fivetech commited on
Commit
2ea883c
·
verified ·
1 Parent(s): 5b2e278

Fix MiMo: plantilla ChatML explícita (el GGUF la embebe con sintaxis que wllama no parsea: assert chat_templates)

Browse files
Files changed (1) hide show
  1. index.html +3 -0
index.html CHANGED
@@ -546,8 +546,11 @@ async function loadModel(repo, file) {
546
  wllama = new Wllama({
547
  'default': 'https://cdn.jsdelivr.net/npm/@wllama/wllama@3.5.1/esm/wasm/wllama.wasm',
548
  }, { parallelDownloads: 5 });
 
 
549
  await wllama.loadModelFromUrl(url, {
550
  n_gpu_layers: 999,
 
551
  n_ctx: CONFIG.nCtx,
552
  n_batch: 512,
553
  n_threads: Math.max(1, Math.min(8, navigator.hardwareConcurrency || 4)),
 
546
  wllama = new Wllama({
547
  'default': 'https://cdn.jsdelivr.net/npm/@wllama/wllama@3.5.1/esm/wasm/wllama.wasm',
548
  }, { parallelDownloads: 5 });
549
+ // Plantilla ChatML explícita: la embebida en este GGUF no la acepta el parser de wllama (assert chat_templates).
550
+ const CHAT_TEMPLATE = '{% for message in messages %}<|im_start|>{{ message.role }}\n{{ message.content }}<|im_end|>\n{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}';
551
  await wllama.loadModelFromUrl(url, {
552
  n_gpu_layers: 999,
553
+ chat_template: CHAT_TEMPLATE,
554
  n_ctx: CONFIG.nCtx,
555
  n_batch: 512,
556
  n_threads: Math.max(1, Math.min(8, navigator.hardwareConcurrency || 4)),