neurovlm.training.BrainToTextGenerationTrainConfig

neurovlm.training.BrainToTextGenerationTrainConfig#

class neurovlm.training.BrainToTextGenerationTrainConfig(output_root: 'str | Path' = 'runs', run_id: 'str | None' = None, seed: 'int' = 42, device: 'str' = 'auto', epochs: 'int' = 10, batch_size: 'int' = 32, eval_batch_size: 'int' = 8, num_workers: 'int' = 0, learning_rate: 'float' = 0.0001, weight_decay: 'float' = 0.01, gradient_clip: 'float | None' = 1.0, early_stopping_patience: 'int | None' = None, early_stopping_min_delta: 'float' = 0.0, max_train_batches: 'int | None' = None, max_eval_batches: 'int | None' = None, resume: 'str | Path | None' = None, initialization: 'str' = 'released', qformer_checkpoint: 'str | Path | None' = None, qformer_resource: 'str' = 'neurovlm/NeuroQformer', lm_resource: 'str' = 'neurovlm/NeuroQwen3-0.6B', preset: 'str' = 'retained', image_dim: 'int' = 384, semantic_dim: 'int' = 384, lm_dim: 'int' = 1024, num_queries: 'int' = 32, hidden_dim: 'int' = 512, num_heads: 'int' = 8, num_layers: 'int' = 6, dropout: 'float' = 0.05, projection_temp: 'float | None' = 0.05, canonical_basis: 'str' = 'all', use_canonical_projection: 'bool' = True, train_image_projection: 'bool' = False, pad_token_id: 'int | None' = None, generated_samples_limit: 'int' = 0, primary_metric: 'str' = 'val_loss', metric_direction: 'MetricDirection' = <MetricDirection.MIN: 'min'>)[source]#
Parameters:

  • output_root (str | Path)

  • run_id (str | None)

  • seed (int)

  • device (str)

  • epochs (int)

  • batch_size (int)

  • eval_batch_size (int)

  • num_workers (int)

  • learning_rate (float)

  • weight_decay (float)

  • gradient_clip (float | None)

  • early_stopping_patience (int | None)

  • early_stopping_min_delta (float)

  • max_train_batches (int | None)

  • max_eval_batches (int | None)

  • resume (str | Path | None)

  • initialization (str)

  • qformer_checkpoint (str | Path | None)

  • qformer_resource (str)

  • lm_resource (str)

  • preset (str)

  • image_dim (int)

  • semantic_dim (int)

  • lm_dim (int)

  • num_queries (int)

  • hidden_dim (int)

  • num_heads (int)

  • num_layers (int)

  • dropout (float)

  • projection_temp (float | None)

  • canonical_basis (str)

  • use_canonical_projection (bool)

  • train_image_projection (bool)

  • pad_token_id (int | None)

  • generated_samples_limit (int)

  • primary_metric (str)

  • metric_direction (MetricDirection)

__init__(output_root='runs', run_id=None, seed=42, device='auto', epochs=10, batch_size=32, eval_batch_size=8, num_workers=0, learning_rate=0.0001, weight_decay=0.01, gradient_clip=1.0, early_stopping_patience=None, early_stopping_min_delta=0.0, max_train_batches=None, max_eval_batches=None, resume=None, initialization='released', qformer_checkpoint=None, qformer_resource='neurovlm/NeuroQformer', lm_resource='neurovlm/NeuroQwen3-0.6B', preset='retained', image_dim=384, semantic_dim=384, lm_dim=1024, num_queries=32, hidden_dim=512, num_heads=8, num_layers=6, dropout=0.05, projection_temp=0.05, canonical_basis='all', use_canonical_projection=True, train_image_projection=False, pad_token_id=None, generated_samples_limit=0, primary_metric='val_loss', metric_direction=MetricDirection.MIN)#
Parameters:

  • output_root (str | Path)

  • run_id (str | None)

  • seed (int)

  • device (str)

  • epochs (int)

  • batch_size (int)

  • eval_batch_size (int)

  • num_workers (int)

  • learning_rate (float)

  • weight_decay (float)

  • gradient_clip (float | None)

  • early_stopping_patience (int | None)

  • early_stopping_min_delta (float)

  • max_train_batches (int | None)

  • max_eval_batches (int | None)

  • resume (str | Path | None)

  • initialization (str)

  • qformer_checkpoint (str | Path | None)

  • qformer_resource (str)

  • lm_resource (str)

  • preset (str)

  • image_dim (int)

  • semantic_dim (int)

  • lm_dim (int)

  • num_queries (int)

  • hidden_dim (int)

  • num_heads (int)

  • num_layers (int)

  • dropout (float)

  • projection_temp (float | None)

  • canonical_basis (str)

  • use_canonical_projection (bool)

  • train_image_projection (bool)

  • pad_token_id (int | None)

  • generated_samples_limit (int)

  • primary_metric (str)

  • metric_direction (MetricDirection)

Return type:

None

Methods

__init__([output_root, run_id, seed, ...])

architecture()

Attributes

batch_size

canonical_basis

device

dropout

early_stopping_min_delta

early_stopping_patience

epochs

eval_batch_size

generated_samples_limit

gradient_clip

hidden_dim

image_dim

initialization

learning_rate

lm_dim

lm_resource

max_eval_batches

max_train_batches

metric_direction

num_heads

num_layers

num_queries

num_workers

output_root

pad_token_id

preset

primary_metric

projection_temp

qformer_checkpoint

qformer_resource

resume

run_id

seed

semantic_dim

train_image_projection

use_canonical_projection

weight_decay