Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[P2] Add ESM models #195

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[P2] Add ESM models #195

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Empty file added pyvene/models/esm/__init__.py
View file
Open in desktop
Empty file.
63 changes: 63 additions & 0 deletions pyvene/models/esm/modelings_intervenable_esm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,63 @@
"""
Each modeling file in this library is a mapping between
abstract naming of intervention anchor points and actual
model module defined in the huggingface library.

We also want to let the intervention library know how to
config the dimensions of intervention based on model config
defined in the huggingface library.
"""

from ..constants import (
CONST_INPUT_HOOK,
CONST_OUTPUT_HOOK
)

"""esm base model"""
esm_type_to_module_mapping = dict(
block_input=("encoder.layer[%s]", CONST_INPUT_HOOK),
block_output=("encoder.layer[%s]", CONST_OUTPUT_HOOK),

mlp_input=("encoder.layer[%s].intermediate", CONST_INPUT_HOOK),
mlp_activation=("encoder.layer[%s].intermediate", CONST_OUTPUT_HOOK),
mlp_output=("encoder.layer[%s].output", CONST_OUTPUT_HOOK),

attention_value_output=("encoder.layer[%s].attention.output", CONST_INPUT_HOOK),
head_attention_value_output=("encoder.layer[%s].attention.output", CONST_INPUT_HOOK),

attention_input=("encoder.layer[%s].attention", CONST_INPUT_HOOK),
attention_output=("encoder.layer[%s].attention", CONST_OUTPUT_HOOK),

query_output=("encoder.layer[%s].attention.self.query", CONST_OUTPUT_HOOK),
head_query_output=("encoder.layer[%s].attention.self.query", CONST_OUTPUT_HOOK),
key_output=("encoder.layer[%s].attention.self.key", CONST_OUTPUT_HOOK),
head_key_output=("encoder.layer[%s].attention.self.key", CONST_OUTPUT_HOOK),
value_output=("encoder.layer[%s].attention.self.value", CONST_OUTPUT_HOOK),
head_value_output=("encoder.layer[%s].attention.self.value", CONST_OUTPUT_HOOK),

)
esm_type_to_dimension_mapping = dict(
block_input=("hidden_size",),
block_output=("hidden_size"),

mlp_input=("hidden_size",),
mlp_activation=("intermediate_size",),
mlp_output=("hidden_size",),

attention_value_output=("hidden_size",),
head_attention_value_output=("hidden_size/num_attention_heads",),

attention_input=("num_attention_heads",),
attention_output=("num_attention_heads/num_attention_heads",),

query_output=("num_attention_heads",),
head_query_output=("num_attention_heads/num_attention_heads",),
key_output=("num_attention_heads",),
head_key_output=("num_attention_heads/num_attention_heads",),
value_output=("num_attention_heads",),
head_value_output=("num_attention_heads/num_attention_heads",),
)

"""esm for mlm model"""
esm_mlm_type_to_module_mapping = {k: ("esm." + i, j) for k, (i, j) in esm_type_to_module_mapping.items()}
esm_mlm_type_to_dimension_mapping = esm_type_to_dimension_mapping.copy()
5 changes: 5 additions & 0 deletions pyvene/models/intervenable_modelcard.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
from .backpack_gpt2.modelings_intervenable_backpack_gpt2 import *
from .llava.modelings_intervenable_llava import *
from .olmo.modelings_intervenable_olmo import *
from .esm.modelings_intervenable_esm import *

#########################################################################
"""
Expand Down Expand Up @@ -63,6 +64,8 @@
hf_models.gemma2.modeling_gemma2.Gemma2ForCausalLM: gemma2_lm_type_to_module_mapping,
hf_models.olmo.modeling_olmo.OlmoModel: olmo_type_to_module_mapping,
hf_models.olmo.modeling_olmo.OlmoForCausalLM: olmo_lm_type_to_module_mapping,
hf_models.esm.modeling_esm.EsmModel: esm_type_to_module_mapping,
hf_models.esm.modeling_esm.EsmForMaskedLM: esm_mlm_type_to_module_mapping,
hf_models.blip.modeling_blip.BlipForQuestionAnswering: blip_type_to_module_mapping,
hf_models.blip.modeling_blip.BlipForImageTextRetrieval: blip_itm_type_to_module_mapping,
MLPModel: mlp_type_to_module_mapping,
Expand Down Expand Up @@ -98,6 +101,8 @@
hf_models.gemma2.modeling_gemma2.Gemma2ForCausalLM: gemma2_lm_type_to_dimension_mapping,
hf_models.olmo.modeling_olmo.OlmoModel: olmo_type_to_dimension_mapping,
hf_models.olmo.modeling_olmo.OlmoForCausalLM: olmo_lm_type_to_dimension_mapping,
hf_models.esm.modeling_esm.EsmModel: esm_type_to_dimension_mapping,
hf_models.esm.modeling_esm.EsmForMaskedLM: esm_mlm_type_to_dimension_mapping,
hf_models.blip.modeling_blip.BlipForQuestionAnswering: blip_type_to_dimension_mapping,
hf_models.blip.modeling_blip.BlipForImageTextRetrieval: blip_itm_type_to_dimension_mapping,
MLPModel: mlp_type_to_dimension_mapping,
Expand Down