Classifier Free Guidance One Coordinate Tool

Description

Return one classifier-free-guidance coordinate.

Classifier-free guidance combines conditional and unconditional model predictions. Interpolation and extrapolation depend on the multiplier convention: here zero selects unconditional, one selects conditional, and values above one extend beyond the conditional prediction. One scalar coordinate is distinct from a full prediction tensor or diffusion-sampling process.

Return one classifier-free-guidance coordinate.

When to use this tool

Use Classifier Free Guidance One Coordinate when checking the stated probability, score, guidance or scale contribution.

How to use the tool and how it works

Enter unconditional, conditional, scale. Use the units and conventions below.
Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
Read guided, conditionalIncrement. Use the output convention below.

Return one classifier-free-guidance coordinate. conditionalIncrement = scale*(conditional-unconditional), guided = unconditional+increment.^[1]

Classifier-free guidance combines conditional and unconditional predictions. The multiplier convention here uses scale=1 for the conditional value; compare definitions before transferring a paper or library setting.^[2]

Input field	Meaning and units	Accepted input
unconditional	Unconditional	Required. Finite number.
conditional	Conditional	Required. Finite number.
scale	Signed multiplier in unconditional+scale*(conditional-unconditional); check conventions when comparing to papers/libraries.	Required. Finite number.

Output field	Interpretation	Returned shape
guided	Guided	Required. Finite number.
conditionalIncrement	Conditional Increment	Required. Finite number.

json

{
  "input": {
    "unconditional": 2,
    "conditional": 5,
    "scale": 3
  },
  "output": {
    "guided": 11,
    "conditionalIncrement": 9
  }
}

Limitations and assumptions

Scale may be signed; this is one supplied coordinate convention, not a full diffusion step, noise schedule or image generation.
Inputs must satisfy both the field types and the cross-field conditions above. Invalid input causes an error; a successful false flag or documented placeholder is a different result.
Numeric calculations use binary floating point unless exact decimal-integer arithmetic is explicitly stated. No hidden measurement tolerance or decimal rounding is added; nonfinite or out-of-contract results are rejected. Numeric signed-zero results are exported as zero.

Alternative or Complementary methods

Use the full distribution and model-specific decoding or training objective for normalized probabilities, preference optimization or actual generation.

References

Calculation and input contract: artificial-intelligence/classifier-free-guidance-one-coordinate — tool.io
Classifier-Free Diffusion Guidance — Jonathan Ho and Tim Salimans

Similar or alternative tools

Logit Scale Temperature Composition
Compose a positive logit scale with two temperatures.
Logit Bias One Token Odds Factor
Return one-token logit-bias odds transformation.