Struct QuantInfo 

pub struct QuantInfo {
    pub scheme: String,
    pub stored_bytes: u64,
    pub dequantized_bytes: u64,
}

Quantization scheme + size estimates for a cached .safetensors file.

Populated by inspect_safetensors_local (the cache-hit path) via anamnesis::InspectInfo::from(&header). Absent (None) when:

• the safetensors file has no detected quantization (QuantScheme::Unquantized),
• the inspect ran over HTTP Range (the remote path’s bespoke parser doesn’t go through anamnesis until v0.11.1), or
• the file format isn’t safetensors (GGUF / NPZ / PTH carry no quant-method metadata).

Decoupled from anamnesis::QuantScheme (a #[non_exhaustive] enum) so downstream library consumers (candle-mi, anamnesis) aren’t forced to match every variant. The scheme field stores QuantScheme’s Display output ("FineGrainedFp8", "Bnb4", "Gptq", "Awq", …); consumers that need to match exact variants should call anamnesis::parse_safetensors_header themselves.

Fields

scheme: String

Detected quantization scheme as the Display form of anamnesis::QuantScheme (e.g. "FineGrainedFp8", "Bnb4").

stored_bytes: u64

Bytes stored on disk for tensor data (header excluded).

dequantized_bytes: u64

Estimated bytes after dequantising to BF16. For BnB-NF4/FP4 (U8-packed nibbles), this is stored_bytes × 4; for FP8 / GPTQ / AWQ / BnB-INT8 it’s num_elements × 2 summed over weight tensors, plus passthrough tensors copied as-is. The formula lives in anamnesis::InspectInfo::from(&SafetensorsHeader) — hf-fm just reads the result.

Trait Implementations

impl Clone for QuantInfo

fn clone(&self) -> QuantInfo

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for QuantInfo

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Serialize for QuantInfo

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.