vllm - 💡(How to fix) Fix [Bug]: Qwen3.5 397B model occurs assertion error during allocating new blocks

Your current environment

<details> <summary>The output of <code>python collect_env.py</code></summary> </details> I find this problem in customer‘s code which is base on vllm 0.18.1.dev35+g36ed358b1.empty vllm_ascend 0.1.dev100+gd773192cd

It is hard to replicate, i keep the log and try to find what happen in the code. I have some questions: 1、Why mamba don’t need to consider the situation that num_required_blocks< len(req_blocks)? This situation is considered in SingleTypeKVCacheManager

<img width="1116" height="720" alt="Image" src="https://github.com/user-attachments/assets/98635ce2-4513-4add-9c00-25518b6f4cca" /> <img width="1002" height="869" alt="Image" src="https://github.com/user-attachments/assets/baedf84f-fb79-4cb1-a7ca-dd40d28725a9" />

🐛 Describe the bug

<img width="1702" height="771" alt="Image" src="https://github.com/user-attachments/assets/5d0eae18-f6c3-4e8b-8cf6-2b131ac857df" /> KeyId: [None]. ERROR 2026-05-14 05:25:28.876 [core.py:1123] EngineCore encountered a fatal error. ERROR 2026-05-14 05:25:28.876 [core.py:1123] Traceback (most recent call last): ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/engine/core.py", line 1114, in run_engine_core ERROR 2026-05-14 05:25:28.876 [core.py:1123] engine_core.run_busy_loop() ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/engine/core.py", line 1155, in run_busy_loop ERROR 2026-05-14 05:25:28.876 [core.py:1123] self._process_engine_step() ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/engine/core.py", line 1194, in _process_engine_step ERROR 2026-05-14 05:25:28.876 [core.py:1123] outputs, model_executed = self.step_fn() ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^^^^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/engine/core.py", line 463, in step_with_batch_queue ERROR 2026-05-14 05:25:28.876 [core.py:1123] scheduler_output = self.scheduler.schedule() ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^^^^^^^^^^^^^^^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/core/sched/scheduler.py", line 457, in schedule ERROR 2026-05-14 05:25:28.876 [core.py:1123] new_blocks = self.kv_cache_manager.allocate_slots( ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/core/kv_cache_manager.py", line 365, in allocate_slots ERROR 2026-05-14 05:25:28.876 [core.py:1123] new_blocks = self.coordinator.allocate_new_blocks( ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/core/kv_cache_coordinator.py", line 167, in allocate_new_blocks ERROR 2026-05-14 05:25:28.876 [core.py:1123] return tuple( ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/core/kv_cache_coordinator.py", line 168, in <genexpr> ERROR 2026-05-14 05:25:28.876 [core.py:1123] manager.allocate_new_blocks( ERROR 2026-05-14 05:25:28.876 [core.py:1123] File "/usr/local/python3.11.14/lib/python3.11/site-packages/vllm/v1/core/single_type_kv_cache_manager.py", line 955, in allocate_new_blocks ERROR 2026-05-14 05:25:28.876 [core.py:1123] assert num_required_blocks > len(req_blocks), ( ERROR 2026-05-14 05:25:28.876 [core.py:1123] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ERROR 2026-05-14 05:25:28.876 [core.py:1123] AssertionError: num_required_blocks 4 < len(req_blocks) 5 ERROR 2026-05-14 05:25:28.880 [async_llm.py:818] AsyncLLM output_handler failed.

attach the pip list、env and full service log assert-problem.zip

The start command: - name: MODEL_PATH value: /home/admin/model/ - name: PORT value: '8000' - name: TASK_QUEUE_ENABLE value: '1' - name: VLLM_USE_V1 value: '1' - name: VLLM_VERSION value: '0.18.0' - name: OTEL_EXPORTER_OTLP_TRACES_PROTOCOL value: 'http/protobuf' - name: VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS value: '380' - name: PROMETHEUS_MULTIPROC_DIR value: /tmp/ - name: ASCEND_PROCESS_LOG_PATH value: /home/admin/logs/ascend/ - name: VLLM_LOGGING_CONFIG_PATH value: /home/admin/vllm/production_logging_config.json - name: VLLM_ASCEND_ENABLE_NZ value: '1' - name: HCCL_OP_EXPANSION_MODE value: 'AIV' - name: PYTORCH_NPU_ALLOC_CONF value: 'expandable_segments:True' - name: HCCL_BUFFSIZE value: '1024' - name: VLLM_IMAGE_FETCH_TIMEOUT value: '30' - name: VLLM_VIDEO_FETCH_TIMEOUT value: '60' - name: RFORK_SCHEDULER_URL value: 'http://rfork-planner.rfork-planner.svc:1223' - name: OMP_NUM_THREADS value: '1' - name: ENGINE_EXTRA_ARGS value: > --trust-remote-code --served-model-name auto --distributed-executor-backend mp --model-loader-extra-config {"enable_multithread_load":true,"num_threads":8} --enable-log-requests --enable-prefix-caching --mamba-cache-mode align --enable-prompt-tokens-details --reasoning-padding <think> --otlp-traces-endpoint https://antcollector.alipay.com/namespace/aicloud/task/otlptrace/otlp/api/v1/traces --token-level-profiling --data-parallel-size 1 --tensor-parallel-size 16 --max-model-len 262144 --max-num-batched-tokens 16384 --max-num-seqs 128 --gpu-memory-utilization 0.90 --compilation-config {"cudagraph_capture_sizes":[4,16,32,64,96,128,160,192,224,256,288,320,352,384,416,448,480,512],"cudagraph_mode":"FULL_DECODE_ONLY"} --speculative-config {"method":"qwen3_5_mtp","num_speculative_tokens":3,"enforce_eager":true} --trust-remote-code --async-scheduling --quantization ascend --mm-processor-cache-gb 4 --mm-processor-cache-type shm --load-format rfork --enable-auto-tool-choice --tool-call-parser qwen3_coder --enable-expert-parallel --additional-config {"enable_cpu_binding":true,"multistream_overlap_shared_expert":false}

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.