A buyer's checklist for machine learning models

Use this checklist when you buy a machine learning model, so you get everything you're paying for.

Model files

• Model weights file (.pt, .h5, .onnx, .safetensors)
• Model architecture code or config
• Tokenizer or preprocessor files (for NLP models)
• Label mapping (for classification models)

Code & documentation

• Inference script with example usage
• requirements.txt or environment.yml
• README with setup instructions
• API documentation (if the model is served as an API)
• Training script (for fine-tuning)

Performance

• Benchmark results on standard datasets
• Comparison with baseline models
• Latency measurements
• Memory use during inference

Legal & licensing

• A license that allows commercial use
• Training data sources disclosed
• No GPL or AGPL code you can't accept
• Third-party model dependencies and their licenses listed

Support

• Post-sale support offered
• Update and bug-fix policy written down
• Contact details provided

Using escrow

Work through this checklist before you confirm receipt. If something is missing, raise it with the seller in the deal chat, and if the seller can't provide it, open a dispute before you confirm. Aegis, EPICSWP's assistant, can summarise the case, and a human moderator decides based on the evidence.