n_batch configurable por URL (?batch=) para entornos con poca RAM

index.html

@@ -550,7 +550,7 @@ async function loadModel(repo, file) {
     await wllama.loadModelFromUrl(url, {
       n_gpu_layers: 999,
       n_ctx: CONFIG.nCtx,
-      n_batch: 512,
+      n_batch: parseInt(params.get('batch')) || 512,
       n_threads: Math.max(1, Math.min(8, navigator.hardwareConcurrency || 4)),
       progressCallback: ({ loaded, total }) => {
         if (total > 0) {