Implement support for selecting basic keep_alive times for Ollama (#1920)

frontend/src/components/LLMSelection/OllamaLLMOptions/index.jsx

@@ -111,6 +111,35 @@ export default function OllamaLLMOptions({ settings }) {
               Enter the URL where Ollama is running.
             </p>
           </div>
+
+          <div className="flex flex-col w-60">
+            <label className="text-white text-sm font-semibold block mb-2">
+              Ollama Keep Alive
+            </label>
+            <select
+              name="OllamaLLMKeepAliveSeconds"
+              required={true}
+              className="bg-zinc-900 border-gray-500 text-white text-sm rounded-lg block w-full p-2.5"
+              defaultValue={settings?.OllamaLLMKeepAliveSeconds ?? "300"}
+            >
+              <option value="0">No cache</option>
+              <option value="300">5 minutes</option>
+              <option value="3600">1 hour</option>
+              <option value="-1">Forever</option>
+            </select>
+            <p className="text-xs leading-[18px] font-base text-white text-opacity-60 mt-2">
+              Choose how long Ollama should keep your model in memory before
+              unloading.
+              <a
+                className="underline text-blue-300"
+                href="https://github.com/ollama/ollama/blob/main/docs/faq.md#how-do-i-keep-a-model-loaded-in-memory-or-make-it-unload-immediately"
+                target="_blank"
+              >
+                {" "}
+                Learn more &rarr;
+              </a>
+            </p>
+          </div>
         </div>
       </div>
     </div>

server/models/systemSettings.js

@@ -411,6 +411,7 @@ const SystemSettings = {
       OllamaLLMBasePath: process.env.OLLAMA_BASE_PATH,
       OllamaLLMModelPref: process.env.OLLAMA_MODEL_PREF,
       OllamaLLMTokenLimit: process.env.OLLAMA_MODEL_TOKEN_LIMIT,
+      OllamaLLMKeepAliveSeconds: process.env.OLLAMA_KEEP_ALIVE_TIMEOUT ?? 300,
 
       // TogetherAI Keys
       TogetherAiApiKey: !!process.env.TOGETHER_AI_API_KEY,

server/utils/AiProviders/ollama/index.js

@@ -13,6 +13,9 @@ class OllamaAILLM {
 
     this.basePath = process.env.OLLAMA_BASE_PATH;
     this.model = modelPreference || process.env.OLLAMA_MODEL_PREF;
+    this.keepAlive = process.env.OLLAMA_KEEP_ALIVE_TIMEOUT
+      ? Number(process.env.OLLAMA_KEEP_ALIVE_TIMEOUT)
+      : 300; // Default 5-minute timeout for Ollama model loading.
     this.limits = {
       history: this.promptWindowLimit() * 0.15,
       system: this.promptWindowLimit() * 0.15,
@@ -28,6 +31,7 @@ class OllamaAILLM {
     return new ChatOllama({
       baseUrl: this.basePath,
       model: this.model,
+      keepAlive: this.keepAlive,
       useMLock: true,
       temperature,
     });

server/utils/helpers/updateENV.js

@@ -101,6 +101,10 @@ const KEY_MAPPING = {
     envKey: "OLLAMA_MODEL_TOKEN_LIMIT",
     checks: [nonZero],
   },
+  OllamaLLMKeepAliveSeconds: {
+    envKey: "OLLAMA_KEEP_ALIVE_TIMEOUT",
+    checks: [isInteger],
+  },
 
   // Mistral AI API Settings
   MistralApiKey: {
@@ -454,6 +458,11 @@ function nonZero(input = "") {
   return Number(input) <= 0 ? "Value must be greater than zero" : null;
 }
 
+function isInteger(input = "") {
+  if (isNaN(Number(input))) return "Value must be a number";
+  return Number(input);
+}
+
 function isValidURL(input = "") {
   try {
     new URL(input);