
----------------------------------------------Summary-----------------------------------------------
batch_size_per_gpu: 1
seq_len: 512
num_tokens_to_generate: 10
use_kv_cache: True
kv_cache_memory_per_gpu: 427.62 MB
kv_cache_latency: 550.0 us
layernorm_dtype_bytes: 2
embedding_memory_per_gpu: 514.79 MB
weight_memory_per_gpu: 25.68 GB
prefill_max_batch_size_per_gpu: 2
prefill_activation_memory_per_gpu: 5.03 GB
prefill_num_flops_fwd_total: 13.36 T
decode_max_batch_size_per_gpu: 32
decode_activation_memory_per_gpu: 6.56 MB
decode_num_flops_fwd_total: 25.68 G
prefill_latency: 43.7 ms
prefill_latency_fwd_attn: 14.45 ms
prefill_latency_fwd_mlp: 27.53 ms
prefill_latency_fwd_layernorm: 539.46 us
prefill_latency_fwd_tp_comm: 0.0 us
prefill_latency_fwd_input_embedding: 331.05 us
prefill_latency_fwd_output_embedding_loss: 844.78 us
decode_latency: 17.07 ms
decode_latency_fwd_attn: 5.4 ms
decode_latency_fwd_mlp: 10.79 ms
decode_latency_fwd_layernorm: 1.05 us
decode_latency_fwd_tp_comm: 0.0 us
decode_latency_fwd_input_embedding: 331.05 us
decode_latency_fwd_output_embedding_loss: 1.65 us
total_decode_latency: 170.71 ms
total_latency: 214.41 ms
total_per_token_latency: 21.44 ms
----------------------------------------------------------------------------------------------------
