bugfix: correct the calculation of num_prompt_tokens_per_request and num_generated_tokens_per_request metrics.

Author: XuZhang99

xllm/core/framework/batch/batch_factory.cpp

@@ -69,8 +69,10 @@ std::vector<Batch> BatchFactory::create_batches(
   COUNTER_ADD(num_processing_tokens_total_prompt, num_prompt_tokens);
   COUNTER_ADD(num_processing_tokens_total_generated, num_generated_tokens);
 
-  HISTOGRAM_OBSERVE(num_prompt_tokens_per_request, num_prompt_tokens);
-  HISTOGRAM_OBSERVE(num_generated_tokens_per_request, num_generated_tokens);
+  HISTOGRAM_OBSERVE(num_prompt_tokens_per_request,
+                    num_prompt_tokens / running_sequences.size());
+  HISTOGRAM_OBSERVE(num_generated_tokens_per_request,
+                    num_generated_tokens / running_sequences.size());
 
   return batches;
 }