[Bug] deepseek v3 deployment on h200

[zhyncs]

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

pip3 install lmdeploy==0.7.0
lmdeploy serve api_server deepseek-ai/DeepSeek-V3 --tp 8 --backend pytorch

LICENSE-MODEL: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████| 13.8k/13.8k [00:00<00:00, 48.1MB/s]
.gitattributes: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████| 1.52k/1.52k [00:00<00:00, 4.03MB/s]
README_WEIGHTS.md: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 3.65k/3.65k [00:00<00:00, 16.8MB/s]
LICENSE-CODE: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.06k/1.06k [00:00<00:00, 4.70MB/s]
inference/configs/config_16B.json: 100%|███████████████████████████████████████████████████████████████████████████████████████| 417/417 [00:00<00:00, 1.70MB/s]
README.md: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████| 22.6k/22.6k [00:00<00:00, 12.8MB/s]
figures/benchmark.png: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 184k/184k [00:00<00:00, 1.62MB/s]
figures/niah.png: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████| 108k/108k [00:00<00:00, 1.32MB/s]
inference/configs/config_236B.json: 100%|██████████████████████████████████████████████████████████████████████████████████████| 455/455 [00:00<00:00, 2.02MB/s]
inference/configs/config_671B.json: 100%|██████████████████████████████████████████████████████████████████████████████████████| 503/503 [00:00<00:00, 2.26MB/s]
inference/convert.py: 100%|████████████████████████████████████████████████████████████████████████████████████████████████| 3.25k/3.25k [00:00<00:00, 14.4MB/s]
inference/fp8_cast_bf16.py: 100%|██████████████████████████████████████████████████████████████████████████████████████████| 3.24k/3.24k [00:00<00:00, 14.3MB/s]
inference/generate.py: 100%|███████████████████████████████████████████████████████████████████████████████████████████████| 5.39k/5.39k [00:00<00:00, 21.6MB/s]
inference/kernel.py: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 4.33k/4.33k [00:00<00:00, 18.2MB/s]
inference/requirements.txt: 100%|█████████████████████████████████████████████████████████████████████████████████████████████| 66.0/66.0 [00:00<00:00, 293kB/s]
inference/model.py: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 17.6k/17.6k [00:00<00:00, 50.4MB/s]
modeling_deepseek.py: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 75.8k/75.8k [00:00<00:00, 684kB/s]
Fetching 185 files: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████| 185/185 [00:01<00:00, 184.46it/s]

The phenomenon is stuck here.

Reproduction

as mentioned above

Environment

sys.platform: linux
Python: 3.10.16 (main, Dec  4 2024, 08:53:37) [GCC 9.4.0]
CUDA available: True
MUSA available: False
numpy_random_seed: 2147483648
GPU 0,1,2,3,4,5,6,7: NVIDIA H200
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.4, V12.4.131
GCC: x86_64-linux-gnu-gcc (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0
PyTorch: 2.5.1+cu124
PyTorch compiling details: PyTorch built with:
  - GCC 9.3
  - 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 v3.5.3 (Git Hash 66f0cb9eb66affd2da3bf5f8d897376f04aae6af)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX512
  - CUDA Runtime 12.4
  - 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 90.1
  - Magma 2.6.1
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=12.4, CUDNN_VERSION=9.1.0, 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 -DLIBKINETO_NOXPUPTI=ON -DUSE_FBGEMM -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-strict-overflow -Wno-strict-aliasing -Wno-stringop-overflow -Wsuggest-override -Wno-psabi -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, TORCH_VERSION=2.5.1, 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: 0.20.1+cu124
LMDeploy: 0.7.0+c2f212d
transformers: 4.48.0
gradio: Not Found
fastapi: 0.115.6
pydantic: 2.10.5
triton: 3.1.0
NVIDIA Topology:
	GPU0	GPU1	GPU2	GPU3	GPU4	GPU5	GPU6	GPU7	NIC0	NIC1	NIC2	NIC3	NIC4	NIC5	NIC6	NIC7	NIC8	CPU Affinity	NUMA Affinity	GPU NUMA ID
GPU0	 X 	NV18	NV18	NV18	NV18	NV18	NV18	NV18	NODE	NODE	PIX	NODE	SYS	SYS	SYS	SYS	SYS	0-95	0		N/A
GPU1	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NV18	NODE	NODE	NODE	PIX	SYS	SYS	SYS	SYS	SYS	0-95	0		N/A
GPU2	NV18	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NODE	PIX	NODE	NODE	SYS	SYS	SYS	SYS	SYS	0-95	0		N/A
GPU3	NV18	NV18	NV18	 X 	NV18	NV18	NV18	NV18	PIX	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS	0-95	0		N/A
GPU4	NV18	NV18	NV18	NV18	 X 	NV18	NV18	NV18	SYS	SYS	SYS	SYS	NODE	NODE	PIX	NODE	NODE	96-191	1		N/A
GPU5	NV18	NV18	NV18	NV18	NV18	 X 	NV18	NV18	SYS	SYS	SYS	SYS	NODE	NODE	NODE	PIX	NODE	96-191	1		N/A
GPU6	NV18	NV18	NV18	NV18	NV18	NV18	 X 	NV18	SYS	SYS	SYS	SYS	NODE	PIX	NODE	NODE	PHB	96-191	1		N/A
GPU7	NV18	NV18	NV18	NV18	NV18	NV18	NV18	 X 	SYS	SYS	SYS	SYS	PIX	NODE	NODE	NODE	NODE	96-191	1		N/A
NIC0	NODE	NODE	NODE	PIX	SYS	SYS	SYS	SYS	 X 	NODE	NODE	NODE	SYS	SYS	SYS	SYS	SYS
NIC1	NODE	NODE	PIX	NODE	SYS	SYS	SYS	SYS	NODE	 X 	NODE	NODE	SYS	SYS	SYS	SYS	SYS
NIC2	PIX	NODE	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	 X 	NODE	SYS	SYS	SYS	SYS	SYS
NIC3	NODE	PIX	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	NODE	 X 	SYS	SYS	SYS	SYS	SYS
NIC4	SYS	SYS	SYS	SYS	NODE	NODE	NODE	PIX	SYS	SYS	SYS	SYS	 X 	NODE	NODE	NODE	NODE
NIC5	SYS	SYS	SYS	SYS	NODE	NODE	PIX	NODE	SYS	SYS	SYS	SYS	NODE	 X 	NODE	NODE	PHB
NIC6	SYS	SYS	SYS	SYS	PIX	NODE	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	 X 	NODE	NODE
NIC7	SYS	SYS	SYS	SYS	NODE	PIX	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	NODE	 X 	NODE
NIC8	SYS	SYS	SYS	SYS	NODE	NODE	PHB	NODE	SYS	SYS	SYS	SYS	NODE	PHB	NODE	NODE	 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_0
  NIC1: mlx5_1
  NIC2: mlx5_2
  NIC3: mlx5_3
  NIC4: mlx5_4
  NIC5: mlx5_5
  NIC6: mlx5_8
  NIC7: mlx5_9
  NIC8: mlx5_bond_0

Error traceback

[zhyncs]

lmdeploy serve api_server deepseek-ai/DeepSeek-V3 --tp 8 --backend pytorch --log-level INFO

[zhyncs]

15 minutes have passed and it still shows loading.

[grimoire]

Errr, it is just Desperately SLOW. We have not performed much optimization on weight loading.

[RunningLeon]

@zhyncs hi, can you try on this PR #2886 ?

[zhyncs]

@zhyncs hi, can you try on this PR #2886 ?

I'll try it today. Thanks!

[zhyncs]

I did a test, and it still takes over 20 minutes to finish loading. Is this within expectations?

[RunningLeon]

I did a test, and it still takes over 20 minutes to finish loading. Is this within expectations?

@zhyncs hi, we don't have deepseekv3 so it tested on deepseekv2-chat with tp=8 and loading timing can reduce from 15min to 8 min.

[zhyncs]

Hi @RunningLeon, I've granted @grimoire access to the H200. Could you @grimoire please help verify? Thanks!

[grimoire]

Is the model placed on NFS? The uncached first-time load could be slow.

[zhyncs]

Is the model placed on NFS? The uncached first-time load could be slow.

@grimoire I think it's placed on SSD.