Local Chinese AI Model Setup

I want to run an open-source Chinese LLM—most likely Qwen, Deepseek, or Kimi—entirely under my own roof for document analysis, document preparation, and workflow management. My main priority is the hardware side: advising me on whether to buy a single GPU for the long haul, rent a cloud GPU, or start with a more modest card and upgrade later. All three options are on the table, so I need a clear comparison of cost, VRAM requirements, and expected performance.

Once the hardware decision is made, I’ll need hands-on help getting the chosen model installed, optimised, and benchmarked on that setup. That includes configuring the runtime, loading the model weights, and confirming that everything runs smoothly on my local network. After it works, please walk me through everyday usage—prompting, fine-tuning on my own documents, setting up retrieval-augmented generation for PDF and Word files, and basic troubleshooting steps I can repeat later.

Freelancers based in Singapore or similar timezone will be preferred.

Key deliverables
• Hardware recommendation report with buy/rent links and a simple ROI breakdown
• Full installation and configuration of the selected model on the agreed hardware or cloud instance
• Performance test results (latency, token throughput, VRAM usage)
• One live call or screen-share to demonstrate workflows and a follow-up session to answer additional questions

If you’re fluent with Chinese LLMs and small-scale self-hosting, I’m ready to get started.

Hourly rate is indicative only and based on my very limited experience. If you have a different rate, please put that in your response.