[Bug] v0.12.2-cu12.8部署qwen3.5的时候请，求任何内容出现 illegal memory

### Checklist

- [ ] 1. I have searched related issues but cannot get the expected help.
- [x] 2. The bug has not been fixed in the latest version.
- [x] 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.

### Describe the bug

latest-cu128 在单卡2080TI（ 22G显存 ） + 8卡成功部署了Qwen3.5-35B-A3B。但是用请求测试的时候出现illegal memory 错误

### Reproduction

docker run -d \
--runtime nvidia \
--gpus '"device=0,1,2,3,4,5,6,7"'\
--name lmdeploy_ht \
-v /data/model/hugginface:/models \
-v ~/.cache/huggingface:/root/.cache/huggingface \
-p 23333:23333 \
--ipc=host \
--shm-size=128g \
docker.1ms.run/openmmlab/lmdeploy:v0.12.2-cu12.8 \
tail -f /dev/null
docker exec -it lmdeploy_ht bash

lmdeploy serve api_server /models/Qwen/Qwen3.5-35B-A3B \
--tp 8 \
--cache-max-entry-count 0.8 \
--log-level INFO \
--max-concurrent-requests 4 \
--model-name pkumlm_txt \
--backend turbomind\
--max_batch_size 64 \
--cache-block-seq-len 32

### Environment

```Shell
v0.12.2-cu12.8
单卡2080TI（ 22G显存 ） * 8卡
Qwen3.5-35B-A3B
```

### Error traceback

```Shell
curl -H "Accept: application/json" -H "Content-type: application/json" -X POST -d '{"model": "pkumlm_txt","messages":[{"role":"user","content":"你是谁"} ]}' http://192.168.1.36:23333/v1/chat/completions

报错
2026-03-19 12:52:50,954 - lmdeploy - INFO - turbomind.py:687 - [async_stream_infer] session 1 start
[TM][INFO] [SeqMgr][Create] ID 1
[TM][WARNING] [ProcessInferRequests] [1] total sequence length (11 + 262133) exceeds `session_len` (258368), `max_new_tokens` is truncated to 258357
invalid argument
```