apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-31 11:56:44 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

[BugFix] Fix Configs (#2849)

Browse Source 
* fix config

* fix config
This commit is contained in:
YuanRisheng
2025-07-16 10:50:36 +08:00
committed by GitHub
parent 0fad10b35a
commit 101ad33332
10 changed files with 30 additions and 37 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
fastdeploy/spec_decode/mtp.py
View File

@@ -93,7 +93,7 @@ class MTPProposer(Proposer):
expected_decode_len: int):
"""Set dummy prefill inputs to model_inputs"""
max_dec_len = expected_decode_len + 1
self.num_gpu_blocks = self.parallel_config.max_block_num
self.num_gpu_blocks = self.parallel_config.total_block_num
self.initialize_kv_cache()
full_length = min(num_tokens // batch_size,
self.parallel_config.max_model_len - max_dec_len)
@@ -327,8 +327,8 @@ class MTPProposer(Proposer):
self.free_list = list(
range(
self.parallel_config.max_block_num - 1,
int(self.parallel_config.max_block_num *
self.parallel_config.total_block_num - 1,
int(self.parallel_config.total_block_num *
self.parallel_config.kv_cache_ratio) - 1,
-1,
))