prism-inference 0.1.0

Bounded Rust CPU inference and tokenization for Prism-compatible GGUF models
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13

//! Checked scheduling bounds for multi-token prediction.

/// Whether catch-up rows and sequence anchors fit the first decoder batch.
///
/// The addition uses a widened integer so valid `i32` inputs never overflow.
/// A nonpositive batch or a negative row/anchor count is rejected.
#[must_use]
pub fn mtp_first_decode_fits(batch: i32, catchup_rows: i32, anchors: i32) -> bool {
    batch > 0
        && catchup_rows >= 0
        && anchors >= 0
        && i64::from(catchup_rows) + i64::from(anchors) <= i64::from(batch)
}