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

[Bugfix] Replace PoolingParams.normalize with use_activation (#32243)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-13 18:45:42 +08:00
committed by GitHub
parent eb28e8068d
commit 232214b2ae
21 changed files with 68 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/pooling/embed/openai_embedding_long_text/README.md
View File

@@ -47,7 +47,7 @@ The key parameters for chunked processing are in the `--pooler-config`:
```json
{
"pooling_type": "auto",
"normalize": true,
"use_activation": true,
"enable_chunked_processing": true,
"max_embed_len": 3072000
}

4
examples/pooling/embed/openai_embedding_long_text/client.py
View File

@@ -14,7 +14,7 @@ Prerequisites:
# MEAN pooling (processes all chunks, recommended for complete coverage)
vllm serve intfloat/multilingual-e5-large \
--pooler-config \
'{"pooling_type": "MEAN", "normalize": true, ' \
'{"pooling_type": "MEAN", "use_activation": true, ' \
'"enable_chunked_processing": true, "max_embed_len": 3072000}' \
--served-model-name multilingual-e5-large \
--trust-remote-code \
@@ -24,7 +24,7 @@ Prerequisites:
# OR CLS pooling (native CLS within chunks, MEAN aggregation across chunks)
vllm serve BAAI/bge-large-en-v1.5 \
--pooler-config \
'{"pooling_type": "CLS", "normalize": true, ' \
'{"pooling_type": "CLS", "use_activation": true, ' \
'"enable_chunked_processing": true, "max_embed_len": 1048576}' \
--served-model-name bge-large-en-v1.5 \
--trust-remote-code \

2
examples/pooling/embed/openai_embedding_long_text/service.sh
View File

@@ -96,7 +96,7 @@ echo ""
echo "🔧 Starting server with enhanced chunked processing configuration..."
# Build pooler config JSON
POOLER_CONFIG="{\"pooling_type\": \"$POOLING_TYPE\", \"normalize\": true, \"enable_chunked_processing\": ${VLLM_ENABLE_CHUNKED_PROCESSING}, \"max_embed_len\": ${MAX_EMBED_LEN}}"
POOLER_CONFIG="{\"pooling_type\": \"$POOLING_TYPE\", \"use_activation\": true, \"enable_chunked_processing\": ${VLLM_ENABLE_CHUNKED_PROCESSING}, \"max_embed_len\": ${MAX_EMBED_LEN}}"
# Start vLLM server with enhanced chunked processing
vllm serve "$MODEL_NAME" \