Update index.html

index.html

@@ -140,7 +140,7 @@
     const device = document.getElementById("device").value;
     const imageUrl = document.getElementById("image-url").value;
     const maxTokens = parseInt(document.getElementById("max-tokens").value) || 32;
-    const numRuns = parseInt(document.getElementById("num-runs").value) || 5;
+    const numRuns = parseInt(document.getElementById("num-runs").value) || 3;
     const doImageSplitting = document.getElementById("do-split").checked;
 
     const dtypeSettings = { decoder: decoder_dtype, embed: embed_dtype, vision: vision_dtype };
@@ -183,28 +183,23 @@
           const [processor, model] = await SmolVLM.getInstance(modelId, dtypeSettings, device);
           const text = processor.apply_chat_template(messages, { add_generation_prompt: true });
           const inputs = await processor(text, [image], { do_image_splitting: doImageSplitting });
-
-          let numTokens = 0;
-          let startTime;
-          let tps = 0;
-          const token_callback_function = () => {
-            startTime = startTime || performance.now();
-            tps = (numTokens++ / (performance.now() - startTime)) * 1000;
-          };
+          
+          const start = performance.now();
           const streamer = new TextStreamer(processor.tokenizer, {
             skip_prompt: true,
             skip_special_tokens: true,
-            token_callback_function,
           });
-
           await model.generate({
             ...inputs,
             max_new_tokens: maxTokens,
             min_new_tokens: maxTokens,
             streamer,
           });
+          const end = performance.now();
+          const elapsed = end - start;
+          
+          const tps = maxTokens / (elapsed / 1000);  // accurate TPS
 
-          const elapsed = performance.now() - start;
           totalTime += elapsed;
           totalTps += tps;
           runsResults.push({