TensorRT-LLM 安装使用笔记

目录

2026.04.01 linux 安装成功；

docker版；

docker 验证：

2026.04.01 linux 安装成功；

pip install tensorrt-llm

上面的安装完，自己安装了torch2.9

匹配版本安装，待测试结果：

pip install torch==2.8.0+cu128 torchvision==0.23.0+cu128

tensorrt-llm应该安装哪个版本

python -c "import torch; print(f'torch=={torch.__version__}')" > /tmp/torch-constraint.txt

tensorrt_llm-1.1.0.tar.gz

conda install -c conda-forge openmpi mpi4py

ImportError: libcublasLt.so.13: cannot open shared object file: No such file or directory

from flash_head.ltx_video.ltx_vae import LtxVAE self.vae = LtxVAE( pretrained_model_type_or_path=vae_dir, dtype=self.param_dtype, device=self.device, )

docker版；

docker pull nvcr.io/nvidia/cuda:12.2.0-base-ubuntu20.04

docker compose \ --env-file ./.env \ -f docker-compose.cosyvoice2.dit.yml up

{ "registry-mirrors_1": [ "https://0qithrlo.mirror.aliyuncs.com" ], "runtimes": { "nvidia": { "args": [], "path": "nvidia-container-runtime" } } }

docker 验证：

import subprocess host = "registry-1.docker.io" try: result = subprocess.run(["ping", "-c", "3", host], capture_output=True, text=True) print(result.stdout) except Exception as e: print(f"无法 ping: {e}")