Skip to main content

dequant_ggml_bytes

fuel_quantized::dequant

Function dequant_ggml_bytes 

Source 
pub fn dequant_ggml_bytes(
    bytes: &[u8],
    dtype: GgmlDType,
    name: &str,
) -> Result<Vec<f32>>
Expand description

Dequantize a raw GGUF tensor byte slice of the given GgmlDType to f32.

name is used only to name the tensor in an error message on a malformed byte count for F32/F16/BF16 (the three unpacked formats, checked exactly; the block formats silently drop a partial trailing block, matching prior per-model behavior).