Types & classes — github.com/Vahe1994/AQLM

CodeHub MCPcopy Create free account

hub / github.com/Vahe1994/AQLM / types & classes

Types & classes13 in github.com/Vahe1994/AQLM

⨍ Functions 225◇ Types & classes 13 ↳ Endpoints 1

↓ 3 callersClassQuantizedLinear

↓ 1 callersClassAQEngine

A wrapper class that runs AQ training for a single linear layer. All the important math is in aq.py

aq_engine.py:16

↓ 1 callersClassCatcher

↓ 1 callersClassCatcherExit

↓ 1 callersClassConfigurableAdamW

r""" A version of Adam optimizer that supports custom parameter dtypes, amsgrad, lamb or rmsprop on per-group basis. Adam and Amsgrad based on

src/configurable_adam.py:12

↓ 1 callersClassIntCodes

A storage for integer codes that makes them compatible with FullyShardedDataParallel, see https://github.com/pytorch/pytorch/issues/123528 fo

src/utils.py:155

↓ 1 callersClassQuantizedWeight

↓ 1 callersClassStraightThroughAdamW

A wrapper for a PyTorch optimizer that can perform updates on quantized and/or de-quantized parameters :param update_non_quantized_params: ho

src/pv_optimizer.py:22

↓ 1 callersClassYourQuantizedWeightIsInAnotherRank

This replaces quantized weights that are not held on this rank

src/pv_utils.py:200

↓ 1 callersClass_LayerWrapperThatAccumulatesXTX

ClassParameterRole

src/pv_optimizer.py:16

ClassQuantizedLinear

inference_lib/src/aqlm/inference.py:11

Class_QuantizedMatmul

inference_lib/src/aqlm/inference.py:100