fix get_save_output op and refactor specu_decoding

[Wanglongzhi2001]

PR types

Bug fixes

PR changes

Others

Description

1. 修复 save_output 和 get_output
2. 给投机解码添加 token_penalty
3. 调整投机解码架构
4. 修复 ngram_match 算子的 import 位置给其他暂时不支持投机解码的硬件带来的 import 问题

[paddle-bot]

Thanks for your contribution!

[codecov]

Codecov Report

Attention: Patch coverage is 4.25532% with 45 lines in your changes missing coverage. Please review.

Project coverage is 53.05%. Comparing base (8ed579a) to head (04306b0).
Report is 15 commits behind head on develop.

Files with missing lines	Patch %	Lines
...enlp/experimental/transformers/generation_utils.py	0.00%	37 Missing ⚠️
paddlenlp/trl/llm_utils.py	0.00%	5 Missing ⚠️
paddlenlp/transformers/model_utils.py	0.00%	2 Missing ⚠️
paddlenlp/experimental/transformers/proposers.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #9576      +/-   ##
===========================================
- Coverage    53.07%   53.05%   -0.03%     
===========================================
  Files          703      703              
  Lines       110801   110815      +14     
===========================================
- Hits         58804    58789      -15     
- Misses       51997    52026      +29     

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

[yuanlehome]

LGTM