[Bug] Slower decode speed in v0.17 when compared to v0.14

### Checklist

- [x] 1. I have searched related issues but cannot get the expected help.
- [x] 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

We have tested Qwen3.8-27B model with v0.14 and also in v0.17. Keeping the input, code and all the configs unchanged, we tested the decode speed in v0.14 and in v0.17 in 2 different environments with corresponding deps

Some basic configs are as follows:

```javascript
{
"gpu_memory_utilization": 0.9,
"max_model_len": 32768,
"max_prefill_token_num": 0,
"enable_prefix_caching": true,
"profile_prefill": false,
"dtype": "bfloat16",
"quant_policy": 0,
"engine": "turbomind"
}
```

Logs from V0.17:
```javascript
{
"n_items": 1,
"wall_s": 89.438,
"prefill_s": null,
"decode_s": null,
"input_tokens": 858,
"output_tokens": 4096,
"output_tokens_mean": 4096.0,
"output_tokens_min": 4096,
"output_tokens_max": 4096,
"max_new_tokens": 4096,
"truncated": 1,
"tps": 45.8,
"tps_per_seq": 45.8,
"decode_tps": null,
"per_item": [
{
"wall_s": null,
"input_tokens": 858,
"output_tokens": 4096,
"tps": null,
"truncated": 1
}
]
},
"inference_time_s": 89.67
}
```

Logs from v0.14:
```javascript
{
"n_items": 1,
"wall_s": 65.819,
"prefill_s": null,
"decode_s": null,
"input_tokens": 858,
"output_tokens": 4096,
"output_tokens_mean": 4096.0,
"output_tokens_min": 4096,
"output_tokens_max": 4096,
"max_new_tokens": 4096,
"truncated": 1,
"tps": 62.2,
"tps_per_seq": 62.2,
"decode_tps": null,
"per_item": [
{
"wall_s": null,
"input_tokens": 858,
"output_tokens": 4096,
"tps": null,
"truncated": 1
}
]
},
"inference_time_s": 66.03
}```

This was tested in H200 Single GPU. We haven't used server mode for the above testing. Let me know if we are missing any configs

### Reproduction

Run as library

### Environment

```Shell
**For v0.17:**

sys.platform: linux
Python: 3.12.14 | packaged by Anaconda, Inc. | (main, Aug 27 2026, 14:46:43) [GCC 14.3.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0: NVIDIA H200
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.6, V12.6.77
GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.12.1+cu130
PyTorch compiling details: PyTorch built with:
- GCC 13.3
- C++ Version: 202002
- Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications
- Intel(R) MKL-DNN v3.11.2 (Git Hash 03c022d3ffdcee958cfacbe720048e725fdf644c)
- OpenMP 201511 (a.k.a. OpenMP 4.5)
- LAPACK is enabled (usually provided by MKL)
- NNPACK is enabled
- CPU capability usage: AVX512
- CUDA Runtime 13.0
- NVCC architecture flags: -gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120
- CuDNN 92.0 (built against CUDA 13.2)
- Magma 2.6.1
- Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=7269437d655783a26cba32aa88195b741ff496aa, CUDA_FLAGS= -DLIBCUDACXX_ENABLE_SIMPLIFIED_COMPLEX_OPERATIONS -Xfatbin -compress-all -DONNX_NAMESPACE=onnx_torch -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90 -gencode arch=compute_100,code=sm_100 -gencode arch=compute_120,code=sm_120 -Xcudafe --diag_suppress=cc_clobber_ignored,--diag_suppress=field_without_dll_interface,--diag_suppress=base_class_has_different_dll_interface,--diag_suppress=dll_interface_conflict_none_assumed,--diag_suppress=dll_interface_conflict_dllexport_assumed,--diag_suppress=bad_friend_decl --expt-relaxed-constexpr --expt-extended-lambda -Xfatbin -compress-all --threads 2 -compress-mode=size -Wno-deprecated-gpu-targets --expt-extended-lambda -DCUB_WRAPPED_NAMESPACE=at_cuda_detail -DDISABLE_CUSPARSE_DEPRECATED -DCUDA_HAS_FP16=1 -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -DC10_NODEPRECATED, CUDA_VERSION=13.0, CUDNN_VERSION=9.20.0, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_MSLK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.12.1, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF,

TorchVision: 0.27.1+cu130
LMDeploy: 0.17.0+a84fb59
transformers: 5.16.1
fastapi: 0.141.1
pydantic: 2.13.5
triton: 3.7.1
NVIDIA Topology:
GPU0 NIC0 NIC1 NIC2 NIC3 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X SYS SYS NODE NODE 169,171,173 1 N/A
NIC0 SYS X PIX SYS SYS
NIC1 SYS PIX X SYS SYS
NIC2 NODE SYS SYS X PIX
NIC3 NODE SYS SYS PIX X

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

NIC Legend:

NIC0: mlx5_1
NIC1: mlx5_2
NIC2: mlx5_7
NIC3: mlx5_8

---------------------------------------------------
**For v0.14**

sys.platform: linux
Python: 3.11.16 (main, Aug 27 2026, 14:44:21) [GCC 14.3.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0: NVIDIA H200
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.6, V12.6.77
GCC: gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
PyTorch: 2.10.0+cu128
PyTorch compiling details: PyTorch built with:
- GCC 13.3
- C++ Version: 201703
- Intel(R) oneAPI Math Kernel Library Version 2024.2-Product Build 20240605 for Intel(R) 64 architecture applications
- Intel(R) MKL-DNN v3.7.1 (Git Hash 8d263e693366ef8db40acc569cc7d8edf644556d)
- OpenMP 201511 (a.k.a. OpenMP 4.5)
- LAPACK is enabled (usually provided by MKL)
- NNPACK is enabled
- CPU capability usage: AVX512
- CUDA Runtime 12.8
- NVCC architecture flags: -gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86;-gencode;arch=compute_90,code=sm_90;-gencode;arch=compute_100,code=sm_100;-gencode;arch=compute_120,code=sm_120
- CuDNN 91.0.2 (built against CUDA 12.9)
- Magma 2.6.1
- Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, COMMIT_SHA=449b1768410104d3ed79d3bcfe4ba1d65c7f22c0, CUDA_VERSION=12.8, CUDNN_VERSION=9.10.2, CXX_COMPILER=/opt/rh/gcc-toolset-13/root/usr/bin/c++, CXX_FLAGS= -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -DUSE_FBGEMM_GENAI -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -DC10_NODEPRECATED -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=range-loop-construct -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-unknown-pragmas -Wno-unused-parameter -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=old-style-cast -faligned-new -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-dangling-reference -Wno-error=dangling-reference -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, TORCH_VERSION=2.10.0, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_GLOO=ON, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=1, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF, USE_ROCM_KERNEL_ASSERT=OFF, USE_XCCL=OFF, USE_XPU=OFF,

TorchVision: 0.25.0+cu128
LMDeploy: 0.14.0+a84fb59
transformers: 5.14.1
fastapi: 0.141.1
pydantic: 2.13.5
triton: 3.6.0
NVIDIA Topology:
GPU0 NIC0 NIC1 NIC2 NIC3 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X SYS SYS NODE NODE 169,171,173 1 N/A
NIC0 SYS X PIX SYS SYS
NIC1 SYS PIX X SYS SYS
NIC2 NODE SYS SYS X PIX
NIC3 NODE SYS SYS PIX X

Legend:

X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks

NIC Legend:

NIC0: mlx5_1
NIC1: mlx5_2
NIC2: mlx5_7
NIC3: mlx5_8
```

### Error traceback

```Shell

```