[Bug]

### Checklist

- [x] 1. I have searched related issues but cannot get the expected help.
- [ ] 2. The bug has not been fixed in the latest version.
- [ ] 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

4卡 nvidia A10 部署报错

[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][WARN][0914.14:00:25.876730][turbomind.cc:123] `max_context_token_num` is not set, default to 16384.
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807188][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807196][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807204][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][FATAL][0914.14:00:37.807214][cast.cu:264] CUDA error: no kernel image is available for execution on the device
*** stacktrace of thread 0x7f30d3fff640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f30d37fe640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f3066283640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f30d88fe640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
/root/startlmdeploy.sh: line 11: 7 Aborted (core dumped) lmdeploy serve api_server /root/hf_model/Qwen/Qwen3.8-27B-FP8 --model-name pkumlm_txt --backend turbomind --server-port 8000 --api-keys pkulesbrain_mlm_txt --model-format fp8 --reasoning-parser default --tool-call-parser qwen3coder --enable-prefix-caching --eager-mode --rope-scaling-factor 0.2 --session-len 16384 --log-level REQUEST --max-batch-size 2 --tp 4 --cache-max-entry-count 0.85

### Reproduction

### Environment

```Shell
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA A10 Off | 00000000:31:00.0 Off | 0 |
| 0% 76C P0 99W / 150W | 6931MiB / 23028MiB | 15% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA A10 Off | 00000000:4B:00.0 Off | 0 |
| 0% 67C P0 69W / 150W | 1469MiB / 23028MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 2 NVIDIA A10 Off | 00000000:98:00.0 Off | 0 |
| 0% 61C P0 64W / 150W | 1469MiB / 23028MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 3 NVIDIA A10 Off | 00000000:B1:00.0 Off | 0 |
| 0% 64C P0 67W / 150W | 1469MiB / 23028MiB | 7% Default |
| | | N/A |
```

### Error traceback

```Shell
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][WARN][0914.14:00:25.876730][turbomind.cc:123] `max_context_token_num` is not set, default to 16384.
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807188][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807196][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][WARN] Invalid TM_LOG_LEVEL='REQUEST'. Using default level.
[TM][FATAL][0914.14:00:37.807204][cast.cu:264] CUDA error: no kernel image is available for execution on the device
[TM][FATAL][0914.14:00:37.807214][cast.cu:264] CUDA error: no kernel image is available for execution on the device
*** stacktrace of thread 0x7f30d3fff640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f30d37fe640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f3066283640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
*** stacktrace of thread 0x7f30d88fe640 ***
[ 0] TM_CUDA_CHECK @ cast.cu:264
/root/startlmdeploy.sh: line 11: 7 Aborted (core dumped) lmdeploy serve api_server /root/hf_model/Qwen/Qwen3.8-27B-FP8 --model-name pkumlm_txt --backend turbomind --server-port 8000 --api-keys pkulesbrain_mlm_txt --model-format fp8 --reasoning-parser default --tool-call-parser qwen3coder --enable-prefix-caching --eager-mode --rope-scaling-factor 0.2 --session-len 16384 --log-level REQUEST --max-batch-size 2 --tp 4 --cache-max-entry-count 0.85
```