[Misc] Simplify max tokens in multimodal registry (#27500)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-10-25 14:56:01 +08:00
parent b853540388
commit 4c5f632165
4 changed files with 17 additions and 38 deletions
--- a/vllm/v1/core/encoder_cache_manager.py
+++ b/vllm/v1/core/encoder_cache_manager.py
@@ -264,8 +264,8 @@ def compute_encoder_budget(
            from the input sequence.
    """
    if mm_registry.supports_multimodal_inputs(model_config):
-        max_tokens_by_modality = (
-            mm_registry.get_max_tokens_per_item_by_nonzero_modality(model_config)
+        max_tokens_by_modality = mm_registry.get_max_tokens_per_item_by_modality(
+            model_config
        )

        return compute_mm_encoder_budget(