Skip to content

5080 Laptop failed to going on😭😭😭 #181

@TeloySXH

Description

@TeloySXH

我的硬件:NVIDIA GeForce RTX 5080 Laptop GPU
我在win11中创建了WSL2,部署环境如下

【 WSL2 环境 】

Linux Ubuntu 22.04.5 LTS
CUDA 12.8.93
GCC 11.4.0
G++ 11.4.0
clang 14.0.0
在WSL2里,我使用了miniconda,创建了虚拟环境T4,部署python3.11.14
之后使用Installation,一键pip拉取依赖,但是出现了如下问题
ImportError: /home/sxh/miniconda3/envs/T4/lib/python3.11/site-packages/flash_attn_2_cuda.cpython-311-x86_64-linux-gnu.so: undefined symbol: _ZN3c104cuda29c10_cuda_check_implementationEiPKcS2_ib
我用AI修复了一段时间,并没有成功,于是我放弃了一键拉取,开始手动搭建依赖

于是我手动创建了虚拟环境T2,并将它连接到win的pycharm上,打开WSL2里的项目
【 虚拟环境 T2 WSL2 】
用途 复现 nano vllm 项目
Python 3.11.14
Pytorch 2.8.0+cu128
Triton 3.2.0
transformers 5.3.0
flash-attn 2.8.3
xxhash 3.6.0
setuptools 82.0.0
nano vllm 0.2.0
现在flash-atten的问题解决了,但是example在运行时候报错
torch_dtype is deprecated! Use dtype instead!
[rank0]: Traceback (most recent call last):
[rank0]: File "/home/sxh/Project/NanovLLM/example.py", line 33, in
[rank0]: main()
[rank0]: File "/home/sxh/Project/NanovLLM/example.py", line 9, in main
[rank0]: llm = LLM(path, enforce_eager=True, tensor_parallel_size=1)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/engine/llm_engine.py", line 30, in init
[rank0]: self.model_runner = ModelRunner(config, 0, self.events)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/engine/model_runner.py", line 31, in init
[rank0]: self.model = Qwen3ForCausalLM(hf_config)
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/models/qwen3.py", line 199, in init
[rank0]: self.model = Qwen3Model(config)
[rank0]: ^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/models/qwen3.py", line 169, in init
[rank0]: self.layers = nn.ModuleList([Qwen3DecoderLayer(config) for _ in range(config.num_hidden_layers)])
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/models/qwen3.py", line 169, in
[rank0]: self.layers = nn.ModuleList([Qwen3DecoderLayer(config) for _ in range(config.num_hidden_layers)])
[rank0]: ^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/models/qwen3.py", line 126, in init
[rank0]: self.self_attn = Qwen3Attention(
[rank0]: ^^^^^^^^^^^^^^^
[rank0]: File "/home/sxh/Project/NanovLLM/nanovllm/models/qwen3.py", line 54, in init
[rank0]: self.rotary_emb = get_rope(
[rank0]: ^^^^^^^^^
[rank0]: TypeError: unhashable type: 'dict'
[rank0]:[W308 20:43:17.782137288 ProcessGroupNCCL.cpp:1538] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())

我花了不少时间修改,但是并没有成功,请问应该如何处理这个问题😭😭

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions