[Bug] Check failed: data_type_v<T> == dtype_ (f16 vs. bf16)

[zwx1225059]

Checklist

• 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

gpt-oss-120b cannot be deployed using version 0.14.0a1. Other models can be deployed normally, and other versions, such as 0.13.0, are normal. The following is the error log.
[TM][WARN][0616.14:35:12.059571][SequenceManager.cc:145] prefix caching is enabled
[TM][WARN][0616.14:35:12.067487][SequenceManager.cc:145] prefix caching is enabled
[TM][WARN][0616.14:35:12.071696][SequenceManager.cc:145] prefix caching is enabled
[TM][WARN][0616.14:35:12.074386][anomaly_handler.cu:141] level: 2
[TM][WARN][0616.14:35:12.074417][anomaly_handler.cu:163] +INF -> 65504
[TM][WARN][0616.14:35:12.074423][anomaly_handler.cu:164] -INF -> -65504
[TM][WARN][0616.14:35:12.074428][anomaly_handler.cu:170] NaN -> 0
[TM][WARN][0616.14:35:12.079330][anomaly_handler.cu:200] fallback: 0
[TM][WARN][0616.14:35:12.079376][anomaly_handler.cu:206] max_batch_size: 1000
[TM][WARN][0616.14:35:12.079393][anomaly_handler.cu:207] vocab_size: 1000
[TM][WARN][0616.14:35:12.079377][anomaly_handler.cu:200] fallback: 0
[TM][WARN][0616.14:35:12.079421][anomaly_handler.cu:206] max_batch_size: 1000
[TM][WARN][0616.14:35:12.079429][anomaly_handler.cu:207] vocab_size: 1000
[TM][WARN][0616.14:35:12.081156][anomaly_handler.cu:200] fallback: 0
[TM][WARN][0616.14:35:12.081188][anomaly_handler.cu:206] max_batch_size: 1000
[TM][WARN][0616.14:35:12.081194][anomaly_handler.cu:207] vocab_size: 1000
[TM][WARN][0616.14:35:12.081232][anomaly_handler.cu:200] fallback: 0
[TM][WARN][0616.14:35:12.081263][anomaly_handler.cu:206] max_batch_size: 1000
[TM][WARN][0616.14:35:12.081271][anomaly_handler.cu:207] vocab_size: 1000
[TM][FATAL][0616.14:35:12.201524][buffer.h:58] Check failed: data_type_v == dtype_ (f16 vs. bf16)
[TM][FATAL][0616.14:35:12.201555][buffer.h:58] Check failed: data_type_v == dtype_ (f16 vs. bf16)
*** stacktrace of thread 0x7f64a97fa700 ***
[ 0] TM_CHECK @ buffer.h:58
[ 1] MoeFfnLayer::Forward() @ moe_ffn_layer.cc:66
[ 2] layer_0 @ unified_decoder.cc:237
[ 3] UnifiedDecoder::Forward() @ unified_decoder.cc:161
[ 4] LanguageModel::Impl::Forward() @ language_model.cc:393
[ 5] ModelExecutor::Impl::Run() @ model_executor.cc:59
[ 6] ModelExecutor::Impl::InternalThreadEntry() @ model_executor.cc:35
[TM][FATAL][0616.14:35:12.201596][buffer.h:58] Check failed: data_type_v == dtype_ (f16 vs. bf16)
*** stacktrace of thread 0x7f64a9ffb700 ***
[ 0] TM_CHECK @ buffer.h:58
[ 1] MoeFfnLayer::Forward() @ moe_ffn_layer.cc:66
[ 2] layer_0 @ unified_decoder.cc:237
[ 3] UnifiedDecoder::Forward() @ unified_decoder.cc:161
[ 4] LanguageModel::Impl::Forward() @ language_model.cc:393
[ 5] ModelExecutor::Impl::Run() @ model_executor.cc:59
[ 6] ModelExecutor::Impl::InternalThreadEntry() @ model_executor.cc:35
*** stacktrace of thread 0x7f64aaffd700 ***
[ 0] TM_CHECK @ buffer.h:58
[ 1] MoeFfnLayer::Forward() @ moe_ffn_layer.cc:66
[ 2] layer_0 @ unified_decoder.cc:237
[ 3] UnifiedDecoder::Forward() @ unified_decoder.cc:161
[ 4] LanguageModel::Impl::Forward() @ language_model.cc:393
[ 5] ModelExecutor::Impl::Run() @ model_executor.cc:59
[ 6] ModelExecutor::Impl::InternalThreadEntry() @ model_executor.cc:35
[TM][FATAL][0616.14:35:12.201692][buffer.h:58] Check failed: data_type_v == dtype_ (f16 vs. bf16)
*** stacktrace of thread 0x7f64aa7fc700 ***
[ 0] TM_CHECK @ buffer.h:58
[ 1] MoeFfnLayer::Forward() @ moe_ffn_layer.cc:66
[ 2] layer_0 @ unified_decoder.cc:237
[ 3] UnifiedDecoder::Forward() @ unified_decoder.cc:161
[ 4] LanguageModel::Impl::Forward() @ language_model.cc:393
[ 5] ModelExecutor::Impl::Run() @ model_executor.cc:59
[ 6] ModelExecutor::Impl::InternalThreadEntry() @ model_executor.cc:35
./start_lmdeploy_oss-120b.sh: line 22: 3842712 Aborted (core dumped) lmdeploy serve api_server openai/gpt-oss-120b --server-port 8000 --backend turbomind --model-name gpt-oss-120b --enable-prefix-caching --cache-max-entry-count 0.90 --enable-prefix-caching --tp 4 --communicator native --log-level WARNING --max-log-len 200

Reproduction

mdeploy serve api_server openai/gpt-oss-120b --server-port 8000 --backend turbomind --model-name gpt-oss-120b --enable-prefix-caching --cache-max-entry-count 0.90 --enable-prefix-caching --tp 4 --communicator native --log-level WARNING --max-log-len 200

Environment

sys.platform: linux
Python: 3.12.0 | packaged by Anaconda, Inc. | (main, Oct  2 2023, 17:29:18) [GCC 11.2.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3: Tesla V100-PCIE-32GB
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.1, V12.1.66
GCC: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.3) 9.4.0
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: 0.14.0a1+
transformers: 5.12.1
fastapi: 0.135.1
pydantic: 2.12.5
triton: 3.6.0
NVIDIA Topology:
        GPU0    GPU1    GPU2    GPU3    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      PIX     PIX     PIX     0-17,36-53      0               N/A
GPU1    PIX      X      PIX     PIX     0-17,36-53      0               N/A
GPU2    PIX     PIX      X      PIX     0-17,36-53      0               N/A
GPU3    PIX     PIX     PIX      X      0-17,36-53      0               N/A

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

Error traceback