a  duF @sddlZddlZddlZddlmZddlmZmZmZm Z m Z ddl m Z ddl mZddlmZmZddlmZddlmZdd lmZmZmZmZdd lmZdd lmZdd lm Z dd l!m"Z"e#e$Z%da&da'e(ej)*dddkZ+gdZ,ddZ-d,ddZ.d-ddZ/d.e j0eeeedddZ1d/e j0eee(e(eee2dddZ3d d!Z4d"d#Z5d$d%Z6d0e7ed&d'd(Z8d1ee7e2eeeeee eee2eeee e7d) d*d+Z9dS)2N)deepcopy)OptionalDictCallableAnyTuple)nn)load_state_dict_from_url)FeatureListNetFeatureHookNet)FeatureGraphNet)load_state_dict) has_hf_hubdownload_cached_filecheck_cached_fileload_state_dict_from_hf)adapt_input_conv) PretrainedCfg)adapt_model_from_file)get_pretrained_cfgFZTIMM_USE_OLD_CACHE) set_pretrained_download_progressset_pretrained_check_hashload_custom_pretrainedload_pretrainedpretrained_cfg_for_featuresresolve_pretrained_cfgbuild_model_with_cfgc Cs|dd}|dd}|dd}|dd}|dd}d}d}|dkrhtdd rhd}|sbJ|}nj|rd}|}t|tsJnN|rd}|}n@d }tr|rt|nd }|s|rtdd rd}|}n |rd}|}|dkr|d dr||d f}||fS) Nsourceurlfile state_dict hf_hub_idhf-hubT)Z necessaryFZhf_hub_filename)getr isinstancedict_USE_OLD_CACHEr) pretrained_cfgZ cfg_sourceZpretrained_urlZpretrained_fileZ pretrained_sdr" load_frompretrained_locZold_cache_validr+]/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/timm/models/_builder.py_resolve_pretrained_sources<      r-TcCs|adS)zA Set download progress for pretrained weights on/off (globally). N)_DOWNLOAD_PROGRESSenabler+r+r,rLsrcCs|adS)z= Set hash checking for pretrained weights on/off (globally). N) _CHECK_HASHr/r+r+r,rRsr)modelr(load_fncCs|pt|dd}|s"tddSt|\}}|s@tddS|dkrTtdn|dkrjt|ttd}|dur~|||n t|d r||n td dS) aLoads a custom (read non .pth) weight file Downloads checkpoint file into cache-dir like torch.hub based loaders, but calls a passed in custom load fun, or the `load_pretrained` model member fn. If the object is already present in `model_dir`, it's deserialized and returned. The default value of `model_dir` is ``/checkpoints`` where `hub_dir` is the directory returned by :func:`~torch.hub.get_dir`. Args: model: The instantiated model to load weights into pretrained_cfg (dict): Default pretrained model cfg load_fn: An external standalone fn that loads weights into provided model, otherwise a fn named 'laod_pretrained' on the model will be called if it exists r(Nz/Invalid pretrained config, cannot load weights.zHNo pretrained weights exist for this model. Using random initialization.r#zKHugging Face hub not currently supported for custom load pretrained models.r) check_hashprogressrzXValid function to load pretrained weights is not available, using random initialization.) getattr_loggerwarningr-rr1r.hasattrr)r2r(r3r)r*r+r+r,rXs*       r)r2r( num_classesin_chans filter_fnstrictc Cs|pt|dd}|stdt|\}}|dkr@td|}n|dkrdtd|dt|}n|d krtd |d|d d rt|tt d }| |dSt |dtt d}nX|dkrtd|dt |t tfrt|}nt|}n|dd} td| d|durXz|||}Wn.tyV} z||}WYd} ~ n d} ~ 00|dd} | dur|dkrt | tr| f} | D]} | d} z.t||| || <td| d|dWnBty } z(|| =d }td| dWYd} ~ n d} ~ 00q|dd}|dd }|durt |trD|f}||d!kr|D]&}||dd||d"dqVd }nV|d kr|D]F}||d}||d||d<||d"}||d||d"<q|j||d#dS)$a Load pretrained checkpoint Args: model (nn.Module) : PyTorch model module pretrained_cfg (Optional[Dict]): configuration for pretrained weights / target dataset num_classes (int): num_classes for target model in_chans (int): in_chans for target model filter_fn (Optional[Callable]): state_dict filter fn for load (takes state_dict, model as args) strict (bool): strict load of checkpoint r(NzWInvalid pretrained config, cannot load weights. Use `pretrained=False` for random init.r!z*Loading pretrained weights from state dictr z&Loading pretrained weights from file ()rz%Loading pretrained weights from url (Z custom_loadF)r5r4cpu)Z map_locationr5r4r#z2Loading pretrained weights from Hugging Face hub ( architecturez this modelz No pretrained weights exist for z). Use `pretrained=False` for random init.Z first_convr;z.weightzConverted input conv z pretrained weights from 3 to z channel(s)zUnable to convert pretrained z+ weights, using random init for this layer. classifier label_offsetrr<z.bias)r?)r6 RuntimeErrorr-r7infor r$rr.r1rr r%listtupler TypeErrorstrrNotImplementedErrorr8pop)r2r(r<r=r>r?r)r*r!Z model_nameeZ input_convsZinput_conv_nameZ weight_name classifiersrDZclassifier_nameZclassifier_weightZclassifier_biasr+r+r,rs                   rcCs&t|}d}|D]}||dq|S)N)r<rC global_pool)rrL)r(Z to_removetrr+r+r,rs rcCs&|r|s dS|D]}||dqdS)N)rL)kwargsnamesnr+r+r,_filter_kwargssrTcCsd}|ddr|d7}|D]}|dkrb|dd}|durt|dksLJ|||d dq|d kr|dd}|durt|dksJ|||d q||d}|dur||||qt||d dS) a& Update the default_cfg and kwargs before passing to model Args: pretrained_cfg: input pretrained cfg (updated in-place) kwargs: keyword args passed to model build fn (updated in-place) kwargs_filter: keyword arg keys that must be removed before model __init__ )r<rOr=Zfixed_input_sizeF)img_sizerU input_sizeNr;r=r)rR)r$len setdefaultrT)r(rQ kwargs_filterZdefault_kwarg_namesrSrVZ default_valr+r+r,_update_default_kwargss$     r[)variantreturncCs|}d}|r8t|tr&tfi|}nt|tr8|}d}|sV|rNd||g}t|}|srtd|dt}|pxi}|js| d|t j |fi|}|S)N.z*No pretrained configuration specified for ze model. Using a default. Please add a config to the model pretrained_cfg registry or pass explicitly.rB) r%r&rrJjoinrr7r8rBrY dataclassesreplace)r\r(pretrained_cfg_overlayZmodel_with_tagZpretrained_tagr+r+r,r s,    r) model_clsr\ pretrainedr(rb model_cfg feature_cfgpretrained_strictpretrained_filter_fnrZc Ks| dd} d} |pi}t|||d}|}t|| | | ddrld} |ddd| vrl| d|d<|dur|fi| } n|fd |i| } || _| j| _| rt| |} | rd nt| d | d d }|rt | ||| d d||d| rt }t| dd}|dur|d|d|vrv|d}t |t rv|}d|vrRt}n$|dkrbt}ndsvJd||| fi|} t|| _| j| _| S)a Build model with specified default_cfg and optional model_cfg This helper fn aids in the construction of a model including: * handling default_cfg and associated pretrained weight loading * passing through optional model_cfg for models with config based arch spec * features_only model adaptation * pruning config / model adaptation Args: model_cls (nn.Module): model class variant (str): model variant name pretrained (bool): load pretrained weights pretrained_cfg (dict): model's pretrained weight/task config model_cfg (Optional[Dict]): model's architecture config feature_cfg (Optional[Dict]: feature extraction adapter config pretrained_strict (bool): load pretrained weights strictly pretrained_filter_fn (Optional[Callable]): filter callable for pretrained weights kwargs_filter (Optional[Tuple]): kwargs to filter before passing to model **kwargs: model args passed through to model __init__ prunedF)r(rbZ features_onlyTZ out_indices)rr;Ncfgrr<r:r=r;)r(r<r=r>r? output_fmt feature_clshookZfxzUnknown feature class )rLrto_dictr[rYr(Z default_cfgrr6r$rr r%rJlowerr r r)rcr\rdr(rbrerfrgrhrZrQrifeaturesr2Znum_classes_pretrainedrornr+r+r,rCsd!                r)T)T)NN)Nr:r;NT)NN)NNNNTNN):r`loggingoscopyrtypingrrrrrZtorchrZ torch.hubr Ztimm.models._featuresr r Ztimm.models._features_fxr Ztimm.models._helpersr Ztimm.models._hubrrrrZtimm.models._manipulaterZtimm.models._pretrainedrZtimm.models._prunerZtimm.models._registryr getLogger__name__r7r.r1intenvironr$r'__all__r-rrModulerboolrrrTr[rJrrr+r+r+r,s          -   0 f & '