64c05eecd6
* higgs init * working with crunches * per-model workspaces * style * style 2 * tests and style * higgs tests passing * protecting torch import * removed torch.Tensor type annotations * torch.nn.Module inheritance fix maybe * hide inputs inside quantizer calls * style structure something * Update src/transformers/quantizers/quantizer_higgs.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * reworked num_sms * Update src/transformers/integrations/higgs.py Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * revamped device checks * docstring upd * Update src/transformers/quantizers/quantizer_higgs.py Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com> * edited tests and device map assertions * minor edits * updated flute cuda version in docker * Added p=1 and 2,3bit HIGGS * flute version check update * incorporated `modules_to_not_convert` * less hardcoding * Fixed comment * Added docs * Fixed gemma support * example in docs * fixed torch_dtype for HIGGS * Update docs/source/en/quantization/higgs.md Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> * Collection link * dequantize interface * newer flute version, torch.compile support * unittest message fix * docs update compile * isort * ValueError instead of assert --------- Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com> Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com>
2.0 KiB
Executable File
Quantization
Quantization techniques reduce memory and computational costs by representing weights and activations with lower-precision data types like 8-bit integers (int8). This enables loading larger models you normally wouldn't be able to fit into memory, and speeding up inference. Transformers supports the AWQ and GPTQ quantization algorithms and it supports 8-bit and 4-bit quantization with bitsandbytes.
Quantization techniques that aren't supported in Transformers can be added with the [HfQuantizer] class.
Learn how to quantize models in the Quantization guide.
QuantoConfig
autodoc QuantoConfig
AqlmConfig
autodoc AqlmConfig
VptqConfig
autodoc VptqConfig
AwqConfig
autodoc AwqConfig
EetqConfig
autodoc EetqConfig
GPTQConfig
autodoc GPTQConfig
BitsAndBytesConfig
autodoc BitsAndBytesConfig
HfQuantizer
autodoc quantizers.base.HfQuantizer
HiggsConfig
autodoc HiggsConfig
HqqConfig
autodoc HqqConfig
FbgemmFp8Config
autodoc FbgemmFp8Config
CompressedTensorsConfig
autodoc CompressedTensorsConfig
TorchAoConfig
autodoc TorchAoConfig
BitNetConfig
autodoc BitNetConfig