o d۷i@sNddlmZddlmZddlmZddgZGdddeZGdddZdS) )Enum)delete_offload_parameter)ModuleQuantizationMetadataKVCacheScaleTypec@seZdZdZdZdS)rk_scalev_scaleN)__name__ __module__ __qualname__KEYVALUErrd/home/ubuntu/vllm_env/lib/python3.10/site-packages/compressed_tensors/quantization/quant_metadata.pyrsc@s.eZdZdZeddZedefddZdS)rz> Container class for metadata related to quantization cCstjjtjjgdddDS)z All quantization parameter names that might be registered onto a module during lifecycle (excluding serialized parameters) cSs$g|]}dD] }|d|qqS)) global_scalescale zero_pointg_idx_r).0 base_namesuffixrrr (s z9QuantizationMetadata.all_qparam_names..)inputweightoutput)rr valuer rrrrall_qparam_names"sz%QuantizationMetadata.all_qparam_namesmodulecCs&|D] }t||rt||qdS)z Remove all parameters related to quantization that might have been registered onto a module previously in lifecycle (excluding serialized parameters) :param module: Module to clear N)rhasattrr)clsrkeyrrrclear_all_qparams3s  z&QuantizationMetadata.clear_all_qparamsN) r r r __doc__ staticmethodr classmethodrr"rrrrrs  N) enumrcompressed_tensors.utilsrtorch.nnr__all__rrrrrrs