3j4SSKJr SSKJr SSKJr SSKJr SSKrSSK r SSK r SSjr SSSjjr SSjrSSS jjr\"S S 55r"S S 5r\"SS55r\"SS55r"SS5rg)) annotations) dataclass)Path)LiteralNcxUbUR5OSnUbUR5OSnURUX"X3S9$)N)outtypeftypeOUTTYPEFTYPE)lowerupperformat)filename output_typeftype_lowercaseftype_uppercases F/home/wildlama/miniconda3/lib/python3.13/site-packages/gguf/utility.pyfill_templated_filenamer sL2=2I;,,.rO2=2I;,,.rO ???#2#2  KKc US:aUS-nSnO#US:aUS-nSnOUS:aUS-nS nOUS -nS n[U[[[U55R S 55- S 5nUSUS3U3$)NgmBg-q=TgeAg& .>Bg.Agư>MgMbP?K0r.f)maxlenstrroundlstrip)model_params_count min_digitsscaled_model_params scale_suffixfixs r#model_weight_count_rounded_notationr)sD 058 c !047 c !047 147 j3s5)<#=>EEcJKKQ OC!!C5 *<. 99rcUS:a([[U5[U5-SS9nUSU3nU$[[U5SS9nU$)Nr)r%x)r)abs) total_params shared_params expert_params expert_count pretty_size size_classs r size_labelr4,sYa9#m:LsS`Oa:anop $~Q{m4  9\9JWXY rcbUb1UR5RSS5RSS5nO6Ub1UR5RSS5RSS5nOSnUbSU3OSnUb#SUR5RSS53OSn Ub#SUR5RSS53OSn Ub1SUR5RSS5R53OSn Ub#SUR5RSS53OSn UUU U U U 3$)N -/z ggml-modelr)stripreplacer) model_name base_namefinetune_stringversion_stringr4r model_typename parametersfinetuneversionencodingkinds rnaming_conventionrF7sJ ((c2::3D  !))#s3;;CE%/%;1ZL!JBQB]?((*223<=>ceH@N@Z.&&(00c:;<`bGFQF];$$&..sC8>>@ABceH9C9OQz!))#s34 5UWDVJ<z'8*TF CCrcR\rSrSr%S\S'S\S'S\S'S\S'S\S 'S S jrS rg ) RemoteTensorNr!dtypetuple[int, ...]shapeint offset_startsizeurlc~[[RURURUR S95nU$)N)rPstartrO) bytearraySafetensorRemoteget_data_by_rangerPrNrO)selfdatas rrWRemoteTensor.dataVs8);;PTPaPahlhqhq;rs rN)returnrS)__name__ __module__ __qualname____firstlineno____annotations__rW__static_attributes__rYrrrHrHNs" J  I HrrHc\rSrSrSrSr\S Sj5r\S Sj5r\SSj5r \SSSjj5r \SSj5r \SS j5r S r g )rT]a Uility class to handle remote safetensor files. This class is designed to work with Hugging Face model repositories. Example (one model has single safetensor file, the other has multiple): for model_id in ["ngxson/TEST-Tiny-Llama4", "Qwen/Qwen2.5-7B-Instruct"]: tensors = SafetensorRemote.get_list_tensors_hf_model(model_id) print(tensors) Example reading tensor data: tensors = SafetensorRemote.get_list_tensors_hf_model(model_id) for name, meta in tensors.items(): dtype, shape, offset_start, size, remote_safetensor_url = meta # read the tensor data data = SafetensorRemote.get_data_by_range(remote_safetensor_url, offset_start, size) print(data) zhttps://huggingface.cocURURSUS35nU(a#URSUS3nURU5$URSUS3nURU5nU(aURUS5nUR S5n[ R "U5nURS5cS5eUSn [[U R555n U R5 0n U HCn URSUSU 3nURU5R5H upXU 'M ME U $[S US 35e) z Get list of tensors from a Hugging Face model repository. Returns a dictionary of tensor names and their metadata. Each tensor is represented as a tuple of (dtype, shape, offset_start, size, remote_safetensor_url) r8z/resolve/main/model.safetensorsz*/resolve/main/model.safetensors.index.jsonrutf-8 weight_mapz"weight_map not found in index filez/resolve/main/z,No safetensor file has been found for model z.If the repo has safetensor files, make sure the model is public or you have a valid Hugging Face token set in the environment variable HF_TOKEN.)check_file_exist BASE_DOMAINget_list_tensorsrUdecodejsonloadsgetlistsetvaluessortitems ValueError)clsmodel_idis_single_filerP index_urlis_multiple_files index_data index_str index_jsonre all_filestensorsfilekeyvals rget_list_tensors_hf_model*SafetensorRemote.get_list_tensors_hf_modelrsr--0A8*Lk.lm __%Qxj0OPC'', ,'q 2\] 00; ..y!>,/; a=a a;#L1JS!2!2!456I NN /1G!)8*N4&I # 4 4S 9 ? ? AHC#&CL!B"N:8*EQ Q  rc URU5up#0nUR5HdupVUS:XaM [U[5(d[ SUSU35eUSnUSnUSupX- n X9-n [ U[ U5XUS9XE'Mf [[UR5S S 95nU$![an [ SUSU S U35eS n A ff=f) z Get list of tensors from a remote safetensor file. Returns a dictionary of tensor names and their metadata. Each tensor is represented as a tuple of (dtype, shape, offset_start, size) __metadata__zInvalid metadata for tensor 'z': rJrL data_offsets)rJrLrNrOrPz$Missing key in metadata for tensor 'z , meta = Nc US$NrrYts r3SafetensorRemote.get_list_tensors..sQqTrr~) get_metadatarq isinstancedictrrrHtupleKeyErrorsorted)rsrPmetadatadata_start_offsetresr@metarJrLoffset_start_relativeoffset_end_relativerOrNes rrh!SafetensorRemote.get_list_tensorss'*&6&6s&;#')"..*JD~%dD)) #@c$!PQQ eW W =A.=Q:%*B0H (uE%LWctwx +"6#))+>:;  e #GvSQRPSS\]a\b!cdd es0B22 C<CCcSnURUSU5n[U5S:a [S5e[R USSSS9nSU-n[U5SU-:a[SSU-S [U535eUSSU-nUR S 5n[ R"U5nX4$![ Ran [S U 35eSn A ff=f) zb Get JSON metadata from a remote safetensor file. Returns tuple of (metadata, data_start_offset) iPrz%Not enough data to read metadata sizeNlittle byteorder'Could not read complete metadata. Need bytes, got rdz-Failed to parse safetensor metadata as JSON: ) rUr rrrM from_bytesrirjrkJSONDecodeError) rsrP read_sizeraw_datametadata_lengthrmetadata_bytes metadata_strrrs rrSafetensorRemote.get_metadatas$ ((a; x=1 DE E.."1.J/ x=1. .Fq?GZF[[ghklthugvwx x"!A$78%,,W5  Rzz,/H. .## RLQCPQ Q RsB..CCCcZSSKnSSKJn U"U5nUR(aUR(d[ SU35eUR 5nUS:a SUSX#-3US'URUS US 9nUR5 UR[US:aU5$S5$) zh Get raw byte data from a remote file by range. If size is not specified, it will read the entire file. rNurlparse Invalid URL: zbytes=r7RangeTallow_redirectsheaders) requests urllib.parserschemenetlocrr_get_request_headersrlraise_for_statuscontentslice) rsrPrRrOrr parsed_urlrresponses rrU"SafetensorRemote.get_data_by_ranges )c]   (9(9}SE23 3**, "9!'wa ~>GG <<T7<K!!#dRid BCCT BCCrc>SSKnSSKJn U"U5nUR(aUR(d[ SU35eUR 5nSUS'URUSUS9nS URs=:*=(a S :$s $!URa g f=f) z\ Check if a file exists at the given URL. Returns True if the file exists, False otherwise. rNrrz bytes=0-0rTriF) rrrrrrrrhead status_codeRequestException)rsrPrrrrrs rrf!SafetensorRemote.check_file_exists )c]   (9(9}SE23 3 ..0G*GG }}S$}PH(..444 44 4((  sAB B BBcSS0n[RRS5(aS[RS3US'U$)z$Prepare common headers for requests.z User-Agentconvert_hf_to_ggufHF_TOKENzBearer Authorization)osenvironrl)rsrs rr%SafetensorRemote._get_request_headerssB !56 ::>>* % %)0J1G0H'IGO $rrYN)rtr!rZdict[str, RemoteTensor])rPr!rZr)rPr!rZztuple[dict, int])r)rPr!rRrMrOrMrZbytes)rPr!rZbool)rZzdict[str, str])r[r\r]r^__doc__rg classmethodrrhrrUrfrr`rYrrrTrT]s$+K& & P>RR@DD**rrTc4\rSrSr%S\S'S\S'S\S'Srg) LocalTensorRangeirrrMoffsetrOrYN)r[r\r]r^r_r`rYrrrrsN K Irrc>\rSrSr%S\S'S\S'S\S'S SjrS rg ) LocalTensorir!rJrKrLr data_rangec[R"URRSURRURR S9$)Nc)moderrL)npmemmaprrrrO)rVs r mmap_bytesLocalTensor.mmap_bytess8yy11DOODZDZbfbqbqbvbvwwrrYN)rZz np.ndarray)r[r\r]r^r_rr`rYrrrrs J   xrrc:\rSrSr%SrS\S'S SjrSrSrSr g ) SafetensorsLocali"z Read a safetensors file from the local filesystem. Custom parsing gives a bit more control over the memory usage. The official safetensors library doesn't expose file ranges. zdict[str, LocalTensor]r|c [US5n[RURS5SS9n[R "U5R nUSU-:a[SSU-SU35eURU5RS5n[R"U5nUR5n0n UR5HKupU S :XaM [U S [U S 5[!UXS S -U S SU S S - 5S9X'MM [#[%U R5SS95UlSSS5 g![Ran[SU35eSnAff=f!,(df  g=f)Nrbrrrrrrdz.Failed to parse safetensors metadata as JSON: rrJrLrr)rJrLrc US$rrYrs rr+SafetensorsLocal.__init__..Msadrr)openrMrreadrstatst_sizerrrirjrkrtellrqrrrrrr|) rVrrr file_sizerrrrr|r@rs r__init__SafetensorsLocal.__init__,s{ (D !Q!nnQVVAY(nKO)11I1.. #J1K^J__klukv!wxx66/299'BL W::l3!" .0G&nn. >) +w-W ./ ),@,CC^,Q/$~2Fq2II ! /" w}}N KLDLA" !'' W #QRSQT!UVV W" !s1BE0E%BE0E-E((E--E00 E>cAAUR$Nr|rVargskwargss r __enter__SafetensorsLocal.__enter__Os &||rcAAgrrYrs r__exit__SafetensorsLocal.__exit__Ss &rrN)rr) r[r\r]r^rr_rrrr`rYrrrr"s $#!MFrr)rr!r str | NonerZr!)r+)r$rMr%rMrZr!) r.rMr/rMr0rMr1rMrZr!r)r;rr<rr=rr>rr4rrrr?zLiteral['vocab', 'LoRA'] | NonerZr!) __future__r dataclassesrpathlibrtypingrrrjnumpyrrr)r4rFrHrTrrrrYrrrs"! K:.D.     qqh    xx x22r