
本镜像包含LLaVA视觉大型语言模型(Vision LLM),源自https://github.com/haotian-liu/LLaVA%E9%A1%B9%E7%9B%AE%EF%BC%8C%E4%B8%93%E4%B8%BAJetson%E5%B9%B3%E5%8F%B0%E4%BC%98%E5%8C%96%E3%80%82LLaVA%E6%94%AF%E6%8C%81%E5%9B%BE%E5%83%8F%E7%90%86%E8%A7%A3%E4%B8%8E%E8%87%AA%E7%84%B6%E8%AF%AD%E8%A8%80%E5%AF%B9%E8%AF%9D%E4%BA%A4%E4%BA%92%EF%BC%8C%E5%8F%AF%E7%94%A8%E4%BA%8E%E8%A7%86%E8%A7%89%E9%97%AE%E7%AD%94%E3%80%81%E5%9B%BE%E5%83%8F%E5%86%85%E5%AE%B9%E6%8F%8F%E8%BF%B0%E7%AD%89%E5%A4%9A%E6%A8%A1%E6%80%81%E4%BB%BB%E5%8A%A1%E3%80%82%E6%9C%80%E6%96%B0%E7%89%88%E6%9C%AC%E4%B8%BA%60llava-1.5%60%EF%BC%8C%E6%96%87%E6%A1%A3%E4%B8%AD%E7%A4%BA%E4%BE%8B%E5%9F%BA%E4%BA%8E%60llava-llama-2%60%E7%B3%BB%E5%88%97%E6%A8%A1%E5%9E%8B%EF%BC%8C%E5%8F%AF%E6%9B%BF%E6%8D%A2%E4%B8%BA%60llava-1.5%60%E4%BD%BF%E7%94%A8%E3%80%82%E9%80%9A%E8%BF%87jetson-ai-lab.com/tutorial_llava.html%E5%8F%AF%E4%BA%86%E8%A7%A3%E5%A6%82%E4%BD%95%E7%BB%93%E5%90%88text-generation-webui%E4%BD%BF%E7%94%A8%E9%87%8F%E5%8C%96%E6%A8%A1%E5%9E%8B%E3%80%82
!https://github.com/dusty-nv/jetson-containers/raw/master/data/images/hoover.jpg
llava-llama-2系列(7B/13B)及最新llava-1.5模型llava-llama-2-7b-chat
该模型基于Llama-2-7b-chat的LoRA微调版本,需先申请Llama-2访问权限并获取HuggingFace访问令牌,或使用替代模型SaffalPoosh/llava-llama-2-7B-merged。
运行命令:
bash./run.sh --env HUGGING_FACE_HUB_TOKEN=<你的访问令牌> $(./autotag llava) \ python3 -m llava.serve.cli \ --model-path liuhaotian/llava-llama-2-7b-chat-lightning-lora-preview \ --model-base meta-llama/Llama-2-7b-chat-hf \ --image-file /data/images/hoover.jpg
示例交互:
USER: 路标上写了什么? ASSISTANT: 路标上写着"Hoover Dam"。 USER: 出口还有多远? ASSISTANT: 出口还有1英里远。 USER: 周围环境如何? ASSISTANT: 环境为沙漠地貌,有岩石景观和一条通往出口的土路。
llava-llama-2-13b-chat
运行命令:
bash./run.sh $(./autotag llava) \ python3 -m llava.serve.cli \ --model-path liuhaotian/llava-llama-2-13b-chat-lightning-preview \ --image-file /data/images/hoover.jpg
示例交互:
USER: 路标上的文字是什么? ASSISTANT: 路标上的文字是"Hoover Dam Exit 2 Mile"。 USER: 出口还有多远? ASSISTANT: 出口距离当前位置2英里。 USER: 这是什么样的环境? ASSISTANT: 环境为沙漠场景,背景中有山脉。
llava | 说明 |
|---|---|
| 构建状态 | |
| 系统要求 | L4T ['>=34.1.0'] |
| 依赖项 | https://github.com/dusty-nv/jetson-containers/tree/master/packages/build/build-essential https://github.com/dusty-nv/jetson-containers/tree/master/packages/cuda/cuda https://github.com/dusty-nv/jetson-containers/tree/master/packages/cuda/cudnn https://github.com/dusty-nv/jetson-containers/tree/master/packages/build/python https://github.com/dusty-nv/jetson-containers/tree/master/packages/numpy https://github.com/dusty-nv/jetson-containers/tree/master/packages/build/cmake/cmake_pip https://github.com/dusty-nv/jetson-containers/tree/master/packages/onnx https://github.com/dusty-nv/jetson-containers/tree/master/packages/pytorch https://github.com/dusty-nv/jetson-containers/tree/master/packages/pytorch/torchvision https://github.com/dusty-nv/jetson-containers/tree/master/packages/llm/huggingface_hub https://github.com/dusty-nv/jetson-containers/tree/master/packages/build/rust https://github.com/dusty-nv/jetson-containers/tree/master/packages/llm/transformers |
| Dockerfile | https://github.com/dusty-nv/jetson-containers/tree/master/packages/llm/llava/Dockerfile |
| 镜像列表 | https://hub.docker.com/r/dustynv/llava/tags (2023-12-15, 6.3GB)https://hub.docker.com/r/dustynv/llava/tags (2023-12-12, 6.3GB)https://hub.docker.com/r/dustynv/llava/tags (2023-12-14, 6.3GB)https://hub.docker.com/r/dustynv/llava/tags (2023-12-18, 8.0GB) |
| 仓库/标签 | 日期 | 架构 | 大小 |
|---|---|---|---|
| https://hub.docker.com/r/dustynv/llava/tags | 2023-12-15 | arm64 | 6.3GB |
| https://hub.docker.com/r/dustynv/llava/tags | 2023-12-12 | arm64 | 6.3GB |
| https://hub.docker.com/r/dustynv/llava/tags | 2023-12-14 | arm64 | 6.3GB |
| https://hub.docker.com/r/dustynv/llava/tags | 2023-12-18 | arm64 | 8.0GB |
容器镜像与其他JetPack/L4T次要版本兼容:
• L4T R32.7容器可在其他L4T R32.7版本(JetPack 4.6+)上运行
• L4T R35.x容器可在其他L4T R35.x版本(JetPack 5.1+)上运行
要启动容器,可使用https://github.com/dusty-nv/jetson-containers/tree/master/docs/run.md%E5%92%8Chttps://github.com/dusty-nv/jetson-containers/tree/master/docs/run.md#autotag%EF%BC%8C%E6%88%96%E6%89%8B%E5%8A%A8%E6%9E%84%E5%BB%BA%60docker run`命令:
bash# 自动拉取或构建兼容的容器镜像 jetson-containers run $(autotag llava) # 或显式指定上述镜像之一 jetson-containers run dustynv/llava:r36.2.0 # 或使用'docker run'(需指定镜像及挂载等参数) sudo docker run --runtime nvidia -it --rm --network=host dustynv/llava:r36.2.0
https://github.com/dusty-nv/jetson-containers/tree/master/docs/run.md%E5%B0%86%E5%8F%82%E6%95%B0%E8%BD%AC%E5%8F%91%E7%BB%99%60docker run
,并添加一些默认配置(如--runtime nvidia、挂载/data`缓存、检测设备)
https://github.com/dusty-nv/jetson-containers/tree/master/docs/run.md#autotag%E4%BC%9A%E6%9F%A5%E6%89%BE%E4%B8%8E%E4%BD%A0%E7%9A%84JetPack/L4T%E7%89%88%E6%9C%AC%E5%85%BC%E5%AE%B9%E7%9A%84%E5%AE%B9%E5%99%A8%E9%95%9C%E5%83%8F%E2%80%94%E2%80%94%E6%9C%AC%E5%9C%B0%E9%95%9C%E5%83%8F%E3%80%81%E4%BB%8E%E4%BB%93%E5%BA%93%E6%8B%89%E5%8F%96%E6%88%96%E6%9E%84%E5%BB%BA%E6%96%B0%E9%95%9C%E5%83%8F%E3%80%82
要将本地目录挂载到容器中,使用-v或--volume标志:
bashjetson-containers run -v /主机路径:/容器路径 $(autotag llava)
要启动容器并运行命令(而非交互式shell):
bashjetson-containers run $(autotag llava) my_app --abc xyz
你可以传递任何docker run支持的选项,命令执行前会打印完整构建的命令。
如果使用上述https://github.com/dusty-nv/jetson-containers/tree/master/docs/run.md#autotag%EF%BC%8C%E5%BF%85%E8%A6%81%E6%97%B6%E4%BC%9A%E8%87%AA%E5%8A%A8%E6%9E%84%E5%BB%BA%E5%AE%B9%E5%99%A8%E3%80%82%E8%8B%A5%E9%9C%80%E6%89%8B%E5%8A%A8%E6%9E%84%E5%BB%BA%EF%BC%8C%E5%85%88%E5%AE%8C%E6%88%90https://github.com/dusty-nv/jetson-containers/tree/master/docs/setup.md%EF%BC%8C%E7%84%B6%E5%90%8E%E8%BF%90%E8%A1%8C%EF%BC%9A
bashjetson-containers build llava
构建过程中会集成上述依赖项并进行测试。使用https://github.com/dusty-nv/jetson-containers/tree/master/jetson_containers/build.py%E6%9F%A5%E7%9C%8B%E6%9E%84%E5%BB%BA%E9%80%89%E9%A1%B9%E3%80%82
您可以使用以下命令拉取该镜像。请将 <标签> 替换为具体的标签版本。如需查看所有可用标签版本,请访问 标签列表页面。






探索更多轩辕镜像的使用方法,找到最适合您系统的配置方式
通过 Docker 登录认证访问私有仓库
无需登录使用专属域名
Kubernetes 集群配置 Containerd
K3s 轻量级 Kubernetes 镜像加速
VS Code Dev Containers 配置
Podman 容器引擎配置
HPC 科学计算容器配置
ghcr、Quay、nvcr 等镜像仓库
Harbor Proxy Repository 对接专属域名
Portainer Registries 加速拉取
Nexus3 Docker Proxy 内网缓存
需要其他帮助?请查看我们的 常见问题Docker 镜像访问常见问题解答 或 提交工单
docker search 限制
站内搜不到镜像
离线 save/load
插件要用 plugin install
WSL 拉取慢
安全与 digest
新手拉取配置
镜像合规机制
manifest unknown
no matching manifest(架构)
invalid tar header(解压)
TLS 证书失败
DNS 超时
域名连通性排查
410 Gone 排查
402 与流量用尽
401 认证失败
429 限流
D-Bus 凭证提示
413 与超大单层
来自真实用户的反馈,见证轩辕镜像的优质服务