A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 18:30 UTC
[V1] Default MLA to V1 (#14921)
Signed-off-by: simon-mo <simon.mo@hey.com>
Simon Mo committed
Mar 17, 2025 at 06:54 UTC
89fca671fbb579d0840818afaece575c71b0b892
1 file changed
+1
-5
vllm/engine/arg_utils.py
+1
-5
@@ -1191,7 +1191,7 @@ class EngineArgs:
1191
NOTE: for autoselection of V0 vs V1 engine, we need to
1192
create the ModelConfig first, since ModelConfig's attrs
1193
(e.g. the model arch) are needed to make the decision.
1194
-
1194
+
1195
This function set VLLM_USE_V1=X if VLLM_USE_V1 is
1196
unspecified by the user.
1197
@@ -1576,10 +1576,6 @@ class EngineArgs:
1576
#############################################################
1577
# Experimental Features - allow users to opt in.
1578
1579
- # MLA is is supported on V1, but off by default for now.
1580
- if model_config.use_mla and _warn_or_fallback("MLA"):
1581
- return False
1582
-
1579
# LoRA is supported on V1, but off by default for now.
1580
if self.enable_lora and _warn_or_fallback("LORA"):
1581
return False