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 8, 22:45 UTC
[Bugfix] Fix `normalize` still being passed to `PoolerConfig` (#33794)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Feb 4, 2026 at 23:56 UTC
80f921ba4bab2ea251d149305ea0f912c6fc218a
2 files changed
+3
-3
tests/models/language/pooling/test_embedding.py
+1
-1
@@ -54,7 +54,7 @@ def test_models(
54
vllm_extra_kwargs = {}
55
if model == "ssmits/Qwen2-7B-Instruct-embed-base":
56
vllm_extra_kwargs["pooler_config"] = PoolerConfig(
57
- seq_pooling_type="MEAN", normalize=False
57
+ seq_pooling_type="MEAN", use_activation=False
58
)
59
60
max_model_len: int | None = 512
vllm/entrypoints/llm.py
+2
-2
@@ -174,8 +174,8 @@ class LLM:
174
multi-modal processor obtained from `AutoProcessor.from_pretrained`.
175
The available overrides depend on the model that is being run.
176
For example, for Phi-3-Vision: `{"num_crops": 4}`.
177
- pooler_config: Initialize non-default pooling config for the pooling
178
- model. e.g. `PoolerConfig(seq_pooling_type="MEAN", normalize=False)`.
177
+ pooler_config: Initialize non-default pooling config for the pooling model,
178
+ e.g., `PoolerConfig(seq_pooling_type="MEAN", use_activation=False)`.
179
compilation_config: Either an integer or a dictionary. If it is an
180
integer, it is used as the mode of compilation optimization. If it
181
is a dictionary, it can specify the full compilation configuration.