>:j;#tSSKJr SSKJrJr SSKJrJr SSKJ r SSK J r SSK J r \"SS\ 55rg ) ) annotations) dataclassfield)OptionalUnion)ACT2FN) PeftConfig)PeftTypecF^\rSrSr%Sr\"SSS0S9rS\S'\"S SS 0S9rS\S '\"S SS 0S9r S\S'\"SSS0S9r S\S'\"SSS0S9r S\S'\"SSS0S9r S\S'\"SSS0S9r S\S'\"SSS0S9rS\S'\"SSS 0S9rS\S!'\"S"SS#0S9rS$\S%'U4S&jrS'rU=r$)( PeanutConfigae This is the configuration class to store the configuration of a [`PeanutModel`]. Args: r (`int`): PEANuT rank. This is the hidden dimension used by the adapters. Similar to LoRA rank, larger `r` increases adapter capacity and trainable parameters. depth (`int`): Number of hidden adapter layers per encoder/decoder side in PEANuT. The input projection `A` and output projection `B` are always present in addition to these hidden layers. Therefore, `depth` must be a non-negative integer. - `depth=0`: `A`, `B`. - `depth=1`: `A`, one encoder, one decoder, `B`. - `depth=2`: `A`, two encoders, two decoders, `B`. - `depth=3`: `A`, three encoders, three decoders, `B`, etc. act_fn (`str`): Non-linear activation applied in the PEANuT network. This corresponds to `non_linear` in the vanilla PyTorch implementation. Default is `"relu"`. Any activation key available in `transformers.activations.ACT2FN` is supported and may perform better on different tasks. scaling (`float`): A scalar multiplier applied to the PEANuT output before adding it to the frozen base layer output. The final adapter contribution is `scaling * (x @ delta_w)`. target_modules (`Union[List[str], str]`, *optional*): The names of the modules to apply PEANuT to. Can be a list of module name strings (e.g. `['q_proj', 'v_proj']`) or a regex pattern. modules_to_save (`List[str]`, *optional*): List of modules apart from PEANuT layers to be set as trainable and saved in the final checkpoint. exclude_modules (`Union[List[str], str]`, *optional*): The names of the modules to not apply the adapter. When passing a string, a regex match will be performed. When passing a list of strings, either an exact match will be performed or it is checked if the name of the module ends with any of the passed strings. layers_to_transform (`Union[list[int], int]`, *optional*): The layer indexes to transform. If this argument is specified, PEFT will transform only the layer indexes that are specified in this list. If a single integer is passed, PEFT will transform only the layer at this index. layers_pattern (`Optional[Union[List[str], str]]`, *optional*): The layer pattern name, used only if `layers_to_transform` is not None and if the layer pattern is not in the common layers pattern. init_weights (`bool`): Whether to initialize PEANuT adapter weights using the default initialization scheme: - If `True`: all weights except `B` are initialized with Kaiming uniform, and `B` is initialized to zero. - If `False`: all weights (including `B`) are initialized with Kaiming uniform. Initializing `B` to zero makes the adapter start as an exact no-op. Notes: PEANuT uses a weight-aware pathway, where the delta weight is conditioned on the base weight. The `A` adapter is applied over the base weight's output dimension, so `A` has shape `(out_dim -> r)` rather than the usual `(in_dim -> r)` used by LoRA-like methods. helpzPEANuT rank. This is the hidden dimension used by the adapter stack. Similar to LoRA rank, larger `r` increases adapter capacity and trainable parameters.)defaultmetadataintrrzNumber of hidden adapter layers per encoder/decoder side in PEANuT. The input projection `A` and output projection `B` are added automatically, so `depth` must be a non-negative integer.depthreluzNon-linear activation applied in the PEANuT pathway. This corresponds to `non_linear` in the vanilla implementation. Must be a key in `transformers.activations.ACT2FN`.stract_fng?zA scalar multiplier applied to the PEANuT output before adding it to the frozen base layer output. The final adapter contribution is `scaling * (x @ delta_w)`.floatscalingNzList of module names or regex expression of the module names to replace with PEANuT. For example, ['q', 'v'] or '.*decoder.*(SelfAttention|EncDecAttention).*(q|v)$'. If not specified, PEANuT will use architecture-specific default target modules.zOptional[Union[list[str], str]]target_modulesa$List of module names or regex expression of the module names to exclude from PEANuT. When passing a string, a regex match will be performed. When passing a list of strings, either an exact match will be performed or it is checked if the name of the module ends with any of the passed strings.exclude_moduleszbList of modules apart from PEANuT layers to be set as trainable and saved in the final checkpoint.zOptional[list[str]]modules_to_savezThe layer indexes to transform, if this argument is specified, PEFT will transform only the layers indexes that are specified inside this list. If a single integer is passed, PEFT will transform only the layer at this index.zOptional[Union[list[int], int]]layers_to_transformzThe layer pattern name, used only if `layers_to_transform` is different to None and if the layer pattern is not in the common layers pattern. This should target the `nn.ModuleList` of the model, which is often called `'layers'` or `'h'`.layers_patternTaWhether to initialize PEANuT adapter weights using the default initialization scheme: if `True`, all weights except `B` are initialized with Kaiming uniform and `B` is initialized to zero; if `False`, all weights including `B` are initialized with Kaiming uniform.bool init_weightsc >[TU]5 [RUl[ UR [5(a[UR 5O UR Ul[ UR[5(a[UR5O URUl UR(aUR(d [S5eURS::a [S5eURS:a [S5eUR[ ;a8[SURS[#[ R$"55S35eg)NzQWhen `layers_pattern` is specified, `layers_to_transform` must also be specified.rz`r` must be a positive integer.z+`depth` can only be a non-negative integer.zUnsupported `act_fn`: z. Must be one of .)super __post_init__r PEANUT peft_type isinstancerlistsetrrr ValueErrorrrrrsortedkeys)self __class__s V/home/mande/repo/quber/.venv/lib/python3.13/site-packages/peft/tuners/peanut/config.pyr$PeanutConfig.__post_init__s  !(243F3F(M(MC## $SWSfSf *4D4H4H$)O)OC$$ %UYUiUi    t'?'?pq q 66Q;>? ? ::>JK K ;;f $5dkk]BSTZ[a[f[f[hTiSjjklm m %)rr&r)__name__ __module__ __qualname____firstlineno____doc__rr__annotations__rrrrrrrrr r$__static_attributes__ __classcell__)r.s@r/r r s~3j K As l E3 ^ FC O GU7< b  7N3 8= 2  8O4 ,1 t ,O(rAs:#("+" fn:fn fnr1