[Pooling] Honor max_embed_len for chunked embeddings (#55551)
Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
Taneem Ibrahim committed
Sep 7, 2026 at 08:31 UTC
ecd600d91e373d939ede33253a75d86bc5bf7338
1 file changed
+1
-1
vllm/entrypoints/pooling/base/protocol.py
+1
-1
@@ -179,7 +179,7 @@ class EmbeddingTokenizeParamsMixin(PoolingTokenizeParamsMixin):
179
pooler_config = model_config.pooler_config
180
if pooler_config is not None:
181
if pooler_config.enable_chunked_processing:
182
- max_total_tokens = None
182
+ max_total_tokens = pooler_config.max_embed_len
183
else:
184
max_embed_len = pooler_config.max_embed_len or default_max_total_tokens
185
max_output_tokens = default_max_total_tokens - max_embed_len