Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.ai-stats.phaseo.app/llms.txt

Use this file to discover all available pages before exploring further.

This page is the entry point for a full guide series on quantization and inference parameters. If you are deciding model variants (for example BF16 vs FP8) or tuning request settings (temperature, top_p, token budgets), start here and work through the linked guides.

​
What is quantization?

Quantization is how model weights and activations are represented with lower precision numbers to reduce memory usage and speed up inference. In general:
  • Higher precision: better quality/stability, higher cost.
  • Lower precision: better throughput/cost, higher quality risk.

​
Guide series

  1. Quantization Formats
  2. Quantization Methods
  3. Choosing a Quantization Strategy
  4. Inference Parameters
  5. Sampling and Decoding Parameters
  6. Context and Token Budgeting
  1. Read formats and methods first.
  2. Use the selection guide to pick an initial variant.
  3. Tune inference parameters on top of that baseline.
  4. Finalize token budgets with production prompt traces.

​
Practical reminder

Quantization and parameter tuning should always be validated on your real prompts and acceptance criteria, not synthetic micro-benchmarks alone.
Last modified on February 18, 2026