Skip to main content

Service

On-Policy Context Distillation (OPCD)

A long system prompt moved into the weights, so every call is shorter.

Talk to us

What we do​

  • Prompt into weightsThe model answers as it does with your full system prompt, without sending it.
  • Fewer input tokensRequests drop the system prompt and few-shot examples, and the tokens they cost.
  • Prompt kept privateYour instructions live in the weights, not in the requests you send.

How we work​

  1. Write the context. We settle the system prompt, examples and instructions that give the behaviour you want.
  2. Generate and train. The model answers varied prompts with the full context, then learns to give those answers without it.
  3. Compare. The distilled model and the prompted model answer the same test set.

What you get​

  • The distilled model, which needs no system prompt.
  • The test results of the two models, side by side.

Start​

Send us the system prompt you use today and a sample of real requests, and we reply with a scope and a quote. Email sale@iapp.co.th or call 02-124-4041.