For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).

Python function

compute_synthetic_acceptance_base_rate

compute_synthetic_acceptance_base_rate()

max.nn.sampling.compute_synthetic_acceptance_base_rate(p_avg, n, tol=1e-09)

source

Solves for the per-position base acceptance rate that matches a target mean.

Under independent per-position Bernoulli acceptance with cascading rejection, the mean joint acceptance across n positions is sum_{i=1..n} base ** i / n. This function binary-searches for the base that produces a mean of p_avg.

Parameters:

• p_avg (float) – Desired mean acceptance rate in [0, 1].
• n (int) – Number of speculative draft steps.
• tol (float) – Binary search tolerance.

Returns:

The per-position base acceptance rate.

Return type:

float