Describe the current behavior / 问题描述 (Mandatory / 必填)

llama2_7b网络在910B环境走KBK流程，性能达不到0.5的lite性能
模型仓地址：https://gitee.com/mindspore/mindformers/blob/dev/docs/model_cards/llama2.md

Environment / 环境信息 (Mandatory / 必填)

• Hardware Environment(Ascend/GPU/CPU) / 硬件环境:

Please delete the backend not involved / 请删除不涉及的后端:
/device ascend

• Software Environment / 软件环境 (Mandatory / 必填):
  -- MindSpore version (e.g., 1.7.0.Bxxx) :
  -- Python version (e.g., Python 3.7.5) :
  -- OS platform and distribution (e.g., Linux Ubuntu 16.04):
  -- GCC/Compiler version (if compiled from source):

【CANN版本】：Milan_C17/20240414
【MindSpore版本】：master_B020
【MindFormers版本】：master_B020

• Excute Mode / 执行模式 (Mandatory / 必填)(PyNative/Graph):

Please delete the mode not involved / 请删除不涉及的模式:
/mode pynative
/mode graph

Related testcase / 关联用例 (Mandatory / 必填)

用例仓地址：MindFormers_Test/cases/llama2/7b/infer/
用例：
test_mf_llama2_7b_kbk_infer_alpaca_batch_incremental_1p_0001

Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)

1. get code from mindformers
2. cd mindformers
3. export GRAPH_OP_RUN=1;export MS_ENABLE_INTERNAL_KERNELS=on;export ASCEND_HOME_PATH=/usr/local/Ascend/latest
   python llama2_generate.py --model_type llama2_7b "
   " --yaml_file {1} "
   "--use_past True --device_id={2} --checkpoint_path
   {3}/llama2/eval/llama2_7b.ckpt "
   "--batch_size {4}
4. 验证网络性能是否可以达到0.5的lite

Describe the expected behavior / 预期结果 (Mandatory / 必填)

网络性能可以达到0.5的lite

Related log / screenshot / 日志 / 截图 (Mandatory / 必填)

2024-04-26 01:29:13,241 - mindformers[mindformers/generation/text_generator.py:881] - INFO - total time: 59.58512878417969 s; generated tokens: 2010 tokens; generate speed: 33.73324923539765 tokens/s
2024-04-26 01:29:37,382 - mindformers[mindformers/generation/text_generator.py:881] - INFO - total time: 24.015976428985596 s; generated tokens: 2010 tokens; generate speed: 83.69428600762913 tokens/s
2024-04-26 01:30:01,670 - mindformers[mindformers/generation/text_generator.py:881] - INFO - total time: 24.162757873535156 s; generated tokens: 2010 tokens; generate speed: 83.18586853868618 tokens/s

Special notes for this issue/备注 (Optional / 选填)

走给谭纬城