sg @ddlmZddlZddlZddlZddlmZmZmZm Z m Z ddl m Z m Z mZe rddl mZmZmZmZmZmZmZdZej.j1dZeGej4ej6j9ej6j;edz nej4eZe eeZ ee Z!dZ"d Z#d Z$d Z%d Z&d Z'dZ(dZ)dZ*dZ+dZ,dZ-dZ.dZ/dZ0dZ1dZ2dZ3dZ4dZ5dZ6dZ7dZ8dZ9d Z:d!Z;d"Zej~ejejejejZBej~ejejZCge j_Eeje j_Ge jZHd%ZId&ZJd'ZKd(ZLd)ZMeLZNdZOeMZPd ZQed*eRZSejZTed+eRZUejZVejZXejZYejeYZ[ejZ\dZ] d Z^ d Z_ d Z` d,Za d-Zb dZcd Zdd Zed Zfd,Zgd-Zhd Zid ZjdZkdZldZmdZndZodZpdZqdZrdZsdZtdZudZvdZwdZxdZydZzdZ{dZ|d Z}d!Z~d&ZdZd xZZdZd Zd Zd Zd,Zd-Zd ZdZd Zd Zd,ZdZdZd.Zd/Zd0Zd1Zd2ZdZd Zd Zd Zd ZdZdZdZdZdZdZdZdZdZdZdZdZdZdZdZdZdZdZd Zd!Zd"Zd#Zd$Zd3Zd.Zd4Zd5Zd6Zd&ZdZd Zd Zd ZeZd&ZdZd Zd Zd Zd,Zd&ZdZd ZdZd Zd ZGd7d8ejZejeɫZGd9d:ejZeje˫Zej~ejejejZGd;dejZGd?d@ejZGdAdBejZGdCdDejZej~dejejejZ GdEdFejZGdGdHejZejeܫZGdIdJejZGdKdLejZejZejejZe!dMge֫ddNZe!dOge׫ddPZe!dQgeޫddRZe!dSge۫ddTZe!dUgddVZdZd Zd Zd Zd,Zd-Ze!dWejgdddXZe!dYgddZZe!d[ejegeT dd\Ze!d]eTgddd^Ze!d_eTegeV dd`Ze!daeVgdddbZe!dcgejdddZe!degejddfZDe!dggejddhZe!digejddjZe!dkgejddlZe!dmgejddnZe!doeVgejddpZe!dqeVgejddrZe!dseVgejddtZe!dueVgejddvZe!dweTgejddxZe!dyeTgejddzZe!d{eTgejdd|Ze!d}eTgejdd~Ze!deTgejddZe!deVgeTddZe!deVgejddZe!deTgejddZe!deTgejddZe!deTgejddZe!deTejejejgej ddZe!deTgejddZ e!deTejejejgej ddZ e!deTejejejgej ddZ e!deTejejgej ddZ e!deTgejddZe!deTgejddZe!deTejgej ddZe!deTgejddZe!deTgejddZe!deTgejddZe!deTgejddZe!dejejeje۫gej ddZe!deTejge ddZe!deVeejgej ddZe!deVegej ddZe!deVgdddZe!degdddZe!deVejejejejejejgej ddZGddejZGddejZejeZe!deVejge ddZe!degdddZ e!deVegdddZ!e!deVgejddZ"e!deVgejddZ#e!deVgdddZ$e!deVe\eXeXgej ddÄZ%e!deVe\e\eXeXgd ddńZ&e!deVe\gdddDŽZ'e!deVe\eXeXeXgd ddɄZ(e!deVe\eXeXejgd dd˄Z)e!deVgddd̈́Z*e!deVgdddτZ+e!deVgejddфZ,e!deVgejddӄZ-e!deVgejddՄZ.e!deVejej^ejgej ddׄZ0e!deVejej^gej ddلZ1e!deVejej^ejgej ddۄZ2e!deVejej^gej dd݄Z3e!deVeje[ejejejgej dd߄Z4e!deVeje[ejejejgej ddZ5e!deVeje[ejgej ddZ6e!deVeje[ejgej ddZ7e!deVe\gejddZ8e!deVejej^eje\gej ddZ9e!deVejej^eje\gej ddZ:e!deVeje\e[ejgej ddZ;e!deVeje\e[ejejejgej ddZ<e!de[ejgeϫ ddZ=e!dejejejgeϫ ddZ>e!degdddZ?e!deVegejddZ@e!deVegejddZAe!deVejejgd ddZBe!deVgejddZCe!deVgejddZDe!deVejgddڐdZEe!deVejgddېdZFe!deVeCejgd dܐdZGe!deVgdddZHe!deVgejejdݐd ZIe!d eVejgejej dސd ZJe!d eVgejejdݐd ZKe!deVejgejej dސdZLe!deVe\gejej dߐdZMe!deTeYgej ddZNe!deTeYgej ddZOe!deTeYgej ddZPe!deTeYgejddZQe!deTeYgej ddZRe!deTgeYddZSe!deTgeYddZTe!d eTgeYdd!ZUe!d"eTgeYdd#ZVe!d$eTgeYdd%ZWe!d&eTgeYdd'ZXe!d(eTgejdd)ZYe!d*eTgejdd+ZZe!d,eTgeYdd-Z[e!d.eTgeYdd/Z\e!d0eTgeYdd1Z]e!d2eTgeYdd3Z^e!d4eTgeYdd5Z_e!d6eTgeYdd7Z`e!d8eTgeYdd9Zae!d:eTgeYdd;Zbe!deTejeje[ejejejgej dd?Zde!d@eTeYejejejejgej ddAZee!dBeTejeYejejejejejgej ddCZfe!dDejejeje߫ejgej ddEZge!dFejejejgej ddGZhejZiGdHdIejZjGdJdKejZke reekZlejekZmej~ejemZnej~demeYZoej~deme̫Zpej~demZqej~ememZrej~demZsdenfdLeofdMepfdNeqfdOerfdPesfgej_te!dQemgejddRZue!dSemeYgdddTZve!dUemegd ddVZwe!dWemgdddXZxe!dYemgemddZZye!d[emgddd\Zze!d]egemdd^Z{e!d_ememgddd`Z|e!daemejgem ddbZ}e!dcemgejdddZ~e!deemejgem ddfZe!dggemddhZe!diejgemddjZe!dkgemddlZe!dmejgemddnZe!doejejgemddpZe!dqejejgemddrZe!dsejejgemddtZe!duejgemddvZe!dwejejejgem ddxZe!dyejejejejgem ddzZe!d{ejejejejejgem dd|Ze!d}ejejejgem dd~Ze!deTejejgem ddZe!dejeYeYejejejejejejg em ddZe!deTejejejejejejejgem ddZe!dejejegem ddZe!deTgemddZe!demgejddZe!demeVejgeY ddZe!dejejejejejgej ddZe!dejejejejejgej ddZe!dgejddZe!dejejgd ddZGddejZGddejZe!deVgeddZe!deVgdddZe!deVgdddZe!demgeddZe!demgdddZe!demgdddZy( ) annotationsN)CallableUnionNewTypeOptional TYPE_CHECKING)load_shared_librarybyref"ctypes_function_for_shared_library) CtypesCData CtypesArray CtypesPointerCtypesVoidPointer CtypesRefCtypesPointerOrRefCtypesFuncPointerllamaLLAMA_CPP_LIB_PATHlib lialgginsggiqsgg llama_model_pllama_context_p @i$%icreZdZUdZerded<ded<ded<defdejfdejfgZ y)llama_token_datazUsed to store token data Attributes: id (llama_token): token id logit (float): log-odds of the token p (float): probability of the token llama_tokenidfloatlogitpN __name__ __module__ __qualname____doc__r__annotations__r@ctypesc_float_fields_F/var/www/html/venv/lib/python3.12/site-packages/llama_cpp/llama_cpp.pyr?r?sD/   { &..! fnnHrOr?ceZdZUdZerded<ded<ded<ded<defdejfdejfdejfgZ y ) llama_token_data_arrayaUsed to sample tokens given logits Attributes: data (ctypes.Array[llama_token_data]): token data size (int): size of the array selected (int): index in the data array (i.e. not the token id) sorted (bool): whether the array is sortedzCtypesArray[llama_token_data]dataintsizeselectedboolsortedN) rFrGrHrIrrJllama_token_data_prKc_size_tc_int64c_boolrMrNrOrPrRrRsW6++    #$ ! V^^$ 6==! HrOrRc eZdZUdZer#ded<ded<ded<ded <d ed <d ed <ded<dejfdeje fdejejfd eje fd ejejfd ejeje fdejejfgZy) llama_batcha#Input data for llama_decode A llama_batch object can contain input about one or many sequences The provided arrays (i.e. token, embd, pos, etc.) must have size of n_tokens Attributes: n_tokens (int): number of tokens token (ctypes.Array[llama_token]): the token ids of the input (used when embd is NULL) embd (ctypes.Array[ctypes.ctypes.c_float]): token embeddings (i.e. float vector of size n_embd) (used when token is NULL) pos (ctypes.Array[ctypes.Array[llama_pos]]): the positions of the respective token in the sequence seq_id (ctypes.Array[ctypes.Array[llama_seq_id]]): the sequence to which the respective token belongs logits (ctypes.Array[ctypes.ctypes.c_int8]): if zero, the logits for the respective token will not be output rTn_tokensCtypesArray[llama_token]tokenCtypesArray[ctypes.c_float]embdz#CtypesArray[CtypesArray[llama_pos]]poszCtypesArray[ctypes.c_int]n_seq_idz&CtypesArray[CtypesArray[llama_seq_id]]seq_idzCtypesArray[ctypes.c_int8]logitsN)rFrGrHrIrrJrKc_int32POINTERr@rL llama_pos llama_seq_idc_int8rMrNrOrPr^r^s  '')) 00++66** V^^$ .&..-. /0 y)* ^V^^FNN34 >6>>.&..">?@ >6>>&--01HrOr^ceZdZUdejfdej fdej fdejdzfgZe rde d<de d<de d<d e d<y y ) llama_model_kv_override_valueval_i64val_f64val_boolval_strr9rTrBrWbytesN) rFrGrHrKr[c_doubler\c_charrMrrJrNrOrPrnrnQs\ FNN# FOO$ V]]# FMMC'( H  rOrncveZdZUdejfdej dzfdefgZerde d<de d<de d<yy) llama_model_kv_overridetagkeyr9valuerTrszUnion[int, float, bool, bytes]N) rFrGrHrKc_intrurnrMrrJrNrOrPrwrw`sI    #$ /0H    --rOrwceZdZUdZer thold, < 0 disabled (default) cb_eval (ggml_backend_sched_eval_callback): callback for scheduling eval cb_eval_user_data (ctypes.ctypes.c_void_p): user data for cb_eval type_k (int): data type for K cache type_v (int): data type for V cache logits_all (bool): the llama_decode() call computes all logits, not just the last one (DEPRECATED - set llama_batch.logits instead) embeddings (bool): if true, extract embeddings (together with logits) offload_kqv (bool): whether to offload the KQV ops (including the KV cache) to GPU flash_attn (bool): whether to use flash attention abort_callback (ggml_abort_callback): abort callback if it returns true, execution of llama_decode() will be aborted abort_callback_data (ctypes.ctypes.c_void_p): data for abort_callback rTn_ctxn_batchn_ubatch n_seq_max n_threadsn_threads_batchrope_scaling_type pooling_typeattention_typerBrope_freq_baserope_freq_scaleyarn_ext_factoryarn_attn_factoryarn_beta_fastyarn_beta_slow yarn_orig_ctx defrag_tholdz'Callable[[ctypes.c_void_p, bool], bool]cb_evalrcb_eval_user_datatype_ktype_vrW logits_all embeddings offload_kqv flash_attnz!Callable[[ctypes.c_void_p], bool]abort_callbackabort_callback_dataN)rFrGrHrIrrJrKc_uint32rhr{rL ggml_backend_sched_eval_callbackrr\ggml_abort_callbackrMrNrOrPrrs@   88**  99,, &//" FOO$ V__% foo& fnn% FNN+ fll+ & 6<<( 6>>* FNN+ FNN+ V^^, 6>>* 6>>* &//* ( 45 foo. 6<< 6<< v}}% v}}%  & v}}% ./ 07HrOrc eZdZUdZer7ded<ded<ded<ded<ded<ded <ded <ded <ded <d ed<d ed<dejfdejfdejfdejfdejfd ejfd ejfd ejfd ejfdejfdejfg Z y)llama_model_quantize_paramsaYParameters for llama_model_quantize Attributes: nthread (int): number of threads to use for quantizing, if <=0 will use std::thread::hardware_concurrency() ftype (int): quantize to this llama_ftype output_tensor_type (int): output tensor type token_embedding_type (int): token embeddings tensor type allow_requantize (bool): allow quantizing non-f32/f16 tensors quantize_output_tensor (bool): quantize output.weight only_copy (bool): only copy tensors - ftype, allow_requantize and quantize_output_tensor are ignored pure (bool): quantize all tensors to the default type keep_split (bool): quantize to the same number of shards imatrix (ctypes.c_void_p): pointer to importance matrix data kv_overrides (ctypes.c_void_p): pointer to vector containing overrides rTnthreadftypeoutput_tensor_typetoken_embedding_typerWallow_requantizequantize_output_tensor only_copypure keep_splitrimatrixrN) rFrGrHrIrrJrKrhr{r\rrMrNrOrPrris   !! $$   %% FNN# &,, v||, . V]]+ !6==1 fmm$  v}}% FOO$ ) HrOrcNeZdZUdZer ded<ded<defdejfgZ y)llama_logit_biaszjUsed to store logit bias Attributes: token (llama_token): token id bias (float): biasr@rarBbiasNrErNrOrPrrs4   +  HrOrc>eZdZUdZerded<dejfgZy)llama_sampler_chain_paramszrParameters for llama_sampler_chain Attributes: no_perf (bool): whether to measure performance timingsrWno_perfN) rFrGrHrIrrJrKr\rMrNrOrPrrs'B   FMM"HrOrcDeZdZdejfdejfgZy)llama_chat_messagerolecontentN)rFrGrHrKrrMrNrOrPrrs! ! FOO$HrOrllama_model_default_paramscy)z&Get default parameters for llama_modelNrNrNrOrPrrrOllama_context_default_paramscy)z(Get default parameters for llama_contextNrNrNrOrPrrrrO"llama_sampler_chain_default_paramscy)z.Get default parameters for llama_sampler_chainNrNrNrOrPrrrrO#llama_model_quantize_default_paramscy)z/Get default parameters for llama_model_quantizeNrNrNrOrPrrrrOllama_backend_initcy)zyInitialize the llama + ggml backend If numa is true, use NUMA optimizations Call once at the start of the programNrNrNrOrPrrrOllama_numa_initcyNrN)numas rPrr" rOllama_backend_freecy)zACall once at the end of the program - currently only used for MPINrNrNrOrPrr7rrOllama_load_model_from_filecyrrN) path_modelparamss rPrrDrOllama_free_modelcyrrNmodels rPrrPrrOllama_new_context_with_modelcyrrN)rrs rPrr\rrO llama_freecy)zFrees all allocated memoryNrNctxs rPrrirrO llama_time_uscyrrNrNrOrPrrtrrOllama_max_devicescyrrNrNrOrPrr~rOllama_supports_mmapcyrrNrNrOrPrrrrOllama_supports_mlockcyrrNrNrOrPrrrrOllama_supports_gpu_offloadcyrrNrNrOrPrrrrOllama_supports_rpccyrrNrNrOrPrrrrO llama_n_ctxcyrrNrs rPrrrrO llama_n_batchcyrrNrs rPrrrrOllama_n_ubatchcyrrNrs rPrrrrOllama_n_seq_maxcyrrNrs rPrrrrO llama_n_vocabcyrrNrs rPrrrrOllama_n_ctx_traincyrrNrs rPrrrrO llama_n_embdcyrrNrs rPrrrrO llama_n_layercyrrNrs rPrrrrO llama_n_headcyrrNrs rPrrrrOllama_get_modelcyrrNrs rPrrrrOllama_pooling_typecyrrNrs rPrrrrOllama_vocab_typecyrrNrs rPr r rrOllama_rope_typecyrrNrs rPr r rrOllama_rope_freq_scale_traincy)z-Get the model's RoPE frequency scaling factorNrNrs rPrrrOllama_model_meta_val_strcy)z*Get metadata value as a string by key nameNrN)rrybufbuf_sizes rPrr$rOllama_model_meta_countcy)z*Get the number of metadata key/value pairsNrNrs rPrrrrOllama_model_meta_key_by_indexcy)zGet metadata key name by indexNrNrirrs rPrrrrO!llama_model_meta_val_str_by_indexcy)z'Get metadata value as a string by indexNrNrs rPrr/rrOllama_model_desccy)z&Get a string describing the model typeNrN)rrrs rPrrFrOllama_model_sizecy)z?Returns the total size of all the tensors in the model in bytesNrNrs rPr!r!WrrOllama_model_n_paramscy)z3Returns the total number of parameters in the modelNrNrs rPr#r#_rrOllama_get_model_tensornamecy)zGet a llama model tensorNrN)rr&s rPr%r%grrOllama_model_has_encodercy)zOReturns true if the model contains an encoder that requires llama_encode() callNrNrs rPr(r(srrOllama_model_has_decodercy)zNReturns true if the model contains a decoder that requires llama_decode() callNrNrs rPr*r*{rrOllama_model_decoder_start_tokency)zFor encoder-decoder models, this function returns id of the token that must be provided to the decoder to start generating output sequence. For other models, it returns -1. NrNrs rPr,r,rrOllama_model_is_recurrentcy)z?Returns true if the model is recurrent (like Mamba, RWKV, etc.)NrNrs rPr.r.rrOllama_model_quantizecy)zReturns 0 on successNrN) fname_inp fname_outrs rPr0r0 rOllama_lora_adapter_initcy)zLoad a LoRA adapter from file The loaded adapter will be associated to the given model, and will be free when the model is deleted NrN)r path_loras rPr5r5rOllama_lora_adapter_setcy)zRAdd a loaded LoRA adapter to given context This will not modify model's weightNrN)radapterscales rPr9r9rOllama_lora_adapter_removecy)zcRemove a LoRA adapter from given context Return -1 if the adapter is not present in the contextNrN)rr;s rPr>r>r=rOllama_lora_adapter_clearcy)z+Remove all LoRA adapters from given contextNrNrs rPr@r@rrOllama_lora_adapter_freecy)zhManually free a LoRA adapter Note: loaded adapters will be free when the associated model is deletedNrN)r;s rPrBrBrrOllama_control_vector_applycy)aApply a loaded control vector to a llama_context, or if data is NULL, clear the currently loaded vector. n_embd should be the size of a single layer's control, and data should point to an n_embd x n_layers buffer starting from layer 1. il_start and il_end are the layer range the vector should apply to (both inclusive) See llama_control_vector_load in common to load a control vector.NrN)lctxrSlenn_embdil_startil_ends rPrDrD6rOc*eZdZUdZerded<defgZy)llama_kv_cache_view_cellzInformation associated with an individual cell in the KV cache view. Attributes: pos (llama_pos): The position for this cell. Takes KV cache shifts into account. May be negative if the cell is not populated.rjrdN)rFrGrHrIrrJrjrMrNrOrPrMrM7s=   "#HrOrMc NeZdZUer(ded<ded<ded<ded<ded<ded<ded <d ed <dej fdej fdej fdej fdej fdej fd ejefd eje fgZ y ) llama_kv_cache_viewrTn_cells n_max_seq token_count used_cellsmax_contiguousmax_contiguous_idxz%CtypesArray[llama_kv_cache_view_cell]cellszCtypesArray[llama_seq_id]cells_sequencesN) rFrGrHrrJrKrhrirMrkrMrNrOrPrOrOds 4422 FNN# fnn% ' v~~& 6>>* v~~. .&..!9:; NFNN<89 HrOrOllama_kv_cache_view_initcy)z@Create an empty KV cache view. (use only for debugging purposes)NrN)rrs rPrXrXrrOllama_kv_cache_view_freecy)z7Free a KV cache view. (use only for debugging purposes)NrN)views rPrZrZrrOllama_kv_cache_view_updatecy)zlUpdate the KV cache view structure with the current state of the KV cache. (use only for debugging purposes)NrN)rr\s rPr]r] rOllama_get_kv_cache_token_countcy)zReturns the number of tokens in the KV cache (slow, use only for debug) If a KV cell has multiple sequences assigned to it, it will be counted multiple times NrNrs rPr`r`rrOllama_get_kv_cache_used_cellscy)zVReturns the number of used KV cells (i.e. have at least one sequence assigned to them)NrNrs rPrbrbr_rOllama_kv_cache_clearcy)zClear the KV cacheNrNrs rPrdrdrrOllama_kv_cache_seq_rmcy)aRemoves all tokens that belong to the specified sequence and have positions in [p0, p1) Returns false if a partial sequence cannot be removed. Removing a whole sequence never fails seq_id < 0 : match any sequence p0 < 0 : [0, p1] p1 < 0 : [p0, inf)NrN)rrfp0p1s rPrfrfs0rOllama_kv_cache_seq_cpcy)zCopy all tokens that belong to the specified sequence to another sequence Note that this does not allocate extra KV cache memory - it simply assigns the tokens to the new sequence p0 < 0 : [0, p1] p1 < 0 : [p0, inf)NrN)r seq_id_src seq_id_dstrhris rPrjrj.rOllama_kv_cache_seq_keepcy)z?Removes all tokens that do not belong to the specified sequenceNrNrrfs rProro r_rOllama_kv_cache_seq_addcy)a2Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1) If the KV cache is RoPEd, the KV data is updated accordingly: - lazily on next llama_decode() - explicitly with llama_kv_cache_update() p0 < 0 : [0, p1] p1 < 0 : [p0, inf)NrN)rrfrhrideltas rPrrrrs2rOllama_kv_cache_seq_divcy)zInteger division of the positions by factor of `d > 1` If the KV cache is RoPEd, the KV data is updated accordingly p0 < 0 : [0, p1] p1 < 0 : [p0, inf)NrN)rrfrhrids rPruruErnrOllama_kv_cache_defragcy)zDefragment the KV cache This will be applied: - lazily on next llama_decode() - explicitly with llama_kv_cache_update()NrNrs rPrxrxdrrOllama_kv_cache_updatecy)zDApply the KV cache updates (such as K-shifts, defragmentation, etc.)NrNrs rPrzrzorrOllama_kv_cache_can_shiftcy)z/Check if the context supports KV cache shiftingNrNrs rPr|r|wrrOllama_state_get_sizecy)zReturns the *actual* size in bytes of the state (rng, logits, embedding and kv_cache) - will often be smaller after compacting tokensNrNrs rPr~r~rrOllama_get_state_sizecy)zReturns the maximum size in bytes of the state (rng, logits, embedding and kv_cache) - will often be smaller after compacting tokensNrNrs rPrrrOllama_state_get_datacyzCopies the state to the specified destination address. Destination needs to have allocated enough memory. Returns the number of bytes copiedNrN)rdstrUs rPrrrrOllama_copy_state_datacyrrN)rrs rPrrrOllama_state_set_datacy)zUSet the state reading from the specified address Returns the number of bytes readNrN)rsrcrUs rPrrrOllama_set_state_datacy)z0Set the state reading from the specified addressNrN)rrs rPrrrrOllama_state_load_filecyrrNr path_session tokens_outn_token_capacityn_token_count_outs rPrr&rOllama_load_session_filecyrrNrs rPrrrrOllama_state_save_filecyrrNrrtokens n_token_counts rPrr+"rOllama_save_session_filecyrrNrs rPrrErrOllama_state_seq_get_sizecy)zCGet the exact size needed to copy the KV cache of a single sequenceNrNrqs rPrr]rrOllama_state_seq_get_datacy)z@Copy the KV cache of a single sequence into the specified bufferNrN)rrrUrfs rPrrmrrOllama_state_seq_set_datacy)zfCopy the sequence data (originally copied with `llama_state_seq_get_data`) into the specified sequenceNrN)rrrU dest_seq_ids rPrrrrOllama_state_seq_save_filecyrrN)rfilepathrfrrs rPrrrrOllama_state_seq_load_filecyrrN)rrrrrrs rPrrs*rOllama_batch_get_onecy)zReturn batch for single sequence of tokens starting at pos_0 NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it NrN)rr_s rPrrrrOllama_batch_initcy)aAllocates a batch of tokens on the heap that can hold a maximum of n_tokens Each token can be assigned up to n_seq_max sequence ids The batch has to be freed with llama_batch_free() If embd != 0, llama_batch.embd will be allocated with size of n_tokens * embd * sizeof(float) Otherwise, llama_batch.token will be allocated to store n_tokens llama_token The rest of the llama_batch members are allocated with size n_tokens All members are left uninitializedNrN)r_rcrs rPrr r4rOllama_batch_freecy)z9Frees a batch of tokens allocated with llama_batch_init()NrN)batchs rPrr rrO llama_encodecy)zProcesses a batch of tokens with the ecoder part of the encoder-decoder model. Stores the encoder output internally for later use by the decoder cross-attention layers. 0 - success < 0 - errorNrNrrs rPrr* rrO llama_decodecy)zPositive return values does not mean a fatal error, but rather a warning. 0 - success 1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context) < 0 - errorNrNrs rPrr: rrOllama_set_n_threadscy)zSet the number of threads used for decoding n_threads is the number of threads used for generation (single token) n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens) NrN)rrrs rPrrG rrOllama_n_threadscy)z?Get the number of threads used for generation of a single tokenNrNrs rPrr_ rrOllama_n_threads_batchcy)zOGet the number of threads used for prompt and batch processing (multiple token)NrNrs rPrrg rrOllama_set_embeddingscy)zpSet whether the model is in embeddings model or not If true, embeddings will be returned but logits will notNrN)rrs rPrrp rrOllama_set_causal_attncy)zlSet whether to use causal attention or not If set to true, the model will only attend to the past tokensNrN)r causal_attns rPrrz rrOllama_set_abort_callbackcy)zSet abort callbackNrN)rrrs rPrr r rOllama_synchronizecy)zWait until all computations are finished This is automatically done when using one of the functions below to obtain the computation results and is not necessary to call it explicitly in most casesNrNrs rPrr r_rOllama_get_logitscy)aZToken logits obtained from the last call to llama_decode() The logits for which llama_batch.logits[i] != 0 are stored contiguously in the order they have appeared in the batch. Rows: number of tokens for which llama_batch.logits[i] != 0 Cols: n_vocab Returns: Pointer to the logits buffer of shape (n_tokens, n_vocab)NrNrs rPrr r rOllama_get_logits_ithcy)zNLogits for the ith token. Equivalent to: llama_get_logits(ctx) + i*n_vocabNrNrrs rPrr r=rOllama_get_embeddingscy)zDGet the embeddings for the input shape: [n_embd] (1-dimensional)NrNrs rPrr rrOllama_get_embeddings_ithcy)zPGet the embeddings for the ith sequence llama_get_embeddings(ctx) + i*n_embdNrNrs rPrr r=rOllama_get_embeddings_seqcy)zGet the embeddings for a sequence id Returns NULL if pooling_type is LLAMA_POOLING_TYPE_NONE shape: [n_embd] (1-dimensional)NrNrqs rPrr r8rOllama_token_get_textcyrrNrras rPrr rrOllama_token_get_scorecyrrNrs rPrr rrOllama_token_get_attrcyrrNrs rPrr rrOllama_token_is_eogcy)zXCheck if the token is supposed to end generation (end-of-generation, eg. EOS, EOT, etc.)NrNrs rPrr r_rOllama_token_is_controlcy)z>Identify if Token Id is a control token or a render-able tokenNrNrs rPrr* rrOllama_token_boscy)zbeginning-of-sentenceNrNrs rPrr8 rrOllama_token_eoscy)zend-of-sentenceNrNrs rPrr? rrOllama_token_eotcy)z end-of-turnNrNrs rPrrF rrOllama_token_clscy)classificationNrNrs rPrrM rrOllama_token_sepcy)zsentence separatorNrNrs rPrrT rrOllama_token_nlcy)z next-lineNrNrs rPrr[ rrOllama_add_bos_tokencyrrNrs rPrrb rrOllama_add_eos_tokencyrrNrs rPrrh rrOllama_token_prefixcy)zcodellama infill tokensNrNrs rPrro rrOllama_token_middlecyrrNrs rPrrv rrOllama_token_suffixcyrrNrs rPrr| rrOllama_token_fim_precyrrNrs rPrr rrOllama_token_fim_sufcyrrNrs rPrr rrOllama_token_fim_midcyrrNrs rPrr rrOllama_token_fim_padcyrrNrs rPrr rrOllama_token_fim_repcyrrNrs rPrr rrOllama_token_fim_sepcyrrNrs rPrr rrOllama_tokenizecy)aConvert the provided text into tokens. Args: model: The model to use for tokenization. text: The text to tokenize. text_len: The length of the text. tokens: The tokens pointer must be large enough to hold the resulting tokens. n_max_tokens: The maximum number of tokens to return. add_special: Allow adding special tokenns if the model is configured to do so. parse_special: Allow parsing special tokens. Returns: Returns the number of tokens on success, no more than n_tokens_max Returns a negative number on failure - the number of tokens that would have been returned NrN)rtexttext_lenr n_tokens_max add_special parse_specials rPrr sLrOllama_token_to_piececy)a#Token Id -> Piece. Uses the vocabulary in the provided context. Does not write null terminator to the buffer. User code is responsible to remove the leading whitespace of the first non-BOS token when decoding multiple tokens. Args: model: The model to use for tokenization. token: The token to convert. buf: The buffer to write the token to. length: The length of the buffer. lstrip: The number of leading spaces to skip. special: If true, special tokens are rendered in the output.NrN)rrarlengthlstripspecials rPr r  BrOllama_detokenizecy)aConvert the provided tokens into text (inverse of llama_tokenize()). Args: model: The model to use for tokenization. tokens: The tokens to convert. n_tokens: The number of tokens. text: The buffer to write the text to. text_len_max: The length of the buffer. remove_special: Allow to remove BOS and EOS tokens if model is configured to do so. unparse_special: If true, special tokens are rendered in the output.NrN)rrr_r text_len_maxremove_specialunparse_specials rPrr- rrOllama_chat_apply_templatecyrrN)rtmplchatn_msgs rPrrh rrOllama_chat_builtin_templatescy)zGet list of built-in chat templates. Args: output: Output buffer to store template names. len: Length of the output buffer. Returns: Number of templates available. Returns a negative number on error. NrN)outputrGs rPrr~ rnrOc eZdZy)llama_sampler_iN)rFrGrHrNrOrPrr srOrc<eZdZdejefdefgZy) llama_samplerifacerN)rFrGrHrKrirllama_sampler_context_trMrNrOrPrr s% .&..12 '(HrOracceptapplyresetclonefreellama_sampler_namecyrrNsmpls rPr'r' rrOllama_sampler_acceptcyrrN)r*ras rPr+r+ rrOllama_sampler_applycyrrN)r*cur_ps rPr-r- rrOllama_sampler_resetcyrrNr)s rPr0r0 rrOllama_sampler_clonecyrrNr)s rPr2r2# rrOllama_sampler_freecyrrNr)s rPr4r4. rrOllama_sampler_chain_initcyrrN)rs rPr6r6; rrOllama_sampler_chain_addcyrrN)chainr*s rPr8r8F rrOllama_sampler_chain_getcyrrNr:rs rPr;r;P rrOllama_sampler_chain_ncyrrNr:s rPr>r>\ rrOllama_sampler_chain_removecyrrNr=s rPrArAg rrOllama_sampler_init_greedycyrrNrNrOrPrCrCu rrOllama_sampler_init_distcyrrN)seeds rPrErE{ rrOllama_sampler_init_softmaxcyrrNrNrOrPrHrH rrOllama_sampler_init_top_kcyrrN)ks rPrJrJ rrOllama_sampler_init_top_pcyrrNrDmin_keeps rPrMrM rrOllama_sampler_init_min_pcyrrNrOs rPrQrQ rrOllama_sampler_init_typicalcyrrNrOs rPrSrS rrOllama_sampler_init_tempcyrrN)ts rPrUrU rrOllama_sampler_init_temp_extcyrrN)rWrtexponents rPrXrX rrOllama_sampler_init_xtccyrrN)rDrWrPrGs rPr[r[ rrOllama_sampler_init_mirostatcyrrN)n_vocabrGtauetams rPr]r] rrOllama_sampler_init_mirostat_v2cyrrN)rGr`ras rPrcrc rrOllama_sampler_init_grammarcyrrN)r grammar_str grammar_roots rPrere rrOllama_sampler_init_penaltiesc yrrN) r_special_eos_id linefeed_idpenalty_last_npenalty_repeat penalty_freqpenalty_present penalize_nl ignore_eoss rPriri rKrOllama_sampler_init_drycyrrN)rdry_multiplierdry_basedry_allowed_lengthdry_penalty_last_n seq_breakers num_breakerss rPrsrs< rnrOllama_sampler_init_logit_biascyrrN)r_ n_logit_bias logit_biass rPr{r{Z rrOllama_sampler_init_infillcy)zGThis sampler is meant to be used for fill-in-the-middle infilling. NrNrs rPrr{ rrOllama_sampler_get_seedcyrrNr)s rPrr rrOllama_sampler_samplecyrrN)r*ridxs rPrr rrOllama_split_pathcy)z-Build a split GGUF final path for this chunk.NrN) split_pathmaxlen path_prefixsplit_no split_counts rPrr rrOllama_split_prefixcy)z^Extract the path prefix from the split_path if and only if the split_no and split_count match.NrN) split_prefixrrrrs rPrr rrOllama_print_system_infocyrrNrNrOrPrr rrO llama_log_setcy)z|Set callback for all future logging events. If this is not called, or NULL is supplied, everything is output on stderr.NrN) log_callback user_datas rPrr rrOceZdZdejfdejfdejfdejfdej fdej fgZy)llama_perf_context_data t_start_ms t_load_ms t_p_eval_ms t_eval_msn_p_evaln_evalNrFrGrHrKrtrhrMrNrOrPrrsQ v' foo& ( foo& V^^$ 6>>" HrOrcDeZdZdejfdej fgZy)llama_perf_sampler_data t_sample_msn_sampleNrrNrOrPrrs! ( V^^$HrOrllama_perf_contextcyrrNrs rPrrrrOllama_perf_context_printcyrrNrs rPrr"rrOllama_perf_context_resetcyrrNrs rPrr,rrOllama_perf_samplercyrrNr@s rPrr7rrOllama_perf_sampler_printcyrrNr@s rPrrArrOllama_perf_sampler_resetcyrrNr@s rPrrKrrO)returnr})rr)rr)rr)rrT)rrsrr}rOptional[llama_model_p])rr3)rr3rrrzOptional[llama_context_p])rr4)rrT)rrW)rr4rrT)rr3rrT)rr4rr)rr3rrB) rr3ryUnion[ctypes.c_char_p, bytes]rrsrrTrrT) rr3rUnion[ctypes.c_int, int]r(Union[bytes, CtypesArray[ctypes.c_char]]rrTrrT)rr3rrrUnion[ctypes.c_size_t, int]rrT)rr3r&rrr)rr3rrW)r2rsr3rsrz/CtypesPointerOrRef[llama_model_quantize_params]rrT)rr3r7rsrzOptional[llama_lora_adapter_p])rr4r;llama_lora_adapter_pr<rBrrT)rr4r;rrrT)r;r)rFr4rSz"CtypesPointerOrRef[ctypes.c_float]rGrTrHrTrIrTrJrTrrT)rr4rUnion[ctypes.c_int32, int]rrO)r\z%'ctypes.pointer[llama_kv_cache_view]')rr4r\z'CtypesPointerOrRef[llama_kv_cache_view]) rr4rfUnion[llama_seq_id, int]rhUnion[llama_pos, int]rirrrW) rr4rlrrmrrhrrir)rr4rfr) rr4rfrrhrrirrtr) rr4rfrrhrrirrwr)rr4rrW)rr4rCtypesArray[ctypes.c_uint8]rUrrrT)rr4rrrrT)rr4rrrUrrrT)rr4rrrrT) rr4rrsrr`rrr#CtypesPointerOrRef[ctypes.c_size_t]rrW) rr4rrsrr`rrrrrrT) rr4rrsrr`rrrrW) rr4rrsrr`rrrrT)rr4rfrkrrT) rr4rrrUrrfrkrrT) rr4rrrUrrrkrrT) rr4rrsrfrkrr`rrrrT)rr4rrsrrkrr`rrrrrrT)rr`r_rrr^)r_rrcrrrrr^)rr^)rr4rr^rrT)rr4rrrr)rr4rrW)rr4rrW)rr4rz!Callable[[ctypes.c_void_p], None]rr)rr4rrb)rr4rrrrb)rr4rfrrrb)rr3raUnion[llama_token, int]rrs)rr3rarrrB)rr3rarrrT)rr3rarrrW)rr3rrsrrrr`rrrUnion[ctypes.c_bool, bool]rrrrT)rr3rarrz9Union[ctypes.c_char_p, bytes, CtypesArray[ctypes.c_char]]r rr rr rrrT)rr3rr`r_rrrsrrrrrrrrT) rr3rrsrzCtypesArray[llama_chat_message]rrTrrT)rCtypesArray[bytes]rGrrrT)r*llama_sampler_prrs)r*rrar)r*rr/z#CtypesArray[llama_token_data_array])r*r)r*rrr)rrrr)r:rr*r)r:rrrrr)r:rrrT)rr)rGrTrr)rLrTrr)rDrBrPrTrr)rWrBrr)rWrBrtrBrZrBrr) rDrBrWrBrPrTrGrTrr) r_rTrGrTr`rBrarBrbrTrr)rGrTr`rBrarBrr)rr3rgrsrhrsrr)r_rTrkrTrlrTrmrTrnrBrorBrprBrqrWrrrWrr)rr3rurBrvrBrwrTrxrTryrrzrTrr)r_rTr}rTr~zCtypesArray[llama_logit_bias]rr)rr3rr)r*rrrT)r*rrr4rrTrrT) rrsrrrrsrrrrrrT) rrsrrrrsrrrrrrT)rrs)rzOptional[CtypesFuncPointer]rr)rr4rr)r:rrr)r:r( __future__rosrKpathlibtypingrrrrrllama_cpp._ctypes_extensionsr r r r r rrrrr_lib_base_nameenvironget_override_base_pathPathpathabspathdirname__file__ _base_path_libctypes_function GGML_TYPE_F32 GGML_TYPE_F16GGML_TYPE_Q4_0GGML_TYPE_Q4_1GGML_TYPE_Q5_0GGML_TYPE_Q5_1GGML_TYPE_Q8_0GGML_TYPE_Q8_1GGML_TYPE_Q2_KGGML_TYPE_Q3_KGGML_TYPE_Q4_KGGML_TYPE_Q5_KGGML_TYPE_Q6_KGGML_TYPE_Q8_KGGML_TYPE_IQ2_XXSGGML_TYPE_IQ2_XSGGML_TYPE_IQ3_XXSGGML_TYPE_IQ1_SGGML_TYPE_IQ4_NLGGML_TYPE_IQ3_SGGML_TYPE_IQ2_SGGML_TYPE_IQ4_XS GGML_TYPE_I8 GGML_TYPE_I16 GGML_TYPE_I32 GGML_TYPE_I64 GGML_TYPE_F64GGML_TYPE_IQ1_MGGML_TYPE_COUNT CFUNCTYPEr\rrrrargtypesrZrestypeLLAMA_MAX_DEVICESLLAMA_DEFAULT_SEEDLLAMA_TOKEN_NULLLLAMA_FILE_MAGIC_GGLALLAMA_FILE_MAGIC_GGSNLLAMA_FILE_MAGIC_GGSQLLAMA_SESSION_MAGICLLAMA_SESSION_VERSIONLLAMA_STATE_SEQ_MAGICLLAMA_STATE_SEQ_VERSIONrTr3llama_model_p_ctypesr4llama_context_p_ctypesrhrjr@ri llama_token_prkLLAMA_VOCAB_TYPE_NONELLAMA_VOCAB_TYPE_SPMLLAMA_VOCAB_TYPE_BPELLAMA_VOCAB_TYPE_WPMLLAMA_VOCAB_TYPE_UGMLLAMA_VOCAB_TYPE_RWKVLLAMA_VOCAB_PRE_TYPE_DEFAULTLLAMA_VOCAB_PRE_TYPE_LLAMA3!LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM#LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODERLLAMA_VOCAB_PRE_TYPE_FALCONLLAMA_VOCAB_PRE_TYPE_MPTLLAMA_VOCAB_PRE_TYPE_STARCODERLLAMA_VOCAB_PRE_TYPE_GPT2LLAMA_VOCAB_PRE_TYPE_REFACTLLAMA_VOCAB_PRE_TYPE_COMMAND_RLLAMA_VOCAB_PRE_TYPE_STABLELM2LLAMA_VOCAB_PRE_TYPE_QWEN2LLAMA_VOCAB_PRE_TYPE_OLMOLLAMA_VOCAB_PRE_TYPE_DBRXLLAMA_VOCAB_PRE_TYPE_SMAUGLLAMA_VOCAB_PRE_TYPE_POROLLAMA_VOCAV_PRE_TYPE_CHATGLM3LLAMA_VOCAB_PRE_TYPE_CHATGLM4LLAMA_VOCAB_PRE_TYPE_VIKINGLLAMA_VOCAB_PRE_TYPE_JAISLLAMA_VOCAB_PRE_TYPE_TEKKENLLAMA_VOCAB_PRE_TYPE_SMOLLMLLAMA_VOCAB_PRE_TYPE_CODESHELLLLAMA_VOCAB_PRE_TYPE_BLOOM!LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISHLLAMA_VOCAB_PRE_TYPE_EXAONELLAMA_VOCAB_PRE_TYPE_CHAMELEONLLAMA_VOCAB_PRE_TYPE_MINERVALLAMA_ROPE_TYPE_NONELLAMA_ROPE_TYPE_NORMLLAMA_ROPE_TYPE_NEOXGGML_ROPE_TYPE_NEOXLLAMA_TOKEN_TYPE_UNDEFINEDLLAMA_TOKEN_TYPE_NORMALLLAMA_TOKEN_TYPE_UNKNOWNLLAMA_TOKEN_TYPE_CONTROLLLAMA_TOKEN_TYPE_USER_DEFINEDLLAMA_TOKEN_TYPE_UNUSEDLLAMA_TOKEN_TYPE_BYTELLAMA_TOKEN_ATTR_UNDEFINEDLLAMA_TOKEN_ATTR_UNKNOWNLLAMA_TOKEN_ATTR_UNUSEDLLAMA_TOKEN_ATTR_NORMALLLAMA_TOKEN_ATTR_CONTROLLLAMA_TOKEN_ATTR_USER_DEFINEDLLAMA_TOKEN_ATTR_BYTELLAMA_TOKEN_ATTR_NORMALIZEDLLAMA_TOKEN_ATTR_LSTRIPLLAMA_TOKEN_ATTR_RSTRIPLLAMA_TOKEN_ATTR_SINGLE_WORDLLAMA_FTYPE_ALL_F32LLAMA_FTYPE_MOSTLY_F16LLAMA_FTYPE_MOSTLY_Q4_0LLAMA_FTYPE_MOSTLY_Q4_1LLAMA_FTYPE_MOSTLY_Q8_0LLAMA_FTYPE_MOSTLY_Q5_0LLAMA_FTYPE_MOSTLY_Q5_1LLAMA_FTYPE_MOSTLY_Q2_KLLAMA_FTYPE_MOSTLY_Q3_K_SLLAMA_FTYPE_MOSTLY_Q3_K_MLLAMA_FTYPE_MOSTLY_Q3_K_LLLAMA_FTYPE_MOSTLY_Q4_K_SLLAMA_FTYPE_MOSTLY_Q4_K_MLLAMA_FTYPE_MOSTLY_Q5_K_SLLAMA_FTYPE_MOSTLY_Q5_K_MLLAMA_FTYPE_MOSTLY_Q6_KLLAMA_FTYPE_MOSTLY_IQ2_XXSLLAMA_FTYPE_MOSTLY_IQ2_XSLLAMA_FTYPE_MOSTLY_Q2_K_SLLAMA_FTYPE_MOSTLY_IQ3_XSLLAMA_FTYPE_MOSTLY_IQ3_XXSLLAMA_FTYPE_MOSTLY_IQ1_SLLAMA_FTYPE_MOSTLY_IQ4_NLLLAMA_FTYPE_MOSTLY_IQ3_SLLAMA_FTYPE_MOSTLY_IQ3_MLLAMA_FTYPE_MOSTLY_IQ2_SLLAMA_FTYPE_MOSTLY_IQ2_MLLAMA_FTYPE_MOSTLY_IQ4_XSLLAMA_FTYPE_MOSTLY_IQ1_MLLAMA_FTYPE_MOSTLY_BF16LLAMA_FTYPE_MOSTLY_TQ1_0LLAMA_FTYPE_MOSTLY_TQ2_0LLAMA_FTYPE_GUESSED#LLAMA_ROPE_SCALING_TYPE_UNSPECIFIEDLLAMA_ROPE_SCALING_TYPE_NONELLAMA_ROPE_SCALING_TYPE_LINEARLLAMA_ROPE_SCALING_TYPE_YARN LLAMA_ROPE_SCALING_TYPE_LONGROPE!LLAMA_ROPE_SCALING_TYPE_MAX_VALUELLAMA_POOLING_TYPE_UNSPECIFIEDLLAMA_POOLING_TYPE_NONELLAMA_POOLING_TYPE_MEANLLAMA_POOLING_TYPE_CLSLLAMA_POOLING_TYPE_LASTLLAMA_POOLING_TYPE_RANK LLAMA_ATTENTION_TYPE_UNSPECIFIEDLLAMA_ATTENTION_TYPE_CAUSALLLAMA_ATTENTION_TYPE_NON_CAUSALLLAMA_SPLIT_MODE_NONELLAMA_SPLIT_MODE_LAYERLLAMA_SPLIT_MODE_ROW Structurer?rYrRllama_token_data_array_prLrr^LLAMA_KV_OVERRIDE_TYPE_INTLLAMA_KV_OVERRIDE_TYPE_FLOATLLAMA_KV_OVERRIDE_TYPE_BOOLLLAMA_KV_OVERRIDE_TYPE_STRrnrwr}rr{rllama_log_callbackrrllama_logit_bias_prrrllama_lora_adapter_p_ctypesrrrrrGGML_NUMA_STRATEGY_DISABLEDGGML_NUMA_STRATEGY_DISTRIBUTEGGML_NUMA_STRATEGY_ISOLATEGGML_NUMA_STRATEGY_NUMACTLGGML_NUMA_STRATEGY_MIRRORGGML_NUMA_STRATEGY_COUNTrrrrrrr[rrrrrrrrrrrrrrrrrr r rrrrrrc_uint64r!r#r%r(r*r,r.r0r5r9r>r@rBrDrMrOllama_kv_cache_view_prXrZr]r`rbrdrfrjrorrrurxrzr|r~rc_uint8rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr rrrr!rrrllama_sampler_p_ctypesllama_sampler_i_namellama_sampler_i_acceptllama_sampler_i_applyllama_sampler_i_resetllama_sampler_i_clonellama_sampler_i_freerMr'r+r-r0r2r4r6r8r;r>rArCrErHrJrMrQrSrUrXr[r]rcrerirsr{rrrrrrrrrrrrrrrrNrOrPrs."  jjnn%9:QdQl\W\\"''//"''//(*CD E Mr~ryr~r~@SsT >:64T:L       $46#3#3 MM6??FMM6??$ 'f&&v}}fooF#%!'**, ###,.- +S1 NN nn {+ ~~ @-'#1D !$%!&'#!"!"!# " "   !#$&! !#!-..* ! !  ! &$  %\')# !" #$ $@!"$$& "#v''($V^^$45V--.*6>>*@A+&** MM6>>6??8!&""!T  FLL  .f.. .d-))-xZ6++ZF&V%%&,,J**&"2"2*bv''"$V^^$45 !1!1 &)),fnnV__=      "    (    )         !  \\N         __() 1     "/0  "6       NN    $b&//:; &FMM:; 'V]];< -r6==AB %r6==9: !7 8&//JK #9":FOOLM !$:#;V__MN "%;$RST %(>'?NO #&:%;V\\JK "%9$:FLLIJ .1E0FWX   NN   &     ),@+A6>>RS#   NN   2    '   NN   2    6??FOO< NN   1*    #&:%;V__MN'*>)?QR3V__Ev  =*-A,BFMMRS*-A,BFMMRS%(<'=v~~+.B-CV]]ST23  OO <  6??+  %*#  8&..I NN  #7@E  89 NN  #7      !   & v~~&  NN   ,         : $v// $Z&**.'':; V^^,  %?  +.C-DdKL #9;P"QSW$'=&>#&<%=v~~'*@)A4HI   MM   $      6      ( )       & 6 Et"     $    !  6       $         ((+A*BDIJ(+A*BDIJ+.D-Ev}}UV'*@)A6??ST'*@)A6??STv~~&  OO    $  &     v~~& OO  :^V^^FNN;V__M OO   $ &    ^V^^FNN;< OO  :  v'  MM  )2  ;   $v'  OO  )2  ;       MM   %/        OO   %/    \* OO   v~~&   OO   $ &     (v~~&   OO   $ &    "  OO    %  /   $v'  OO      )  2  ;     4   $ &   .H+ (  $ *    $#k]D9:"8+!FWX"8+!FWX     ) 0  "%;$EF/0.&..2P  "V^^,FNN6>>"  7   34nfnnV^^6TV^^,FNN6>>"  7   \*FNN6>>"  ":   1;? !8 2K@&.. !8 1;? !8/=v}}3[A6== !8 "%9$:KHI "%9$:KHI "%9$:KHI "%9$:KHI "%9$:KHI !$8#9;GH &)=(> NO &)=(> NO %(<'={KL %(<'={KL %(<'={KL &)=(> LM&)=(> LM&)=(> LM&)=(> LM&)=(> LM&)=(> LM2   NN   ' %  +  , .  P  NN   " C %  %  (   n{#   NN   $'   +  / 0  \)*   NN    *     "v' NN   $ v!//f&&F$$#M2O' 6'v''9OP)))$0F T((( ":)((/EF((()?AWX'v''.DE !" %& #$ #$ #$ !"  OO    [)    56  "E                  34    V^^,  9   LL     V^^,  9  ,b2HIJ *V__,=?UVW-r3IJK +fnn-=?UVW  ^^V__%     ^^V__%      ^^V__%    *V^^,<>TUV ! ^^V^^V^^4  &+   ^^V^^V__fooF  "%-0  $! ^^V__fnnfnnfnnU  "'.38;  $ __fnnfnn5   %   6??FOO<  ',