qualia_core_db/inference/ggml_quants/
mod.rs1pub use etalon_model::ggml::*;
3#[cfg(test)]
4use crate::gguf_sharder::GgufTensorInfo;
5#[cfg(test)]
6mod tests;
7
8#[cfg(test)]
9fn dequant_q4_k(raw: &[u8], n_elems: usize, out: &mut [f32]) -> Result<usize, GgmlDequantError> {
10 dequantize_row_into(raw, GGML_TYPE_Q4_K, n_elems, out)
11}
12
13#[cfg(test)]
14fn dequant_q4_k_soa(
15 raw: &[u8],
16 n_elems: usize,
17 out: &mut [f32],
18) -> Result<usize, GgmlDequantError> {
19 dequantize_row_into(raw, GGML_TYPE_Q4_K_SOA, n_elems, out)
20}
21
22#[cfg(test)]
23fn dequant_q5_k(raw: &[u8], n_elems: usize, out: &mut [f32]) -> Result<usize, GgmlDequantError> {
24 dequantize_row_into(raw, GGML_TYPE_Q5_K, n_elems, out)
25}
26
27#[cfg(test)]
28fn dequant_q8_0(raw: &[u8], n_elems: usize, out: &mut [f32]) -> Result<usize, GgmlDequantError> {
29 dequantize_row_into(raw, GGML_TYPE_Q8_0, n_elems, out)
30}
31
32#[cfg(test)]
33fn quantize_block_f32_to_q4_k_soa(src: &[f32], out: &mut [u8]) {
34 quantize_f32_to_q4_k_soa_tensor(src, src.len(), 1, out)
35 .expect("valid original single-block fixture");
36}