Model parameters

In the Ultra Transcenders books

AI-901

Each book explains Model parameters in context, with comparison tables and the common traps.

Terms in this definition

• SLM

  Smaller than an LLM, with roughly under 10 billion parameters but a similar kind of architecture. Running it costs less and is quicker, at the price of narrower capability.

Related terms

• Checksum

  Confirms that a check digit in text found by a custom sensitive information type's regex is correct, calculated from weights and a mod value. The advanced version can only be set up in PowerShell XML.

• Cloud Secure Score

  Shown in the Microsoft Defender portal, this posture score weights outstanding cloud recommendations by how risky each is, by whether assets face the internet and similar factors, and by how critical those assets are. The Azure portal keeps the older classic score.

• Few-shot learning

  Putting sample inputs with their outputs into a prompt to demonstrate the pattern and format wanted, without altering model weights; zero-shot prompting gives no samples.

• Fine-tuning

  Continuing to train a pretrained base model on task examples, through supervised fine-tuning, DPO or RFT, so its weights shift towards a style, format or task. It neither acts as a safety control nor adds new knowledge well.

• Prompt engineering

  The practice of writing prompts, system messages and examples that guide what a model produces, leaving its weights untouched.

• RRF

  A Cosmos DB system function, used with ORDER BY RANK, that merges the rankings from VectorDistance and FullTextScore, with optional weights, to give hybrid search results.

• Traffic splitting

  A Container Apps capability, configured in configuration.ingress.traffic, that gives revisions percentage weights adding up to 100%, picking each revision by name or by label.

• Weighted

  Distributes traffic among Traffic Manager endpoints in proportion to the weights assigned to them.