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

[V0 Deprecation] Remove pooling model support in V0 (#23434)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Co-authored-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Maximilien de Bayser
2025-08-29 04:04:02 -03:00
committed by GitHub
parent 934bebf192
commit 2554b27baa
38 changed files with 99 additions and 808 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/engine/arg_utils.py
View File

@@ -1566,8 +1566,7 @@ class EngineArgs:
use_spec_decode = self.speculative_config is not None
if (is_gpu and not use_sliding_window and not use_spec_decode
and not self.enable_lora
and model_config.runner_type != "pooling"):
and not self.enable_lora):
self.enable_chunked_prefill = True
logger.warning(
"Chunked prefill is enabled by default for models "
@@ -1585,10 +1584,6 @@ class EngineArgs:
"OOM during the initial memory profiling phase, or result "
"in low performance due to small KV cache size. Consider "
"setting --max-model-len to a smaller value.", max_model_len)
elif (self.enable_chunked_prefill
and model_config.runner_type == "pooling"):
msg = "Chunked prefill is not supported for pooling models"
raise ValueError(msg)
# if using prefix caching, we must set a hash algo
if self.enable_prefix_caching: