Fix wrong token latency when batch size is greater than 1 #4708

Job	Run time
cpp-beam_search_causal_lm-notus-7b-v1	14m 36s
cpp-beam_search_causal_lm-Phi-2	16m 5s
cpp-beam_search_causal_lm-Qwen1_5-7B-Chat	33m 14s
cpp-beam_search_causal_lm-ubuntu (./build/samples/cpp/beam_search_causal_lm/beam_search_causal_lm)	17m 19s
cpp-chat_sample-ubuntu	14m 19s
cpp-beam_search_causal_lm-ubuntu (python ./samples/python/beam_search_causal_lm/beam_search_causa...	18m 26s
cpp-continuous-batching-macos	19m 16s
cpp-continuous-batching-ubuntu	15m 1s
cpp-continuous-batching-windows	25m 18s
cpp-greedy_causal_lm-Qwen-7B-Chat	10m 18s
cpp-greedy_causal_lm-redpajama-3b-chat	15m 51s
cpp-greedy_causal_lm-windows	24m 30s
cpp-Phi-1_5	7m 42s
cpp-multinomial-greedy_causal_lm-ubuntu	13m 22s
cpp-prompt_lookup_decoding_lm-ubuntu	14m 57s
cpp-speculative_decoding_lm-ubuntu	12m 13s
visual_language_chat_sample-ubuntu-internvl2	14m 8s
visual_language_chat_sample-ubuntu-minicpm_v2_6	7m 56s
visual_language_chat_sample-ubuntu-llava_1_5 / visual_language_chat_sample-ubuntu-llava	31m 22s
visual_language_chat_sample-ubuntu-llava_next / visual_language_chat_sample-ubuntu-llava	18m 30s
ci/gha_overall_status_causal_lm	1s
	5h 44m 24s

Provide feedback