[Bug] internlm2_5-7b-chat多卡部署报错 aborted

[SachaHu]

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

4卡T4的服务器，使用lmdeploy部署internlm2_5-7b-chat，张量并行tp=2
模型成功加载到显存，api接口服务正常。
调用推理接口，模型报错aborted，进程结束

使用internlm/internlm2_5-7b-chat-4bit，可以在单卡正常部署

Reproduction

使用modelscope
export LMDEPLOY_USE_MODELSCOPE=True
用cli工具部署服务
lmdeploy serve api_server Shanghai_AI_Laboratory/internlm2_5-7b-chat --backend turbomind --chat-template internlm2 --tp 2
用其他机器post请求推理接口
ip:23333/v1/chat/completions
{ "model": "/root/.cache/modelscope/hub/Shanghai_AI_Laboratory/internlm2_5-7b-chat", "messages": [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "讲一个三国故事"} ], "temperature": 0.7, "top_p": 0.8 }
进程报错跳出
(lmdeploy) [root@local-gpu models]# lmdeploy serve api_server Shanghai_AI_Laboratory/internlm2_5-7b-chat --backend turbomind --chat-template internlm2 --tp 2 [WARNING] gemm_config.in is not found; using default GEMM algo [WARNING] gemm_config.in is not found; using default GEMM algo HINT: Please open http://0.0.0.0:23333 in a browser for detailed api usage!!! HINT: Please open http://0.0.0.0:23333 in a browser for detailed api usage!!! HINT: Please open http://0.0.0.0:23333 in a browser for detailed api usage!!! INFO: Started server process [32752] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:23333 (Press CTRL+C to quit) 已放弃

Environment

'(lmdeploy) [root@local-gpu models]# lmdeploy check_env
sys.platform: linux
Python: 3.8.19 (default, Mar 20 2024, 19:58:24) [GCC 11.2.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3: Tesla T4
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.0, V12.0.140
GCC: gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-28)
PyTorch: 2.3.1+cu121
PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2022.2-Product Build 20220804 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v3.3.6 (Git Hash 86e6af5974177e513fd3fee58425e1063e7f1361)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - CUDA Runtime 12.1
  - NVCC architecture flags: -gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-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
  - CuDNN 8.9.2
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.1, CUDNN_VERSION=8.9.2, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -Wno-error=pedantic -Wno-error=old-style-cast -Wno-missing-braces -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=2.3.1, USE_CUDA=ON, USE_CUDNN=ON, USE_CUSPARSELT=1, USE_EXCEPTION_PTR=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,

TorchVision: 0.18.1+cu121
LMDeploy: 0.6.0+
transformers: 4.44.2
gradio: Not Found
fastapi: 0.115.0
pydantic: 2.9.2
triton: 2.3.1
NVIDIA Topology:
        GPU0    GPU1    GPU2    GPU3    CPU Affinity    NUMA Affinity
GPU0     X      NODE    NODE    SYS     0-15,32-47      0
GPU1    NODE     X      PHB     SYS     0-15,32-47      0
GPU2    NODE    PHB      X      SYS     0-15,32-47      0
GPU3    SYS     SYS     SYS      X      16-31,48-63     1

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
'
模型是

Error traceback

(lmdeploy) [root@local-gpu models]# lmdeploy serve api_server Shanghai_AI_Laboratory/internlm2_5-7b-chat --backend turbomind --chat-template internlm2 --tp 2
[WARNING] gemm_config.in is not found; using default GEMM algo
[WARNING] gemm_config.in is not found; using default GEMM algo
HINT:    Please open http://0.0.0.0:23333 in a browser for detailed api usage!!!
HINT:    Please open http://0.0.0.0:23333 in a browser for detailed api usage!!!
HINT:    Please open http://0.0.0.0:23333 in a browser for detailed api usage!!!
INFO:     Started server process [32752]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:23333 (Press CTRL+C to quit)
已放弃

因为internlm/internlm2_5-7b-chat-4bit可以在单卡正常部署，我认为可能和显卡驱动和nccl有关

[SachaHu]

使用多卡推理internlm2_5-7b-chat-4bit
lmdeploy serve api_server /data/models/internlm-7b-chat-int4 --backend turbomind --model-format awq --chat-template internlm2 --tp 2
推理服务正常运行。所以大概能排除多卡推理的问题。

[lvhan028]

export TM_DEBUG_LEVEL=DEBUG
启动服务时，加上选项 --log-level=DEBUG， 看看日志报错在哪里

[SachaHu]

export TM_DEBUG_LEVEL=DEBUG 启动服务时，加上选项 --log-level=DEBUG， 看看日志报错在哪里

@lvhan028 这是日志最后的部分，报错大概是[TM][DEBUG] getPtr with type i4, but data type is: x

[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: decoder_output
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: last_token_hidden_units
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: last_token_hidden_units
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_decoder.cc:148
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_decoder.cc:148
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: layer_id
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: layer_id
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: hidden_features
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_k_len
[TM][DEBUG] void turbomind::UnifiedAttentionLayer::forward(turbomind::TensorMap*, const turbomind::TensorMap*, const WeightType*) [with T = __nv_bfloat16; turbomind::UnifiedAttentionLayer::WeightType = turbomind::LlamaAttentionWeight<__nv_bfloat16>]
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: hidden_features
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] void turbomind::UnifiedAttentionLayer::forward(turbomind::TensorMap*, const turbomind::TensorMap*, const WeightType*) [with T = __nv_bfloat16; turbomind::UnifiedAttentionLayer::WeightType = turbomind::LlamaAttentionWeight<__nv_bfloat16>]
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: layer_id
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: layer_id
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: dc_batch_size
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: dc_batch_size
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: pf_batch_size
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: pf_batch_size
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T turbomind::Tensor::getVal() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_q_len
[TM][DEBUG] T turbomind::Tensor::getVal(size_t) const [with T = int; size_t = long unsigned int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_k_len
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_q_len
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_k_len
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_q_len
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_q_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_k_len
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: h_cu_k_len
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: finished
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = bool] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: finished
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: rope_theta
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = bool] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = float] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: rope_theta
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: block_ptrs
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = float] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = void*] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: block_ptrs
[TM][DEBUG] getPtr with type x, but data type is: u8
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = void*] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_block_counts
[TM][DEBUG] getPtr with type x, but data type is: u8
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: cu_block_counts
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: input_query
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: hidden_features
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: hidden_features
[TM][DEBUG] void turbomind::UnifiedAttentionLayer::allocateBuffer(size_t, size_t, size_t, const WeightType*) [with T = __nv_bfloat16; size_t = long unsigned int; turbomind::UnifiedAttentionLayer::WeightType = turbomind::LlamaAttentionWeight<__nv_bfloat16>]
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = __nv_bfloat16] start
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] void turbomind::UnifiedAttentionLayer::allocateBuffer(size_t, size_t, size_t, const WeightType*) [with T = __nv_bfloat16; size_t = long unsigned int; turbomind::UnifiedAttentionLayer::WeightType = turbomind::LlamaAttentionWeight<__nv_bfloat16>]
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] malloc buffer 0x6d0394400 with size 153600
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] malloc buffer 0xe1a394400 with size 153600
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] malloc buffer 0x6d03b9c00 with size 102400
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] malloc buffer 0xe1a3b9c00 with size 102400
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] void* turbomind::IAllocator::reMalloc(T*, size_t, bool, bool) [with T = __nv_bfloat16; size_t = long unsigned int]
[TM][DEBUG] Cannot find buffer (nil), mallocing new one.
[TM][DEBUG] malloc buffer 0x6d03d2c00 with size 182272
[TM][DEBUG] virtual void* turbomind::Allocatorturbomind::AllocatorType::CUDA::malloc(size_t, bool, bool)
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: lora_mask
[TM][DEBUG] malloc buffer 0xe1a3d2c00 with size 182272
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] bool turbomind::TensorMap::isExist(const string&) const for key: lora_mask
[TM][DEBUG] T* turbomind::Tensor::getPtr() const [with T = int] start
[TM][DEBUG] getPtr with type i4, but data type is: x
[TM][DEBUG] getPtr with type i4, but data type is: x
[TM][DEBUG] void turbomind::cublasMMWrapper::Gemm(cublasOperation_t, cublasOperation_t, int, int, int, const void*, int, const void*, int, void*, int, float, float)
[TM][DEBUG] void turbomind::cublasMMWrapper::Gemm(cublasOperation_t, cublasOperation_t, int, int, int, const void*, int, const void*, int, void*, int, float, float)
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/utils/cublasMMWrapper.cc:326
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/LlamaLinear.cu:105
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:217
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/utils/cublasMMWrapper.cc:326
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/LlamaLinear.cu:105
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:217
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:346
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:346
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:350
[TM][DEBUG] run syncAndCheck at /lmdeploy/src/turbomind/models/llama/unified_attention_layer.cc:350
已放弃