Functions — github.com/arcprize/hierarchical-reasoning-model-analysis

hub / github.com/arcprize/hierarchical-reasoning-model-analysis / functions

Functions116 in github.com/arcprize/hierarchical-reasoning-model-analysis

⨍ Functions 116◇ Types & classes 38

↓ 9 callersMethodload

(filename, callback, fetchArgs)

assets/npyjs.js:155

↓ 8 callersFunctiongrid_hash

(grid: np.ndarray)

dataset/build_arc_dataset.py:79

↓ 6 callersFunctionarc_grid_to_np

(grid: List[List[int]])

dataset/build_arc_dataset.py:41

↓ 6 callersFunctiontrunc_normal_init_

(tensor: torch.Tensor, std: float = 1.0, lower: float = -2.0, upper: float = 2.0)

models/common.py:7

↓ 4 callersMethod__init__

(self, hidden_size: int, expansion: float)

models/layers.py:140

↓ 4 callersFunctioncreate_dataloader

(config: PretrainConfig, split: str, rank: int, world_size: int, **kwargs)

pretrain.py:96

↓ 4 callersFunctiondihedral_transform

8 dihedral symmetries by rotate, flip and mirror

dataset/common.py:27

↓ 4 callersFunctionrms_norm

(hidden_states: torch.Tensor, variance_epsilon: float)

models/layers.py:152

↓ 3 callersMethod__init__

(self, config_dict: dict)

models/hrm/hrm_act_v2.py:250

↓ 3 callersMethod__init__

(self, config_dict: dict)

models/hrm/hrm_act_v1.py:221

↓ 3 callersFunctionload_model_class

(identifier: str, prefix: str = "models.")

utils/functions.py:5

↓ 2 callersMethod_collate_batch

(self, batch)

puzzle_dataset.py:98

↓ 2 callersFunction_crop

Find maximum-sized rectangle without any EOS token inside.

evaluators/arc.py:15

↓ 2 callersFunction_map_grid

(grid: np.ndarray)

dataset/build_arc_dataset.py:107

↓ 2 callersFunction_seq_to_numpy

(seq)

dataset/build_maze_dataset.py:91

↓ 2 callersFunction_seq_to_numpy

(seq)

dataset/build_sudoku_dataset.py:117

↓ 2 callersFunctionapply_transformation

(x: np.ndarray)

dataset/build_sudoku_dataset.py:50

↓ 2 callersFunctionconvert_subset

(set_name: str, config: DataProcessConfig)

dataset/build_maze_dataset.py:32

↓ 2 callersFunctionconvert_subset

(set_name: str, config: DataProcessConfig)

dataset/build_sudoku_dataset.py:62

↓ 2 callersFunctioncreate_evaluators

(config: PretrainConfig, eval_metadata: PuzzleDatasetMetadata)

pretrain.py:248

↓ 2 callersFunctioncreate_model

(config: PretrainConfig, train_metadata: PuzzleDatasetMetadata, rank: int, world_size: int)

pretrain.py:109

↓ 2 callersFunctionevaluate

( config: PretrainConfig, train_state: TrainState, eval_loader: torch.utils.data.DataLoader, e

pretrain.py:330

↓ 2 callersFunctionevaluate_checkpoint

Evaluate a trained model checkpoint on a specified dataset. Args: checkpoint_path: Path to the model checkpoint data_pat

evaluate_trained_model.py:97

↓ 2 callersFunctionget_model_source_path

(identifier: str, prefix: str = "models.")

utils/functions.py:15

↓ 2 callersMethodinitial_carry

(self, *args, **kwargs)

models/losses.py:46

↓ 2 callersFunctionpuzzle_hash

(puzzle: dict)

dataset/build_arc_dataset.py:89

↓ 2 callersFunctionrotate_half

Rotates half the hidden dims of the input.

models/layers.py:23

↓ 1 callersMethod__init__

(self, num_embeddings: int, embedding_dim: int, batch_size: int, init_std: float, cast_to: torch.dtype)

models/sparse_embedding.py:12

↓ 1 callersFunction_find_multiple

(a, b)

models/layers.py:19

↓ 1 callersMethod_input_embeddings

(self, input: torch.Tensor, puzzle_identifiers: torch.Tensor)

models/hrm/hrm_act_v2.py:183

↓ 1 callersMethod_input_embeddings

(self, input: torch.Tensor, puzzle_identifiers: torch.Tensor)

models/hrm/hrm_act_v1.py:148

↓ 1 callersMethod_iter_test

(self)

puzzle_dataset.py:121

↓ 1 callersMethod_iter_train

(self)

puzzle_dataset.py:154

↓ 1 callersMethod_lazy_load_dataset

(self)

puzzle_dataset.py:75

↓ 1 callersMethod_load_metadata

(self)

puzzle_dataset.py:71

↓ 1 callersFunction_sample_batch

(rng: np.random.Generator, group_order: np.ndarray, puzzle_indices: np.ndarray, group_indices: np.ndarray, sta

puzzle_dataset.py:14

↓ 1 callersFunction_sparse_emb_signsgd_dist

( local_weights_grad: torch.Tensor, local_ids: torch.Tensor, weights: torch.Tensor, lr: f

models/sparse_embedding.py:98

↓ 1 callersFunctionapply_rotary_pos_emb

(q: torch.Tensor, k: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor)

models/layers.py:30

↓ 1 callersFunctionaug

(name: str)

dataset/build_arc_dataset.py:100

↓ 1 callersMethodbegin_eval

(self)

evaluators/arc.py:60

↓ 1 callersFunctioncompute_lr

(base_lr: float, config: PretrainConfig, train_state: TrainState)

pretrain.py:238

↓ 1 callersFunctionconvert_dataset

(config: DataProcessConfig)

dataset/build_arc_dataset.py:226

↓ 1 callersFunctionconvert_single_arc_puzzle

(results: dict, name: str, puzzle: dict, aug_count: int, dest_mapping: Dict[str, Tuple[str, str]])

dataset/build_arc_dataset.py:128

↓ 1 callersFunctionconvert_to_serializable

Convert numpy types to Python native types for JSON serialization

evaluate_trained_model.py:283

↓ 1 callersFunctioncosine_schedule_with_warmup_lr_lambda

( current_step: int, *, base_lr: float, num_warmup_steps: int, num_training_steps: int,

pretrain.py:159

↓ 1 callersMethodempty_carry

(self, batch_size: int)

models/hrm/hrm_act_v2.py:207

↓ 1 callersMethodempty_carry

(self, batch_size: int)

models/hrm/hrm_act_v1.py:170

↓ 1 callersMethodfloat16ToFloat32

(float16)

assets/npyjs.js:89

↓ 1 callersFunctioninit_train_state

( config: PretrainConfig, train_metadata: PuzzleDatasetMetadata, rank: int, world_size: int )

pretrain.py:178

↓ 1 callersFunctioninverse_aug

(name: str)

dataset/build_arc_dataset.py:113

↓ 1 callersFunctioninverse_dihedral_transform

(arr: np.ndarray, tid: int)

dataset/common.py:50

↓ 1 callersFunctionlaunch

(hydra_config: DictConfig)

pretrain.py:519

↓ 1 callersFunctionload_checkpoint

(model: nn.Module, config: PretrainConfig)

pretrain.py:213

↓ 1 callersFunctionload_config_from_checkpoint

Load the config from checkpoint directory's config files.

evaluate_trained_model.py:69

↓ 1 callersFunctionload_puzzles_arcagi

(config: DataProcessConfig)

dataset/build_arc_dataset.py:169

↓ 1 callersFunctionload_synced_config

(hydra_config: DictConfig, rank: int, world_size: int)

pretrain.py:497

↓ 1 callersFunctionlog_stablemax

(x, dim=-1)

models/losses.py:19

↓ 1 callersFunctionmain

()

evaluate_trained_model.py:326

↓ 1 callersFunctionmain

()

run_augmentation_ablation_eval.py:198

↓ 1 callersFunctionnp_grid_to_seq_translational_augment

(inp: np.ndarray, out: np.ndarray, do_translation: bool)

dataset/build_arc_dataset.py:52

↓ 1 callersMethodparse

(arrayBufferContents)

assets/npyjs.js:116

↓ 1 callersMethodpuzzle_emb

(self)

models/hrm/hrm_act_v2.py:256

↓ 1 callersMethodpuzzle_emb

(self)

models/hrm/hrm_act_v1.py:227

↓ 1 callersMethodreset_carry

(self, reset_flag: torch.Tensor, carry: HierarchicalReasoningModel_ACTV2InnerCarry)

models/hrm/hrm_act_v2.py:217

↓ 1 callersMethodreset_carry

(self, reset_flag: torch.Tensor, carry: HierarchicalReasoningModel_ACTV1InnerCarry)

models/hrm/hrm_act_v1.py:176

↓ 1 callersMethodresult

(self, save_path: Optional[str], rank: int, world_size: int, group: Optional[torch.distributed.ProcessGroup] =

evaluators/arc.py:104

↓ 1 callersFunctionrun_augmentation_ablation

Run augmentation ablation study using the existing evaluation infrastructure. For each augmentation count, creates a custom evaluator co

run_augmentation_ablation_eval.py:29

↓ 1 callersFunctions

(x, epsilon=1e-30)

models/losses.py:11

↓ 1 callersFunctionsave_code_and_config

(config: PretrainConfig)

pretrain.py:474

↓ 1 callersFunctionsave_train_state

(config: PretrainConfig, train_state: TrainState)

pretrain.py:202

↓ 1 callersFunctionsetup_distributed

Initialize distributed training if in distributed environment.

evaluate_trained_model.py:47

↓ 1 callersFunctionshuffle_sudoku

(board: np.ndarray, solution: np.ndarray)

dataset/build_sudoku_dataset.py:29

↓ 1 callersMethodstep

(self, closure=None)

models/sparse_embedding.py:63

↓ 1 callersFunctiontrain_batch

( config: PretrainConfig, train_state: TrainState, batch: Any, global_batch_size: int, ran

pretrain.py:260

↓ 1 callersMethodupdate_batch

(self, batch: Dict[str, torch.Tensor], preds: Dict[str, torch.Tensor])

evaluators/arc.py:66

Method__init__

(self, config: PuzzleDatasetConfig, split: str = "train")

puzzle_dataset.py:54

Method__init__

(self, data_path: str, eval_metadata: PuzzleDatasetMetadata, submission_K: int = 2, pass_Ks: Sequence[int] = (

evaluators/arc.py:43

Method__init__

( self, data_path: str, eval_metadata, submission_K: int = 2, pass_Ks:

evaluators/arc_augmentation_ablation.py:29

Method__init__

( self, params: ParamsT, world_size: int, lr: Union[float, torch.Tensor] = 1e

models/sparse_embedding.py:42

Method__init__

(self, in_features: int, out_features: int, bias: bool)

models/layers.py:44

Method__init__

(self, num_embeddings: int, embedding_dim: int, init_std: f

models/layers.py:63

Method__init__

(self, dim, max_position_embeddings, base, device=None)

models/layers.py:81

Method__init__

(self, hidden_size, head_dim, num_heads, num_key_value_heads, causal=False)

models/layers.py:99

Method__init__

(self, model: nn.Module, loss_type: str)

models/losses.py:41

Method__init__

(self, config: HierarchicalReasoningModel_ACTV2Config)

models/hrm/hrm_act_v2.py:75

Method__init__

(self, layers: List[HierarchicalReasoningModel_ACTV2Block])

models/hrm/hrm_act_v2.py:104

Method__init__

(self, config: HierarchicalReasoningModel_ACTV2Config)

models/hrm/hrm_act_v2.py:120

Method__init__

(self, config: HierarchicalReasoningModel_ACTV1Config)

models/hrm/hrm_act_v1.py:63

Method__init__

(self, layers: List[HierarchicalReasoningModel_ACTV1Block])

models/hrm/hrm_act_v1.py:89

Method__init__

(self, config: HierarchicalReasoningModel_ACTV1Config)

models/hrm/hrm_act_v1.py:105

Method__iter__

(self)

puzzle_dataset.py:192

Methodconstructor

(opts)

assets/npyjs.js:3

Methodfloat16ToFloat32Array

(float16Array)

assets/npyjs.js:78

Methodforward

(self, inputs: torch.Tensor)

models/sparse_embedding.py:28

Methodforward

(self, input: torch.Tensor)

models/layers.py:58

Methodforward

(self, input: torch.Tensor)

(self, cos_sin: CosSin, hidden_states: torch.Tensor)

( self, return_keys: Set[str], # Model args **model_kwargs, )

models/losses.py:49

next →1–100 of 116, ranked by callers