Update index.html

index.html

@@ -354,7 +354,8 @@
         await model.generate(enc.input_ids, {
           attention_mask: enc.attention_mask,
           max_new_tokens: 1,
-          do_sample: false
+          do_sample: false,
+          use_cache: false    // avoid cached branch on first pass
         });
       } catch (e) {
         console.warn("Warm-up skipped:", e);
@@ -392,15 +393,31 @@
       const padId = tokenizer?.pad_token_id ?? eosId;
     
       const t0 = performance.now();
-      const gen = await model.generate(enc.input_ids, {
-        attention_mask: enc.attention_mask,
-        max_new_tokens: 1,
-        do_sample: false,
-        output_scores: true,
-        return_dict_in_generate: true,
-        eos_token_id: eosId,
-        pad_token_id: padId
-      });
+      let gen;
+      try {
+        // try normal path first
+        gen = await model.generate(enc.input_ids, {
+          attention_mask: enc.attention_mask,
+          max_new_tokens: 1,
+          do_sample: false,
+          output_scores: true,
+          return_dict_in_generate: true,
+          eos_token_id: eosId,
+          pad_token_id: padId
+        });
+      } catch (e) {
+        console.warn("generate() failed; retrying with use_cache:false", e);
+        gen = await model.generate(enc.input_ids, {
+          attention_mask: enc.attention_mask,
+          max_new_tokens: 1,
+          do_sample: false,
+          output_scores: true,
+          return_dict_in_generate: true,
+          eos_token_id: eosId,
+          pad_token_id: padId,
+          use_cache: false      // <-- key change
+        });
+      }
       const dt = (performance.now() - t0) | 0;
     
       // --- from here, keep your existing logits -> softmax -> Top-K code ---