Skip to content

Latest commit

 

History

History
85 lines (65 loc) · 3.24 KB

quantization.md

File metadata and controls

85 lines (65 loc) · 3.24 KB
Raw

Quantization and Sparsity

torchtune integrates with torchao for QAT and QLoRA. Currently only some quantization techniques are integrated, see the docstrings in the quantization recipe and the QAT recipe for more details.

For post training quantization, we recommend using torchao directly: https://github.com/pytorch/ao/blob/main/torchao/quantization/README.md to quantize their model and do eval/benchmark in torchao as well: https://github.com/pytorch/ao/tree/main/torchao/_models/llama.

Quantization-Aware Training (QAT)

(PyTorch 2.4+)

QAT refers to applying fake quantization to weights and/or activations during finetuning, which means simulating only the quantization math without actually casting the original dtype to a lower precision. You can run QAT with finetuning using the following command:

tune run --nproc_per_node 4 qat_distributed --config llama3/8B_qat_full

This produces an unquantized model in the original data type. To get an actual quantized model, follow this with tune run quantize while specifying the same quantizer in the config, e.g.

# QAT specific args
quantizer:
  _component_: torchtune.training.quantization.Int8DynActInt4WeightQATQuantizer
  groupsize: 256

Currently only torchtune.training.quantization.Int8DynActInt4WeightQATQuantizer is supported. This refers to int8 dynamic per token activation quantization combined with int4 grouped per axis weight quantization. For more details, please refer to the torchao implementation.

Eval

To evaluate a quantized model, make the following changes to the default evaluation config

# Currently we only support torchtune checkpoints when
# evaluating quantized models. For more details on checkpointing see
# https://pytorch.org/torchtune/main/deep_dives/checkpointer.html
# Make sure to change the default checkpointer component
checkpointer:
  _component_: torchtune.training.FullModelTorchTuneCheckpointer
  ..
  checkpoint_files: [<quantized_model_checkpoint>]

# Quantization specific args
quantizer:
  _component_: torchtune.training.quantization.Int8DynActInt4WeightQuantizer
  groupsize: 256

Noet: we can use Int8DynActInt4WeightQuantizer to load a QAT quantized model since it's the same type of quantization.

and run evaluation:

tune run eleuther_eval --config eleuther_evaluation

Generate

To run inference using a quantized model, make the following changes to the default generation config

# Currently we only support torchtune checkpoints when
# evaluating quantized models. For more details on checkpointing see
# https://pytorch.org/torchtune/main/deep_dives/checkpointer.html
# Make sure to change the default checkpointer component
checkpointer:
  _component_: torchtune.training.FullModelTorchTuneCheckpointer
  ..
  checkpoint_files: [<quantized_model_checkpoint>]

# Quantization Arguments
quantizer:
  _component_: torchtune.training.quantization.Int8DynActInt4WeightQuantizer
  groupsize: 256

and run generation:

tune run generate --config generation