From ecd600d91e373d939ede33253a75d86bc5bf7338 Mon Sep 17 00:00:00 2001 From: Taneem Ibrahim Date: Mon, 7 Sep 2026 08:31:40 -0500 Subject: [PATCH] [Pooling] Honor max_embed_len for chunked embeddings (#55551) Signed-off-by: Taneem Ibrahim --- vllm/entrypoints/pooling/base/protocol.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/vllm/entrypoints/pooling/base/protocol.py b/vllm/entrypoints/pooling/base/protocol.py index 3de958fc1a..e30195a7b1 100644 --- a/vllm/entrypoints/pooling/base/protocol.py +++ b/vllm/entrypoints/pooling/base/protocol.py @@ -179,7 +179,7 @@ class EmbeddingTokenizeParamsMixin(PoolingTokenizeParamsMixin): pooler_config = model_config.pooler_config if pooler_config is not None: if pooler_config.enable_chunked_processing: - max_total_tokens = None + max_total_tokens = pooler_config.max_embed_len else: max_embed_len = pooler_config.max_embed_len or default_max_total_tokens max_output_tokens = default_max_total_tokens - max_embed_len