Struct QuantizationConfig 

pub struct QuantizationConfig {
    pub quant_method: String,
    pub quant_algo: String,
    pub format: String,
    pub ignore_modules: Vec<String>,
}

Advertised weight-quantization layout, as declared in the HF config.json’s quantization_config block (or a sibling hf_quant_config.json). This is the authoritative signal for format dispatch — the QuantFormat trait prefers this over tensor-name sniffing, matching the dispatch model used by vLLM / TensorRT-LLM / SGLang.

quant_method is the serialization scheme:

• "compressed-tensors" — Neural Magic / llm-compressor. Uses weight_packed + weight_global_scale + input_global_scale. Commonly paired with format = "nvfp4-pack-quantized" or "float-quantized".
• "modelopt" — NVIDIA TensorRT ModelOpt. Uses weight (as the packed FP4 payload when quant_algo == "NVFP4") + weight_scale
  • weight_scale_2 + input_scale.
• "fp8" — native FP8 block-scaled (e.g. Qwen/Qwen3.5-35B-A3B-FP8) with weight_scale_inv sibling tensors.

ignore_modules holds the already-expanded list of module-path patterns that should be loaded as dense BF16 rather than quantized. Patterns use HF glob semantics (* matches any non-. sub-path).

Fields

quant_method: String

Raw quant_method string from the config. Stable values: "compressed-tensors", "modelopt", "fp8".

quant_algo: String

ModelOpt-specific algorithm label: "NVFP4", "FP8", … Empty string for schemes that don’t declare one (e.g. plain FP8).

format: String

Optional format string (compressed-tensors uses this for "nvfp4-pack-quantized" and friends).

ignore_modules: Vec<String>

Module-path globs that should stay BF16 (the “ignore list” in ModelOpt terminology; targets/exclude_modules in compressed- tensors). Example entries: "lm_head", "model.layers.*.self_attn*".

Trait Implementations

impl Clone for QuantizationConfig

fn clone(&self) -> QuantizationConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for QuantizationConfig

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

Formats the value using the given formatter.