[Bugfix] Fix illegal memory access error with chunked prefill, prefix caching, block manager v2 and xformers enabled together (vllm-project#9532)

Signed-off-by: sasha0552 <[email protected]>

Author: sasha0552

tests/prefix_caching/test_prefix_caching.py

@@ -11,13 +11,21 @@
 from vllm.block import PhysicalTokenBlock
 from vllm.core.block_manager_v1 import CachedBlockAllocator
 from vllm.utils import Device
-
+from vllm import SamplingParams, TokensPrompt
 from ..models.utils import check_outputs_equal
 
 MODELS = [
     "facebook/opt-125m",
 ]
 
+UNSTABLE_PROMPT_SEQUENCE = [
+    ([0] * 588) + ([1] * 1332) + ([2] * 30) + ([3] * 1),
+    ([0] * 588) + ([1] * 1332) + ([4] * 3) + ([5] * 50),
+    ([0] * 588) + ([1] * 1332) + ([2] * 30) + ([6] * 95),
+    ([0] * 588) + ([1] * 1332) + ([4] * 3) + ([7] * 174),
+    ([0] * 588) + ([8] * 1539),
+]
+
 
 @pytest.fixture(scope="module", autouse=True)
 def check_deprecated_block_manager():
@@ -146,3 +154,22 @@ def test_mixed_requests(
         name_0="hf",
         name_1="vllm",
     )
+
+
+@pytest.mark.parametrize("backend", ["FLASH_ATTN", "FLASHINFER", "XFORMERS"])
+def test_unstable_prompt_sequence(
+    vllm_runner,
+    backend: str,
+    monkeypatch,
+) -> None:
+    override_backend_env_variable(monkeypatch, backend)
+
+    with vllm_runner(
+            "Qwen/Qwen2.5-0.5B-Instruct",
+            enable_chunked_prefill=True,
+            enable_prefix_caching=True,
+            max_model_len=4096,
+    ) as vllm_model:
+        for prompt in UNSTABLE_PROMPT_SEQUENCE:
+            vllm_model.generate(TokensPrompt(prompt_token_ids=prompt),
+                                SamplingParams(max_tokens=1))

vllm/attention/backends/utils.py

@@ -146,7 +146,6 @@ def _add_seq_group(
             chunked_prefill_enabled: bool):
         is_prompt = inter_data.is_prompt
         block_tables = inter_data.block_tables
-        computed_block_nums = inter_data.computed_block_nums
 
         for (seq_id, token_len, seq_len, curr_seq_len, query_len, context_len,
              curr_sliding_window_block) in zip(
@@ -172,10 +171,14 @@ def _add_seq_group(
             # NOTE: This only works for oooooooxxx style attention.
             block_table = []
             if inter_data.prefix_cache_hit:
-                block_table = computed_block_nums
+                block_table = block_tables[seq_id]
             elif ((chunked_prefill_enabled or not is_prompt)
                   and block_tables is not None):
-                block_table = block_tables[seq_id][-curr_sliding_window_block:]
+                if curr_sliding_window_block == 0:
+                    block_table = block_tables[seq_id]
+                else:
+                    block_table = block_tables[seq_id][
+                        -curr_sliding_window_block:]
             self.block_tables.append(block_table)
 
             # Compute slot mapping.