From 6858f23da6d5f20ab68bcbbcbb42909c352e2ea0 Mon Sep 17 00:00:00 2001
From: sroy745 <142070531+sroy745@users.noreply.github.com>
Date: Sun, 6 Oct 2024 20:52:42 -0700
Subject: [PATCH] [BugFix][Core] Fix BlockManagerV2 when Encoder Input is None
 (#9103)

---
 vllm/core/block/block_table.py | 2 --
 vllm/core/block_manager_v2.py  | 4 +++-
 vllm/engine/arg_utils.py       | 5 -----
 3 files changed, 3 insertions(+), 8 deletions(-)

diff --git a/vllm/core/block/block_table.py b/vllm/core/block/block_table.py
index a9f4bd871dfda..d10cb29ef4a7c 100644
--- a/vllm/core/block/block_table.py
+++ b/vllm/core/block/block_table.py
@@ -220,7 +220,6 @@ def free(self) -> None:
         occupied by each block. After freeing all the blocks, the `_blocks` list
         is set to `None`.
         """
-        assert self._is_allocated
         for block in self.blocks:
             self._allocator.free(block)
         self._blocks.reset()
@@ -239,7 +238,6 @@ def physical_block_ids(self) -> List[int]:
             List[int]: A list of physical block indices for the blocks in the
                 BlockTable.
         """
-        assert self._is_allocated
         return self._blocks.ids()
 
     def get_unseen_token_ids(self, sequence_token_ids: List[int]) -> List[int]:
diff --git a/vllm/core/block_manager_v2.py b/vllm/core/block_manager_v2.py
index 0fad5fa99daf8..c7ee6609306d7 100644
--- a/vllm/core/block_manager_v2.py
+++ b/vllm/core/block_manager_v2.py
@@ -151,7 +151,9 @@ def _allocate_sequence(self, seq: Sequence) -> BlockTable:
             block_allocator=self.block_allocator,
             max_block_sliding_window=self.max_block_sliding_window,
         )
-        block_table.allocate(seq.get_token_ids())
+        if seq.get_token_ids():
+            # Add blocks to the block table only if the sequence is non empty.
+            block_table.allocate(seq.get_token_ids())
 
         return block_table
 
diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py
index 1623ebb3aa74c..cae95d20ca23d 100644
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -903,11 +903,6 @@ def create_engine_config(self) -> EngineConfig:
                     "--enable-prefix-caching is currently not "
                     "supported for multimodal models and has been disabled.")
             self.enable_prefix_caching = False
-        if model_config.is_encoder_decoder_model:
-            logger.warning(
-                "Block Manager v2 does not support encoder-decoder models"
-                " currently. Using Block Manager v1 as fallback.")
-            self.use_v2_block_manager = False
 
         cache_config = CacheConfig(
             block_size=self.block_size if self.device != "neuron" else