Qwen

Use this skill when a task names Qwen, Qwen-7B/Qwen-14B/Qwen-72B, Qwen-Chat, Qwen/Qwen-*, ModelScope Qwen checkpoints, Qwen's Transformers examples, qwenllm/qwen Docker images, or the repository's fine-tuning, evaluation, serving, tokenizer, and tool-use workflows.

This is a repository operating graph, not a general model card. Start by deciding whether the request is about a base model or a chat model, whether the checkpoint is remote or local, and whether the desired action needs CPU, CUDA, Docker, a vendor accelerator, a cloud API, credentials, or a sandbox.

Fast start

• Base install for the documented Transformers workflows: pip install -r requirements.txt.
• Add only the optional dependency family needed by the route: web demo, API server, PEFT/DeepSpeed, AutoGPTQ, vLLM/FastChat, ModelScope, or a vendor runtime.
• Safe environment check: python scripts/check_qwen_environment.py --check-dependencies.
• Minimal public-package check: python -c "import torch, transformers, tiktoken; print(torch.__version__, transformers.__version__)".
• Do not download a checkpoint, start a service, run training, run a benchmark, or use a cloud key merely to prove that the package imports.

Route by task