Use Ollama /api/embed endpoint for vector embeddings (#5221) * Use Ollama /api/embed endpoint for vector embeddings The deprecated /api/embeddings endpoint does not properly support the truncate parameter, causing "input length exceeds context length" errors when vectorizing files. Migrate to /api/embed which correctly handles truncation and supports native batch input. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * Wrap single Ollama vector calculation into batch Fixes https://github.com/SillyTavern/SillyTavern/pull/5221/changes#r2850052729 --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com> Co-authored-by: Cohee <18619528+Cohee1207@users.noreply.github.com>

d789efba0708097fb5716b6365ea3fcf830aa6f2

shifusen329 <quanhong329@gmail.com>

Signed

src/vectors/ollama-vectors.js+19 -23 Viewed

  * @returns {Promise<number[][]>} - The array of vectors for the texts
  */
 export async function getOllamaBatchVector(texts, apiUrl, model, keep, directories) {
-    const result = [];
-    for (const text of texts) {
-        const vector = await getOllamaVector(text, apiUrl, model, keep, directories);
-        result.push(vector);
-    }
-    return result;
-}
-/**
- * Gets the vector for the given text from Ollama
- * @param {string} text - The text to get the vector for
- * @param {string} apiUrl - The API URL
- * @param {string} model - The model to use
- * @param {boolean} keep - Keep the model loaded in memory
- * @param {import('../users.js').UserDirectoryList} directories - The directories object for the user
- * @returns {Promise<number[]>} - The vector for the text
- */
-export async function getOllamaVector(text, apiUrl, model, keep, directories) {
     const url = new URL(apiUrl);
-    url.pathname = '/api/embeddings';
+    url.pathname = '/api/embed';
     const headers = {};
     setAdditionalHeadersByType(headers, TEXTGEN_TYPES.OLLAMA, apiUrl, directories);

             ...headers,
         },
         body: JSON.stringify({
-            prompt: text,
+            input: texts,
             model: model,
             keep_alive: keep ? -1 : undefined,
             truncate: true,

     if (!response.ok) {
         const responseText = await response.text();
-        throw new Error(`Ollama: Failed to get vector for text: ${response.statusText} ${responseText}`);
+        throw new Error(`Ollama: Failed to get batch vectors: ${response.statusText} ${responseText}`);
+    }
     /** @type {any} */
     const data = await response.json();
-    if (!Array.isArray(data?.embedding)) {
+    if (!Array.isArray(data?.embeddings)) {
         throw new Error('API response was not an array');
+    }
-    return data.embedding;
+    return data.embeddings;
+}
+/**
+ * Gets the vector for the given text from Ollama
+ * @param {string} text - The text to get the vector for
+ * @param {string} apiUrl - The API URL
+ * @param {string} model - The model to use
+ * @param {boolean} keep - Keep the model loaded in memory
+ * @param {import('../users.js').UserDirectoryList} directories - The directories object for the user
+ * @returns {Promise<number[]>} - The vector for the text
+ */
+export async function getOllamaVector(text, apiUrl, model, keep, directories) {
+    const vectors = await getOllamaBatchVector([text], apiUrl, model, keep, directories);
+    return vectors[0];
+}