PoliBa-DeepLearning/NanoSocrates
2
0
Fork 0
Code Issues 3 Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Merge branch 'dev.train' of https://repositories.communitynotfound.work/PoliBa-DeepLearning/NanoSocrates into dev.train

Browse Source
This commit is contained in:
GassiGiuseppe
2025-10-11 11:28:35 +02:00
parent 82462078f8 625f79f7c3
commit 586f021276
7 changed files with 14 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
Project_Model/Libs/Transformer/Utils/__init__.py
View File

@@ -4,7 +4,7 @@ from .post_tokenization import truncate_sequence, pad_sequence, normalize_sequen
from .inference_masking import inference_masking
from .truncate_rdf_list import truncate_rdf_list
from .decode_out import tensor2token
from .model_utils import decompose_nano_socrates, create_standalone_model
__all__ = [
"TaskType",
@@ -17,6 +17,4 @@ __all__ = [
"inference_masking",
"truncate_rdf_list",
"tensor2token",
"decompose_nano_socrates",
"create_standalone_model"
]

53
Project_Model/Libs/Transformer/Utils/model_utils.py
View File

@@ -1,53 +0,0 @@
import torch
from Project_Model.Libs.Embedder import NanoSocratesEmbedder
from ..Models import TrainingModel, NanoSocraDecoder, NanoSocratEncoder
from ..Classes import DeToken, Encoder, Decoder
from ..Enums import ModelType
def decompose_nano_socrates(
model: TrainingModel, vocabulary_size: int, embedding_size: int
) -> tuple[TrainingModel, NanoSocratEncoder, NanoSocraDecoder]:
encoder_pieces, decoder_pieces = model.take_pieces()
encoder_embedder, encoder = encoder_pieces
encoder_detokener = DeToken(embedding_size, vocabulary_size)
decoder_embedder, decoder, decoder_detokener = decoder_pieces
return (
model,
NanoSocratEncoder(encoder_embedder, encoder, encoder_detokener),
NanoSocraDecoder(decoder_embedder, decoder, decoder_detokener),
)
def create_standalone_model(
model_type: ModelType,
vocabulary_size: int,
latent_space: int = 256,
feed_forward_multiplier: int = 4,
attention_heads: int = 4,
layer_number: int = 2,
) -> NanoSocratEncoder | NanoSocraDecoder:
feed_forward_latent_space = latent_space * feed_forward_multiplier
embedder = NanoSocratesEmbedder(vocabulary_size, latent_space)
detokener = DeToken(latent_space, vocabulary_size)
if model_type == ModelType.ENCODER_ONLY:
TMP_ENCODERS = [
Encoder(latent_space, feed_forward_latent_space, attention_heads)
] * layer_number
encoder = torch.nn.Sequential(*TMP_ENCODERS)
return NanoSocratEncoder(embedder, encoder, detokener)
TMP_DECODERS = [
Decoder(latent_space, feed_forward_latent_space, attention_heads)
] * layer_number
decoder = torch.nn.Sequential(*TMP_DECODERS)
return NanoSocraDecoder(embedder, decoder, detokener)