Replace approximate formula with exact one for throughput (#251)

megatron/training.py

@@ -668,23 +668,23 @@ def add_to_logging(name):
         elapsed_time_per_iteration = elapsed_time / total_iterations
 
         seq_len = args.curriculum_seqlen if args.curriculum_learning else args.seq_length
+        hidden_size = args.hidden_size
+        num_layers = args.num_layers
+        vocab_size = args.padded_vocab_size
 
-        # throughput
+        # Compute throughput.
         samples_per_sec = batch_size / elapsed_time_per_iteration
         samples_per_sec_per_replica = samples_per_sec / args.data_parallel_size
         tokens_per_sec = samples_per_sec * seq_len
         tokens_per_sec_per_replica = tokens_per_sec / args.data_parallel_size
 
-        # general TFLOPs formula
-        # model_size_in_B * 4 * 2 * seqlen * global_batch_size / (time_in_sec_per_interation * total_gpus * 1e3)
-        #
+        # General TFLOPs formula (borrowed from Equation 3 in Section 5.1 of
+        # https://arxiv.org/pdf/2104.04473.pdf).
         # The factor of 4 is when used with activation check-pointing,
         # otherwise it will be 3, but for 200B model, activation check-pointing will always be on.
-        #
-        # here:
-        # model_size_in_B * 4 * 2 * seqlen * batch_size / (time_in_msec_per_interation * total_gpus * 1e3)
         checkpoint_activations_factor = 4 if args.checkpoint_activations else 3
-        tflops = args.parameters_in_billions_no_embedding * checkpoint_activations_factor * 2 * seq_len * batch_size / (elapsed_time_per_iteration * args.world_size * 1e3)
+        flops_per_iteration = (24 * checkpoint_activations_factor * batch_size * seq_len * num_layers * (hidden_size**2)) * (1. + (seq_len / (6. * hidden_size)) + (vocab_size / (16. * num_layers * hidden_size)))
+        tflops = flops_per_iteration / (elapsed_time_per_iteration * args.world_size * (10**12))
 
         # only the last rank process has a non-None _GLOBAL_TENSORBOARD_WRITER
         if writer and is_last_rank():