apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 09:07:10 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[Feature][MTP]Support new mtp (#3656)

Browse Source 
* update multi-draft-token strategy

* fix format

* support hybrid mtp with ngram speculative decoding method
This commit is contained in:
freeliuzc
2025-08-27 19:38:26 +08:00
committed by GitHub
parent 62659a7a73
commit c753f1fc9e
20 changed files with 501 additions and 579 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
fastdeploy/spec_decode/base.py
View File

@@ -45,6 +45,10 @@ class Proposer(ABC):
self.max_model_len = self.parallel_config.max_model_len
self.speculative_method = self.speculative_config.method
self.max_draft_token_num = self.speculative_config.num_speculative_tokens
self.num_model_steps = self.speculative_config.num_model_steps
self.max_ngram_size = self.speculative_config.max_ngram_size
self.min_ngram_size = self.speculative_config.min_ngram_size
spec_logger.info(f"Speculate config: {self.speculative_config}")