apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[Metax] fix shape error & output garbled code when reasoning big picture or video (#5965)

Browse Source 
Co-authored-by: root <root@lt-wks-10-0-180-15.pub.metax-tech.com>
This commit is contained in:
MingkunZhang
2026-01-09 13:41:45 +08:00
committed by GitHub
parent 2d2b156252
commit cb09b52e66
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/engine/args_utils.py
+5 -2
View File
@@ -1374,7 +1374,7 @@ class EngineArgs:
speculative_cfg = self.create_speculative_config()
if not self.enable_chunked_prefill:
if current_platform.is_cuda() and self.splitwise_role == "mixed":
if (current_platform.is_cuda() or current_platform.is_maca()) and self.splitwise_role == "mixed":
# default enable chunked prefill
self.enable_chunked_prefill = True
@@ -1384,7 +1384,10 @@ class EngineArgs:
if self.max_num_batched_tokens is None:
if int(envs.ENABLE_V1_KVCACHE_SCHEDULER):
self.max_num_batched_tokens = 8192 # if set to max_model_len, it's easy to be OOM
if current_platform.is_maca():
self.max_num_batched_tokens = self.max_model_len
else:
self.max_num_batched_tokens = 8192 # if set to max_model_len, it's easy to be OOM
else:
if self.enable_chunked_prefill:
self.max_num_batched_tokens = 2048