Module validate 

Tensor-level (not full-model) quantization error metrics: for each quantized tensor, dequantize and compare against the original f32 weights. No forward pass is needed – this is a pure diff between the original and quantized-then-reconstructed safetensors files.

Structs

TensorReport

Per-tensor quantization error metrics.

Functions

compare_tensors

Compares original against reconstructed element-wise, computing TensorReport metrics. Both slices must be the same length (the caller is responsible for shape bookkeeping).

validate_checkpoint

Compares every quantized tensor in quantized_path (as recorded by its embedded quantization_config) against the corresponding tensor in original_path, returning one TensorReport per quantized tensor, in the checkpoint’s tensor order.