mr-memory 2.18.2 → 2.18.4

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mr-memory",
-  "version": "2.18.2",
+  "version": "2.18.4",
   "description": "MemoryRouter persistent memory plugin for OpenClaw — your AI remembers every conversation",
   "type": "module",
   "files": [

package/upload.ts

@@ -143,7 +143,8 @@ type MemoryLine = {
 const MAX_ITEM_CHARS = 8000;
 const TARGET_CHUNK_CHARS = 4000;
 const MAX_BATCH_BYTES = 2_000_000;
-const MAX_BATCH_COUNT = 100;
+const MAX_BATCH_COUNT_DEFAULT = 100;
+const MAX_BATCH_COUNT_QWEN = 25;
 const BATCH_SLEEP_MS = 150;
 const MAX_HTTP_RETRIES = 3;
 
@@ -402,7 +403,8 @@ export async function runUpload(params: {
     return;
   }
 
-  console.log(`Uploading ${files.length} files to MemoryRouter...`);
+  const modelLabel = embeddings ? `(model: ${embeddings})` : "(model: bge)";
+  console.log(`Uploading ${files.length} files to MemoryRouter ${modelLabel}...`);
 
   const allLines: MemoryLine[] = [];
   let skippedEmpty = 0;
@@ -429,14 +431,16 @@ export async function runUpload(params: {
     return;
   }
 
-  // Batch
+  // Batch — smaller batches for Qwen (4096-dim vectors are 4x heavier)
+  const isQwen = embeddings && ['qwen', 'qwen3', 'qwen3-8b', 'qwen3-embedding', 'qwen3-embedding-8b'].includes(embeddings.toLowerCase());
+  const maxBatchCount = isQwen ? MAX_BATCH_COUNT_QWEN : MAX_BATCH_COUNT_DEFAULT;
   const batches: MemoryLine[][] = [];
   let currentBatch: MemoryLine[] = [];
   let currentBytes = 0;
 
   for (const line of allLines) {
     const lineBytes = JSON.stringify(line).length + 1;
-    if (currentBytes + lineBytes > MAX_BATCH_BYTES || currentBatch.length >= MAX_BATCH_COUNT) {
+    if (currentBytes + lineBytes > MAX_BATCH_BYTES || currentBatch.length >= maxBatchCount) {
       if (currentBatch.length > 0) batches.push(currentBatch);
       currentBatch = [line];
       currentBytes = lineBytes;