Quick fix for wrong number of multi-outputs sometimes when streaming

PiperOrigin-RevId: 676573033

mediapipe/tasks/web/genai/llm_inference/llm_inference.ts

@@ -148,7 +148,7 @@ class Deferred<T> {
  */
 function roundUpToNearestEven(n: number): number {
   if (n === 1) return 1;
-  return (n + (n % 2));
+  return n + (n % 2);
 }
 
 /**
@@ -851,6 +851,11 @@ export class LlmInference extends TaskRunner {
         // Don't trigger the user progress listener if there are WebGPU errors.
         if (this.userProgressListener && this.wgpuErrors.length === 0) {
           if (this.isMultiResponseGeneration) {
+            // TODO: Remove this when we no longer need to have an
+            // even number of responses in multi-output.
+            if (decodedText.length > this.options.getNumResponses()) {
+              decodedText.pop();
+            }
             (this.userProgressListener as MultiResponseProgressListener)(
               decodedText,
               /* done= */ false,
@@ -1047,8 +1052,9 @@ export class LlmInference extends TaskRunner {
     llmGpuOptions.setSequenceBatchSize(0);
     // TODO: Remove this when we no longer need to have an even
     // number of responses in multi-output.
-    llmGpuOptions.setNumOutputHeads(roundUpToNearestEven(
-        this.options.getNumResponses()));
+    llmGpuOptions.setNumOutputHeads(
+      roundUpToNearestEven(this.options.getNumResponses()),
+    );
     llmGpuOptions.setSamplerParams(this.options.getSamplerParams());
 
     const gpuModelInfo = new LlmGpuCalculatorOptions.GpuModelInfo();
@@ -1097,8 +1103,9 @@ export class LlmInference extends TaskRunner {
     const detokenizerOptions = new DetokenizerCalculatorOptions();
     // TODO: Remove this when we no longer need to have an even
     // number of responses in multi-output.
-    detokenizerOptions.setNumOutputHeads(roundUpToNearestEven(
-        this.options.getNumResponses()));
+    detokenizerOptions.setNumOutputHeads(
+      roundUpToNearestEven(this.options.getNumResponses()),
+    );
     // No need to set spm model, instead reuse TokenizerCalculator's side input.
     detokenizerOptions.addStopTokens('<eos>');
     detokenizerOptions.addStopTokens('<|endoftext|>');