Index

_

__init__() (AdditionIntervention method) (AutoencoderIntervention method) (AutoencoderLayer method) (AutoencoderLayerBase method) (BackpackGPT2BaseModelOutput method) (BackpackGPT2Config method) (BackpackGPT2LMHeadModel method) (BackpackGPT2LMHeadModelOutput method) (BackpackGPT2Model method) (BackpackGPT2PreTrainedModel method) (BackpackMLP method) (BackpackNoMixBlock method) (BackpackSenseNetwork method) (BackpackWeightNetwork method) (BaseModel method) (BasisAgnosticIntervention method) (BlipITMWrapper method) (BlipWrapper method) (BoundlessRotatedSpaceIntervention method) (CausalModel method) (CollectIntervention method) (ConstantSourceIntervention method) (DistributedRepresentationIntervention method) (GRUCell method) (GRUConfig method) (GRUForClassification method) (GRULMHeadModel method) (GRUModel method) (GRUModelOutput method) (GRUPreTrainedModel method) (HandlerList method) (IntervenableConfig method) (IntervenableModel method) (IntervenableModelOutput method) (IntervenableNdifModel method) (Intervention method) (InterventionOutput method) (InterventionState method) (InverseRotateLayer method) (JumpReLUAutoencoderIntervention method) (LocalistRepresentationIntervention method) (LowRankRotatedSpaceIntervention method) (LowRankRotateLayer method) (MLPBlock method) (MLPConfig method) (MLPForClassification method) (MLPModel method) (MLPModelOutput method) (NoiseIntervention method) (PCARotatedSpaceIntervention method) (RepresentationConfig method) (RotatedSpaceIntervention method) (RotateLayer method) (SharedWeightsTrainableIntervention method) (SigmoidMaskIntervention method) (SigmoidMaskRotatedSpaceIntervention method) (SkipIntervention method) (SourcelessIntervention method) (SubspaceLowRankRotateLayer method) (SubtractionIntervention method) (TrainableIntervention method) (VanillaIntervention method) (ZeroIntervention method)

A

AdditionIntervention (class in pyvene.models.interventions) AutoencoderIntervention (class in pyvene.models.interventions)

AutoencoderLayer (class in pyvene.models.layers) AutoencoderLayerBase (class in pyvene.models.layers)

B

b_sd_to_bsd() (in module pyvene.models.modeling_utils) BackpackGPT2BaseModelOutput (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackGPT2Config (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackGPT2LMHeadModel (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackGPT2LMHeadModelOutput (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackGPT2Model (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackGPT2PreTrainedModel (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackMLP (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackNoMixBlock (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackSenseNetwork (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BackpackWeightNetwork (class in pyvene.models.backpack_gpt2.modelings_backpack_gpt2) BaseModel (class in pyvene.models.intervenable_base)

BasisAgnosticIntervention (class in pyvene.models.interventions) bhsd_to_bs_hd() (in module pyvene.models.modeling_utils) blip_itm_type_to_dimension_mapping (in module pyvene.models.blip.modelings_intervenable_blip_itm) blip_type_to_dimension_mapping (in module pyvene.models.blip.modelings_intervenable_blip) BlipITMWrapper (class in pyvene.models.blip.modelings_blip_itm) BlipWrapper (class in pyvene.models.blip.modelings_blip) BoundlessRotatedSpaceIntervention (class in pyvene.models.interventions) broadcast_tensor_v1() (in module pyvene.models.intervention_utils) broadcast_tensor_v2() (in module pyvene.models.intervention_utils) bs_hd_to_bhsd() (in module pyvene.models.modeling_utils) bsd_to_b_sd() (in module pyvene.models.modeling_utils) build_intervenable_model() (in module pyvene.models.intervenable_base)

C

CausalModel (class in pyvene.data_generators.causal_model) closeness_to_permutation_loss() (in module pyvene.models.basic_utils) CollectIntervention (class in pyvene.models.interventions) component (RepresentationConfig attribute) config_class (BackpackGPT2PreTrainedModel attribute) ConstantSourceIntervention (class in pyvene.models.interventions) count_parameters() (BaseModel method) (in module pyvene.models.basic_utils) create_backpack_gpt2() (in module pyvene.models.backpack_gpt2.modelings_intervenable_backpack_gpt2) create_blip() (in module pyvene.models.blip.modelings_intervenable_blip) create_blip_itm() (in module pyvene.models.blip.modelings_intervenable_blip_itm) create_directory() (in module pyvene.models.basic_utils) create_gemma() (in module pyvene.models.gemma.modelings_intervenable_gemma)

create_gemma2() (in module pyvene.models.gemma2.modelings_intervenable_gemma2) create_gpt2() (in module pyvene.models.gpt2.modelings_intervenable_gpt2) create_gpt2_classifier() (in module pyvene.models.gpt2.modelings_intervenable_gpt2) create_gpt2_lm() (in module pyvene.models.gpt2.modelings_intervenable_gpt2) create_gpt_neo() (in module pyvene.models.gpt_neo.modelings_intervenable_gpt_neo) create_gpt_neox() (in module pyvene.models.gpt_neox.modelings_intervenable_gpt_neox) create_gru() (in module pyvene.models.gru.modelings_intervenable_gru) create_gru_classifier() (in module pyvene.models.gru.modelings_intervenable_gru) create_gru_lm() (in module pyvene.models.gru.modelings_intervenable_gru) create_llama() (in module pyvene.models.llama.modelings_intervenable_llama) create_llava() (in module pyvene.models.llava.modelings_intervenable_llava) create_mistral() (in module pyvene.models.mistral.modellings_intervenable_mistral) create_mlp_classifier() (in module pyvene.models.mlp.modelings_intervenable_mlp) create_olmo() (in module pyvene.models.olmo.modelings_intervenable_olmo)

D

disable_intervention_gradients() (BaseModel method) disable_model_gradients() (BaseModel method)

DistributedRepresentationIntervention (class in pyvene.models.interventions) do_intervention() (in module pyvene.models.modeling_utils)

E

embed_to_distrib() (in module pyvene.models.basic_utils) enable_model_gradients() (BaseModel method)

eval() (IntervenableModel method) eval_alignment() (IntervenableModel method)

F

format_token() (in module pyvene.models.basic_utils) forward() (AdditionIntervention method) (AutoencoderIntervention method) (AutoencoderLayer method) (BackpackGPT2LMHeadModel method) (BackpackGPT2Model method) (BackpackMLP method) (BackpackNoMixBlock method) (BackpackSenseNetwork method) (BackpackWeightNetwork method) (BaseModel method) (BlipITMWrapper method) (BlipWrapper method) (BoundlessRotatedSpaceIntervention method) (CollectIntervention method) (GRUCell method) (GRUForClassification method) (GRULMHeadModel method) (GRUModel method) (IntervenableModel method) (IntervenableNdifModel method) (Intervention method) (InverseRotateLayer method) (JumpReLUAutoencoderIntervention method) (LowRankRotatedSpaceIntervention method) (LowRankRotateLayer method) (MLPBlock method) (MLPForClassification method) (MLPModel method) (NoiseIntervention method) (PCARotatedSpaceIntervention method) (RotatedSpaceIntervention method) (RotateLayer method) (SigmoidMaskIntervention method) (SigmoidMaskRotatedSpaceIntervention method) (SkipIntervention method) (SubspaceLowRankRotateLayer method) (SubtractionIntervention method) (VanillaIntervention method) (ZeroIntervention method)

G

gather_neurons() (in module pyvene.models.modeling_utils) gemma2_lm_type_to_dimension_mapping (in module pyvene.models.gemma2.modelings_intervenable_gemma2) gemma2_type_to_dimension_mapping (in module pyvene.models.gemma2.modelings_intervenable_gemma2) gemma_lm_type_to_dimension_mapping (in module pyvene.models.gemma.modelings_intervenable_gemma) gemma_type_to_dimension_mapping (in module pyvene.models.gemma.modelings_intervenable_gemma) generate() (IntervenableModel method) get_batch_size() (in module pyvene.models.basic_utils) get_cached_activations() (BaseModel method) get_cached_hot_activations() (BaseModel method) get_device() (BaseModel method) get_dimension_by_component() (in module pyvene.models.modeling_utils) get_input_embeddings() (GRUModel method) get_internal_model_type() (in module pyvene.models.modeling_utils) get_list_depth() (in module pyvene.models.basic_utils) GET_LOC() (in module pyvene.models.basic_utils) get_module_hook() (in module pyvene.models.modeling_utils) get_output_embeddings() (GRULMHeadModel method)

get_trainable_parameters() (BaseModel method) get_type_from_string() (in module pyvene.models.basic_utils) getattr_for_torch_module() (in module pyvene.models.modeling_utils) gpt2_lm_type_to_dimension_mapping (in module pyvene.models.gpt2.modelings_intervenable_gpt2) gpt2_type_to_dimension_mapping (in module pyvene.models.gpt2.modelings_intervenable_gpt2) gpt_neo_type_to_dimension_mapping (in module pyvene.models.gpt_neo.modelings_intervenable_gpt_neo) gpt_neox_type_to_dimension_mapping (in module pyvene.models.gpt_neox.modelings_intervenable_gpt_neox) group_key (RepresentationConfig attribute) gru_classifier_type_to_dimension_mapping (in module pyvene.models.gru.modelings_intervenable_gru) gru_type_to_dimension_mapping (in module pyvene.models.gru.modelings_intervenable_gru) GRUCell (class in pyvene.models.gru.modelings_gru) GRUConfig (class in pyvene.models.gru.modelings_gru) GRUForClassification (class in pyvene.models.gru.modelings_gru) GRULMHeadModel (class in pyvene.models.gru.modelings_gru) GRUModel (class in pyvene.models.gru.modelings_gru) GRUModelOutput (class in pyvene.models.gru.modelings_gru) GRUPreTrainedModel (class in pyvene.models.gru.modelings_gru)

H

HandlerList (class in pyvene.models.modeling_utils)

harmonic_sigmoid_boundary() (in module pyvene.models.basic_utils) hidden_source_representation (RepresentationConfig attribute)

I

IntervenableConfig (class in pyvene.models.configuration_intervenable_model) IntervenableModel (class in pyvene.models.intervenable_base) IntervenableModelOutput (class in pyvene.models.intervenable_base) IntervenableNdifModel (class in pyvene.models.intervenable_base) Intervention (class in pyvene.models.interventions) intervention (RepresentationConfig attribute) intervention_link_key (RepresentationConfig attribute)

intervention_type (RepresentationConfig attribute) InterventionOutput (class in pyvene.models.interventions) InterventionState (class in pyvene.models.intervention_utils) InverseRotateLayer (class in pyvene.models.layers) is_gru() (in module pyvene.models.modeling_utils) is_mlp() (in module pyvene.models.modeling_utils) is_stateless() (in module pyvene.models.modeling_utils) is_transformer() (in module pyvene.models.modeling_utils)

J

JumpReLUAutoencoderIntervention (class in pyvene.models.interventions)

L

latent_dim (RepresentationConfig attribute) layer (RepresentationConfig attribute) llama_lm_type_to_dimension_mapping (in module pyvene.models.llama.modelings_intervenable_llama) llama_type_to_dimension_mapping (in module pyvene.models.llama.modelings_intervenable_llama) llava_lm_type_to_dimension_mapping (in module pyvene.models.llava.modelings_intervenable_llava) llava_type_to_dimension_mapping (in module pyvene.models.llava.modelings_intervenable_llava)

load() (IntervenableModel static method) (IntervenableNdifModel static method) load_intervention() (IntervenableModel method) LocalistRepresentationIntervention (class in pyvene.models.interventions) low_rank_dimension (RepresentationConfig attribute) LowRankRotatedSpaceIntervention (class in pyvene.models.interventions) LowRankRotateLayer (class in pyvene.models.layers)

M

max_number_of_units (RepresentationConfig attribute) mistral_type_to_dimension_mapping (in module pyvene.models.mistral.modellings_intervenable_mistral) mlp_type_to_dimension_mapping (in module pyvene.models.mlp.modelings_intervenable_mlp) MLPBlock (class in pyvene.models.mlp.modelings_mlp) MLPConfig (class in pyvene.models.mlp.modelings_mlp) MLPForClassification (class in pyvene.models.mlp.modelings_mlp) MLPModel (class in pyvene.models.mlp.modelings_mlp) MLPModelOutput (class in pyvene.models.mlp.modelings_mlp) module pyvene pyvene.data_generators pyvene.data_generators.causal_model pyvene.models pyvene.models.backpack_gpt2 pyvene.models.backpack_gpt2.modelings_backpack_gpt2 pyvene.models.backpack_gpt2.modelings_intervenable_backpack_gpt2 pyvene.models.basic_utils pyvene.models.blip pyvene.models.blip.modelings_blip pyvene.models.blip.modelings_blip_itm pyvene.models.blip.modelings_intervenable_blip pyvene.models.blip.modelings_intervenable_blip_itm pyvene.models.configuration_intervenable_model pyvene.models.constants pyvene.models.gemma pyvene.models.gemma.modelings_intervenable_gemma pyvene.models.gemma2 pyvene.models.gemma2.modelings_intervenable_gemma2 pyvene.models.gpt2 pyvene.models.gpt2.modelings_intervenable_gpt2 pyvene.models.gpt_neo pyvene.models.gpt_neo.modelings_intervenable_gpt_neo pyvene.models.gpt_neox pyvene.models.gpt_neox.modelings_intervenable_gpt_neox pyvene.models.gru pyvene.models.gru.modelings_gru pyvene.models.gru.modelings_intervenable_gru pyvene.models.intervenable_base pyvene.models.intervenable_modelcard pyvene.models.intervention_utils pyvene.models.interventions pyvene.models.layers pyvene.models.llama pyvene.models.llama.modelings_intervenable_llama pyvene.models.llava pyvene.models.llava.modelings_intervenable_llava pyvene.models.mistral pyvene.models.mistral.modellings_intervenable_mistral pyvene.models.mlp pyvene.models.mlp.modelings_intervenable_mlp pyvene.models.mlp.modelings_mlp pyvene.models.modeling_utils pyvene.models.olmo pyvene.models.olmo.modelings_intervenable_olmo

moe_key (RepresentationConfig attribute)

N

named_parameters() (BaseModel method)

NoiseIntervention (class in pyvene.models.interventions)

O

olmo_lm_type_to_dimension_mapping (in module pyvene.models.olmo.modelings_intervenable_olmo)

olmo_type_to_dimension_mapping (in module pyvene.models.olmo.modelings_intervenable_olmo) output_to_subcomponent() (in module pyvene.models.modeling_utils)

P

PCARotatedSpaceIntervention (class in pyvene.models.interventions) print_forward_hooks() (in module pyvene.models.modeling_utils) pyvene module pyvene.data_generators module pyvene.data_generators.causal_model module pyvene.models module pyvene.models.backpack_gpt2 module pyvene.models.backpack_gpt2.modelings_backpack_gpt2 module pyvene.models.backpack_gpt2.modelings_intervenable_backpack_gpt2 module pyvene.models.basic_utils module pyvene.models.blip module pyvene.models.blip.modelings_blip module pyvene.models.blip.modelings_blip_itm module pyvene.models.blip.modelings_intervenable_blip module pyvene.models.blip.modelings_intervenable_blip_itm module pyvene.models.configuration_intervenable_model module pyvene.models.constants module pyvene.models.gemma module pyvene.models.gemma.modelings_intervenable_gemma module pyvene.models.gemma2 module pyvene.models.gemma2.modelings_intervenable_gemma2 module pyvene.models.gpt2 module pyvene.models.gpt2.modelings_intervenable_gpt2 module pyvene.models.gpt_neo module

pyvene.models.gpt_neo.modelings_intervenable_gpt_neo module pyvene.models.gpt_neox module pyvene.models.gpt_neox.modelings_intervenable_gpt_neox module pyvene.models.gru module pyvene.models.gru.modelings_gru module pyvene.models.gru.modelings_intervenable_gru module pyvene.models.intervenable_base module pyvene.models.intervenable_modelcard module pyvene.models.intervention_utils module pyvene.models.interventions module pyvene.models.layers module pyvene.models.llama module pyvene.models.llama.modelings_intervenable_llama module pyvene.models.llava module pyvene.models.llava.modelings_intervenable_llava module pyvene.models.mistral module pyvene.models.mistral.modellings_intervenable_mistral module pyvene.models.mlp module pyvene.models.mlp.modelings_intervenable_mlp module pyvene.models.mlp.modelings_mlp module pyvene.models.modeling_utils module pyvene.models.olmo module pyvene.models.olmo.modelings_intervenable_olmo module

R

random_permutation_matrix() (in module pyvene.models.basic_utils) remove_forward_hooks() (in module pyvene.models.modeling_utils)

RepresentationConfig (class in pyvene.models.configuration_intervenable_model) RotatedSpaceIntervention (class in pyvene.models.interventions) RotateLayer (class in pyvene.models.layers)

S

save() (IntervenableModel method) save_intervention() (IntervenableModel method) scatter_neurons() (in module pyvene.models.modeling_utils) set_device() (BaseModel method) set_input_embeddings() (GRUModel method) set_seed() (in module pyvene.models.basic_utils) set_temperature() (BaseModel method) set_zero_grad() (BaseModel method) SharedWeightsTrainableIntervention (class in pyvene.models.interventions) sigmoid_boundary() (in module pyvene.models.basic_utils) SigmoidMaskIntervention (class in pyvene.models.interventions) SigmoidMaskRotatedSpaceIntervention (class in pyvene.models.interventions) simple_example() (in module pyvene.data_generators.causal_model)

simple_output_to_subcomponent() (in module pyvene.models.modeling_utils) simple_scatter_intervention_output() (in module pyvene.models.modeling_utils) SkipIntervention (class in pyvene.models.interventions) source_representation (RepresentationConfig attribute) SourcelessIntervention (class in pyvene.models.interventions) split_and_select() (in module pyvene.models.constants) split_half() (in module pyvene.models.constants) split_head_and_permute() (in module pyvene.models.constants) split_heads() (in module pyvene.models.constants) split_three() (in module pyvene.models.constants) subspace_partition (RepresentationConfig attribute) SubspaceLowRankRotateLayer (class in pyvene.models.layers) SubtractionIntervention (class in pyvene.models.interventions)

T

tie_weights() (BackpackGPT2LMHeadModel method) top_vals() (in module pyvene.models.basic_utils)

train() (IntervenableModel method) train_alignment() (IntervenableModel method) TrainableIntervention (class in pyvene.models.interventions)

U

unit (RepresentationConfig attribute)

V

VanillaIntervention (class in pyvene.models.interventions)

W

weighted_average() (in module pyvene.models.modeling_utils)

Z

zero_grad() (BaseModel method)

ZeroIntervention (class in pyvene.models.interventions)