nim-interactive-chat

Generate interactive chat scripts for NVIDIA NIM models, enabling one-click start, auto health check, OpenAI-compatible API, and seamless local conversation.

Security Scan

The skill's requested actions and requirements match its description: it generates start/stop scripts and an interactive client for NVIDIA NIM models and does not ask for unrelated credentials or install arbitrary code.

Purpose & Capability

Name/description (create interactive chat scripts for NIM models) matches the SKILL.md content. Required runtime tools (Docker, Python, openai package) are consistent with starting containers and providing an OpenAI-compatible client.

Instruction Scope

SKILL.md instructs creating scripts in the user's home directory, starting/stopping Docker containers, removing old containers, and waiting for health checks. Those actions are within scope for a start/stop/chat helper but are potentially disruptive (container removal, port exposure) so users should review generated scripts before running them.

Install Mechanism

Instruction-only skill with no install specification or downloaded artifacts — the lowest-risk form. Nothing will be written by the skill itself until the agent/user runs the generated scripts.

Credentials

The skill declares no required environment variables (none in metadata), which is consistent with the package being instruction-only. However, the generated Python client is 'OpenAI-compatible' and the SKILL.md lists the openai Python package as a requirement — a runtime API key (e.g., OPENAI_API_KEY or an equivalent) may be needed when using that client but is not declared here. This omission is plausible but worth awareness.

Persistence & Privilege

No always: true or other elevated privileges requested. The skill writes files under ~/ (start/stop scripts and a chat client) and controls Docker containers — expected for its purpose but users should confirm file locations and permissions before executing.

Assessment

This skill is coherent for creating local start/stop scripts and an interactive OpenAI-compatible client for NIM models, but review the generated files before running them. Points to consider: 1) The scripts will create/stop/remove Docker containers and write files to your home directory — ensure you don't overwrite important containers or files. 2) Verify the Docker image source (e.g., tgcr.turing-agi.com) is trusted before pulling. 3) The Python client requires the openai package and may need an API key at runtime even though no env vars are declared; be prepared to provide credentials when you run the client and avoid exposing them in plain script files. 4) Check port mappings and firewall rules to avoid unintended network exposure. If unsure, run the generated scripts in a controlled environment (VM or sandbox) and inspect the scripts line-by-line before execution.

为 NVIDIA NIM 格式大模型一键创建交互式对话启动脚本，自动等待服务就绪，一键启动直接进入对话，使用体验和本地聊天程序一样流畅。

功能特点

• ✅ 一键创建全套脚本：启动 + 停止 + 交互式对话客户端
• ✅ 自动删除旧容器避免冲突
• ✅ 自动等待健康检查通过，就绪后自动打开对话
• ✅ OpenAI 兼容 API，支持流式输出
• ✅ 保持对话上下文
• ✅ 输入 exit 退出，容器继续后台运行
• ✅ 支持自定义 GPU 显存利用率

使用方法

create-nim-interactive <image-name> <container-name> <port> <model-name> [gpu-memory-utilization]

参数说明

参数	必填	说明
image-name	✅	Docker 镜像名称
container-name	✅	容器名称
port	✅	本地端口映射
model-name	✅	模型名称（API 用）
gpu-memory-utilization	⭕	GPU 显存利用率，默认 0.9

示例

为 Qwen3.5-4B 创建：

create-nim-interactive \
  tgcr.turing-agi.com/public/qwen/qwen3.5-4b-spark-nim:latest \
  qwen3.5-4b-nim \
  18000 \
  Qwen/Qwen3.5-4B \
  0.88

创建完成后：

# 一键启动（自动等待服务就绪 + 进入交互式对话）
./start-qwen3.5-4b-nim.sh

# 一键停止容器
./stop-qwen3.5-4b-nim.sh

生成文件

• ~/start-<container-name>.sh - 一键启动脚本
• ~/stop-<container-name>.sh - 一键停止脚本
• ~/<container-name>-chat.py - Python 交互式对话客户端（OpenAI 兼容）

requirements

• Docker
• Python 3
• openai Python package

适用场景

• 本地部署 NVIDIA NIM 格式模型（NGC / TGC 等）
• 想要一键启动直接对话的便捷体验
• 和 Gemma 一样友好的使用方式