iklobato/claude-qwen: Run Claude Code on a self-hosted uncensored Qwen model

This guide details how to run the Claude Code CLI against a self-hosted, uncensored Qwen model on a rented Vast.ai GPU using llama-server. The setup uses a patched Jinja chat template and an SSH tunnel to provide full tool calling and a 262k context length. Measured performance on an A100 GPU reaches high token generation speeds suitable for authorized penetration testing. Open article →