++++++++++
 gpt4all-chat/main.qml       |  2 +-
 gpt4all-chat/mysettings.cpp | 21 +++++++++++++++++++++
 gpt4all-chat/mysettings.h   |  4 ++++
 4 files changed, 37 insertions(+), 1 deletion(-)

diff --git a/gpt4all-chat/chatllm.cpp b/gpt4all-chat/chatllm.cpp
index 4f87d150..3a42021b 100644
--- a/gpt4all-chat/chatllm.cpp
+++ b/gpt4all-chat/chatllm.cpp
@@ -211,6 +211,15 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo)
     // Store the file info in the modelInfo in case we have an error loading
     m_llModelInfo.fileInfo = fileInfo;
 
+    // Check if we've previously tried to load this file and failed/crashed
+    if (MySettings::globalInstance()->attemptModelLoad() == filePath) {
+        MySettings::globalInstance()->setAttemptModelLoad(QString()); // clear the flag
+        if (!m_isServer)
+            LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store
+        m_llModelInfo = LLModelInfo();
+        emit modelLoadingError(QString("Previous attempt to load model resulted in crash for `%1` most likely due to out of memory. You should either remove this model or decrease your system RAM by closing other applications.").arg(modelInfo.filename()));
+    }
+
     if (fileInfo.exists()) {
         if (isChatGPT) {
             QString apiKey;
@@ -239,7 +248,9 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo)
 #endif
 
             if (m_llModelInfo.model) {
+                MySettings::globalInstance()->setAttemptModelLoad(filePath);
                 bool success = m_llModelInfo.model->loadModel(filePath.toStdString());
+                MySettings::globalInstance()->setAttemptModelLoad(QString());
                 if (!success) {
                     delete std::exchange(m_llModelInfo.model, nullptr);
                     if (!m_isServer)
diff --git a/gpt4all-chat/main.qml b/gpt4all-chat/main.qml
index 94f86e4b..5bb723bc 100644
--- a/gpt4all-chat/main.qml
+++ b/gpt4all-chat/main.qml
@@ -163,7 +163,7 @@ Window {
             + "<i>\"" + currentChat.modelLoadingError + "\"</i>"
             + qsTr("<br><br>Model loading failures can happen for a variety of reasons, but the most common "
             + "causes include a bad file format, an incomplete or corrupted download, the wrong file "
-            + "type or an incompatible model type. Here are some suggestions for resolving the problem:"
+            + "type, not enough system RAM or an incompatible model type. Here are some suggestions for resolving the problem:"
             + "<br><ul>"
             + "<li>Ensure the model file has a compatible ggml format and type"
             + "<li>Check the model file is complete in the download folder"
diff --git a/gpt4all-chat/mysettings.cpp b/gpt4all-chat/mysettings.cpp
index 9a4e17ff..71b2334d 100644
--- a/gpt4all-chat/mysettings.cpp
+++ b/gpt4all-chat/mysettings.cpp
@@ -608,3 +608,24 @@ void MySettings::setNetworkUsageStatsActive(bool b)
     setting.sync();
     emit networkUsageStatsActiveChanged();
 }
+
+QString MySettings::attemptModelLoad() const
+{
+    QSettings setting;
+    setting.sync();
+    return setting.value("attemptModelLoad", QString()).toString();
+}
+
+void MySettings::setAttemptModelLoad(const QString &modelFile)
+{
+    if (attemptModelLoad() == modelFile)
+        return;
+
+    QSettings setting;
+    if (modelFile.isEmpty())
+        setting.remove("attemptModelLoad");
+    else
+        setting.setValue("attemptModelLoad", modelFile);
+    setting.sync();
+    emit attemptModelLoadChanged();
+}
diff --git a/gpt4all-chat/mysettings.h b/gpt4all-chat/mysettings.h
index 873cf5ec..2d264666 100644
--- a/gpt4all-chat/mysettings.h
+++ b/gpt4all-chat/mysettings.h
@@ -93,6 +93,9 @@ public:
     bool networkUsageStatsActive() const;
     void setNetworkUsageStatsActive(bool b);
 
+    QString attemptModelLoad() const;
+    void setAttemptModelLoad(const QString &modelFile);
+
 Q_SIGNALS:
     void nameChanged(const ModelInfo &model);
     void filenameChanged(const ModelInfo &model);
@@ -119,6 +122,7 @@ Q_SIGNALS:
     void networkAttributionChanged();
     void networkIsActiveChanged();
     void networkUsageStatsActiveChanged();
+    void attemptModelLoadChanged();
 
 private:
     bool m_forceMetal;