Update index.html

index.html

@@ -1161,7 +1161,9 @@
             const generationEnd = performance.now();
             const totalTime = generationEnd - generationStart;
             const inputTokenCount = inputs.input_ids.dims[1];
-            const outputTokenCount = output.length - inputTokenCount;
+            // Handle tensor/array output properly
+            const totalTokenCount = Array.isArray(output) ? output.length : (output.size || output.length || 0);
+            const outputTokenCount = Math.max(0, totalTokenCount - inputTokenCount);
             const tps = outputTokenCount / (totalTime / 1000);
             
             // Store metrics
@@ -1172,6 +1174,7 @@
             console.log(`📊 Performance Metrics:`);
             console.log(`   - Total generation time: ${totalTime.toFixed(2)}ms`);
             console.log(`   - Input tokens: ${inputTokenCount}`);
+            console.log(`   - Total tokens: ${totalTokenCount}`);
             console.log(`   - Output tokens: ${outputTokenCount}`);
             console.log(`   - Tokens Per Second (TPS): ${tps.toFixed(2)} tok/s`);
             
@@ -1179,7 +1182,7 @@
             document.getElementById("metric-ttft").innerText = `${ttft.toFixed(0)}ms`;
             document.getElementById("metric-tps").innerText = `${tps.toFixed(1)} t/s`;
             
-            const decoded = tokenizer.decode(output.slice(0, [inputs.input_ids.dims[1], null]), { skip_special_tokens: false });
+            const decoded = tokenizer.decode(output.slice(inputs.input_ids.dims[1]), { skip_special_tokens: false });
             // 5. Parse Output
             // Format: <start_function_call>call:add{...}<end_function_call>
             const startTag = "<start_function_call>";