biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CORE] Prompt Embeddings Support for v1 Engine (#24278)

Browse Source 
Signed-off-by: Andrew Sansom <andrew@protopia.ai>
Signed-off-by: Andrew Sansom <qthequartermasterman@gmail.com>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
This commit is contained in:
Andrew Sansom
2025-09-18 19:03:09 -05:00
committed by GitHub
parent 9fac6aa30b
commit 9a4600e4dc
20 changed files with 305 additions and 76 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
vllm/engine/arg_utils.py
View File

@@ -1513,12 +1513,6 @@ class EngineArgs:
recommend_to_remove=False)
return False
# No text embedding inputs so far.
if self.enable_prompt_embeds:
_raise_or_fallback(feature_name="--enable-prompt-embeds",
recommend_to_remove=False)
return False
# No Mamba or Encoder-Decoder so far.
if not model_config.is_v1_compatible:
_raise_or_fallback(feature_name=model_config.architectures,
@@ -1651,6 +1645,13 @@ class EngineArgs:
"models in V0 and has been disabled.")
self.enable_prefix_caching = False
if self.enable_prompt_embeds:
logger.warning(
"--enable-prompt-embeds and --enable-prefix-caching "
"are not supported together in V0. Prefix caching has "
"been disabled.")
self.enable_prefix_caching = False
# Set max_num_seqs to 256 for VLLM_V0.
if self.max_num_seqs is None:
self.max_num_seqs = 256
@@ -1664,6 +1665,17 @@ class EngineArgs:
# For pooling tasks the default is False
if model_config.runner_type != "pooling":
self.enable_chunked_prefill = True
# TODO: When prefix caching supports prompt embeds inputs, this
# check can be removed.
if (self.enable_prompt_embeds
and self.enable_prefix_caching is not False):
logger.warning(
"--enable-prompt-embeds and --enable-prefix-caching "
"are not supported together in V1. Prefix caching has "
"been disabled.")
self.enable_prefix_caching = False
if self.enable_prefix_caching is None:
self.enable_prefix_caching = True
else: