xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 00:15:51 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Doc] Fix batch-level DP example (#23325)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
Co-authored-by: youkaichao <youkaichao@gmail.com>
This commit is contained in:
Cyrus Leung 2025-08-21 21:16:38 +08:00 committed by GitHub
parent 0c6e40bbaa
commit 5cc54f7c5b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
docs/configuration/optimization.md
View File

@ -153,13 +153,14 @@ from vllm import LLM
llm = LLM( llm = LLM(
model="Qwen/Qwen2.5-VL-72B-Instruct", model="Qwen/Qwen2.5-VL-72B-Instruct",
# Create two EngineCore instances, one per DP rank
data_parallel_size=2,
# Within each EngineCore instance:
# The vision encoder uses TP=4 (not DP=2) to shard the input data
# The language decoder uses TP=4 to shard the weights as usual
tensor_parallel_size=4, tensor_parallel_size=4,
# When mm_encoder_tp_mode="data",
# the vision encoder uses TP=4 (not DP=1) to shard the input data,
# so the TP size becomes the effective DP size.
# Note that this is independent of the DP size for language decoder which is used in expert parallel setting.
mm_encoder_tp_mode="data", mm_encoder_tp_mode="data",
# The language decoder uses TP=4 to shard the weights regardless
# of the setting of mm_encoder_tp_mode
) )
``` ```