[CacheKV] Add abstract Cache class.

[lvdongyi]

PR types

New features

PR changes

APIs

Description

1. Add abstract Cache class and DynamicCache, let llama use DynamicCache

[paddle-bot]

Thanks for your contribution!

[lvdongyi]

@DrownFish19 这块我显存不足，本地无法测试。

[codecov]

Codecov Report

Attention: Patch coverage is 80.99174% with 23 lines in your changes missing coverage. Please review.

Project coverage is 52.98%. Comparing base (b5e3f0c) to head (35ea106).
Report is 2 commits behind head on develop.

❗ Current head 35ea106 differs from pull request most recent head dd5c52d

Please upload reports for the commit dd5c52d to get more accurate results.

Files with missing lines	Patch %	Lines
paddlenlp/transformers/cache_utils.py	82.69%	9 Missing ⚠️
paddlenlp/generation/utils.py	75.00%	7 Missing ⚠️
paddlenlp/transformers/llama/modeling.py	82.05%	7 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #9401      +/-   ##
===========================================
- Coverage    53.01%   52.98%   -0.03%     
===========================================
  Files          678      677       -1     
  Lines       108787   107926     -861     
===========================================
- Hits         57668    57184     -484     
+ Misses       51119    50742     -377     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.