apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

cp_fix_bug (#5253)

Browse Source
This commit is contained in:
kevin
2025-11-27 15:15:49 +08:00
committed by GitHub
parent 3d74a4baf6
commit 69b4d058ad
2 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/config.py
View File

@@ -1572,7 +1572,7 @@ class FDConfig:
self.max_prefill_batch = int(os.getenv("MAX_PREFILL_NUM", "3"))
if current_platform.is_xpu():
self.max_prefill_batch = 1
if self.model_config is not None and self.model_config.enable_mm and not envs.ENABLE_V1_KVCACHE_SCHEDULER:
if self.model_config is not None and self.model_config.enable_mm:
self.max_prefill_batch = 1 # TODO:当前多模prefill阶段只支持并行度为1,待优化
else:
self.max_prefill_batch = self.scheduler_config.max_num_seqs

10
fastdeploy/engine/common_engine.py
View File

@@ -691,8 +691,16 @@ class EngineService:
else:
max_num_batched_tokens = self.cfg.model_config.max_model_len
# In multi-mode scenarios, using available_block_num to pull requests to prevent heavy rescheduling
# in the frequency domain due to insufficient blocks
if self.cfg.model_config.enable_mm:
self.resource_manager.check_and_free_block_tables()
available_blocks = self.resource_manager.available_block_num()
else:
available_blocks = self.cfg.cache_config.max_block_num_per_seq
tasks = self.scheduler.get_requests(
available_blocks=self.cfg.cache_config.max_block_num_per_seq,
available_blocks=available_blocks,
block_size=self.cfg.cache_config.block_size,
reserved_output_blocks=self.cfg.cache_config.enc_dec_block_num,
max_num_batched_tokens=max_num_batched_tokens,