o Gi @sLddlmZddlmZmZeeZerddlm Z d de fddZ dS) N)is_accelerate_availablelogging)init_empty_weightsFmodules_to_not_convertcsddlmm}mmmmdtffdd fdd|||}tfdd | D}|sCt |j j d |rI|||S) Nr)QLinearfreezeqfloat8qint2qint4qint8dtypecsd|S)N)float8int8int4int2)r )r r r r rU/home/ubuntu/.local/lib/python3.10/site-packages/diffusers/quantizers/quanto/utils.py_get_weight_typesz5_replace_with_quanto_layers.._get_weight_typec st|}|s |S|D]P\}}|||||vrqt|tjr^t0|j|j|j du|j j |j d}||j |<t||j |_|j |dWdn1sYwYq|S)N) in_features out_featuresbiasr weightsF)listchildrennamed_children isinstancennLinearrrrrweightr weights_dtype_modulestype source_clsrequires_grad_)modelquantization_configr has_childrennamemoduleqlinear)rr_replace_layersrrr+s,     z4_replace_with_quanto_layers.._replace_layersc3s|] \}}t|VqdS)N)r).0_replaced_module)rrr .sz._replace_with_quanto_layers..z does not appear to have any `nn.Linear` modules. Quantization will not be applied. Please check your model architecture, or submit an issue on Github if you think this is a bug. https://github.com/huggingface/diffusers/issues/new)optimum.quantorrr r r r strany named_modulesloggerwarning __class____name__)r%r&r pre_quantizedrhas_been_replacedr)rrr+r r r r r_replace_with_quanto_layers s   r:)F) torch.nnrutilsrr get_loggerr7r4 acceleraterrr:rrrrs