boostr 0.1.0

ML framework built on numr - attention, quantization, model architectures
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

use super::types::GgmlType;

/// Tensor info entry from a GGUF file
#[derive(Debug, Clone)]
pub struct GgufTensorInfo {
    pub name: String,
    pub n_dims: u32,
    pub shape: Vec<usize>,
    pub ggml_type: GgmlType,
    pub offset: u64,
}

impl GgufTensorInfo {
    pub fn numel(&self) -> usize {
        self.shape.iter().product()
    }

    pub fn size_bytes(&self) -> usize {
        let numel = self.numel();
        let bs = self.ggml_type.block_size();
        let bb = self.ggml_type.block_bytes();
        numel.div_ceil(bs) * bb
    }
}