add training speed indicator (#2936)

Co-authored-by: llbdyiu66 <llbdyiu66@users.noreply.github.com>

Author: llbdyiu66

paddleformers/cli/train/sft/workflow.py

@@ -411,11 +411,16 @@ def neft_post_hook(module, input, output):
         if model_args.neftune:
             neft_post_hook_handle.remove()
         if training_args.benchmark:
-            total_effective_tokens = (
-                sum([len(i["input_ids"]) for i in trainer.train_dataset]) * train_result.metrics["progress_or_epoch"]
+            total_tokens = (
+                data_args.max_seq_len
+                * training_args.dataset_world_size
+                * training_args.gradient_accumulation_steps
+                * training_args.max_steps
+            )
+            total_tokens_per_second_per_gpu = (
+                total_tokens / train_result.metrics["train_runtime"] / training_args.world_size
             )
-            effective_tokens_per_second = total_effective_tokens / train_result.metrics["train_runtime"]
-            logger.info(f"Effective_Tokens_per_second: {effective_tokens_per_second} ")
+            logger.info(f"Total_Tokens_per_second_per_gpu: {total_tokens_per_second_per_gpu} ")
             logger.info("Benchmark done.")
         else:
             if not training_args.autotuner_benchmark: