Quanto

New features

- group-wise quantization,
- safe serialization.

New features

- new `QConv2d` quantized module,
- official support for `float8` weights.

Bug fixes

- fix `QbitsTensor.to()` that was not moving the inner tensors,
- prevent shallow `QTensor` copies when loading weights that do not move inner tensors.

New features

- quanto kernels library (not used for now in quantize).

Breaking changes

- quantization types are now all quanto.dtype

New features:

- support int2 and int4 weights.

New contributors:

younesbelkada
a-r-r-o-w

New features:

- calibration streamline option to remove spurious quantize/dequantize,
- calibration debug mode.

New features:

- quantize weights and activations parameters
- float8 activations