feat: Geometric Turing Machine v4 for ARC-AGI

conf/task/gtm.yaml

@@ -0,0 +1,76 @@
+# @package _global_
+name: gtm
+
+algebra:
+  p: 3
+  q: 0
+  r: 1
+  device: cuda
+
+model:
+  channels: 32
+  num_steps: 12
+  max_steps: 24
+  num_hypotheses: 8
+  coord_scale: 1.0
+  head_hidden: 128
+  num_rule_slots: 8
+  num_memory_channels: 4
+  weight_share_steps: false
+
+  log_manifold:
+    gate_init: -5.0
+
+  search_plane:
+    conviction_threshold: 0.9
+    evolve_hidden: 64
+
+  info_geometry:
+    halt_eps: 0.01
+    use_supervised_fim: true
+
+  action_engine:
+    gate_init: 0.0
+
+  attention:
+    num_heads: 4
+    head_dim: 8
+
+dataset:
+  data_dir: data/arc
+  include_toy: true
+  toy_n_examples: 20000
+  toy_max_grid_size: 15
+  num_demos: 3
+  epoch_samples: 4000
+
+training:
+  epochs: 150
+  lr: 0.0005
+  batch_size: 16
+  optimizer_type: riemannian_adam
+  max_bivector_norm: 10.0
+  grad_clip: 1.0
+
+  # CUDA acceleration
+  num_workers: 4
+  pin_memory: true
+  amp: true
+  compile: false
+  cudnn_benchmark: true
+
+  # Three-phase schedule
+  warmup_epochs: 8
+  trim_epochs: 72
+  act_epochs: 70
+
+  # Temperature schedule
+  tau_start: 1.0
+  tau_mid: 0.5
+  tau_end: 0.2
+
+  # Loss weights
+  ortho_weight: 0.005
+  gate_entropy_weight: 0.01
+  info_gain_weight: 0.01
+  eval_every: 5

core/metric.py

@@ -94,8 +94,7 @@ def induced_norm(algebra: CliffordAlgebra, A: torch.Tensor) -> torch.Tensor:
     sq_norm = inner_product(algebra, A, A_rev)
 
     # In mixed signatures, sq_norm can be negative.
-    # We return sqrt(|sq_norm|)
-    return torch.sqrt(torch.abs(sq_norm))
+    return torch.sqrt(torch.abs(sq_norm).clamp(min=1e-12))
 
 def geometric_distance(algebra: CliffordAlgebra, A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:
     """Computes geometric distance.
@@ -238,7 +237,8 @@ def hermitian_norm(algebra: CliffordAlgebra, A: torch.Tensor) -> torch.Tensor:
         Norm [..., 1]. Always >= 0.
     """
     sq = hermitian_inner_product(algebra, A, A)
-    return torch.sqrt(torch.abs(sq))
+    # Clamp before sqrt to avoid inf gradient when sq ≈ 0 (e.g. null multivectors in PGA).
+    return torch.sqrt(torch.abs(sq).clamp(min=1e-12))
 
 
 def hermitian_distance(algebra: CliffordAlgebra, A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:
@@ -277,7 +277,7 @@ def hermitian_angle(algebra: CliffordAlgebra, A: torch.Tensor, B: torch.Tensor)
     sq_b = (signs * B * B).sum(dim=-1, keepdim=True)
     # Use sqrt(sq_a * sq_b) instead of sqrt(sq_a)*sqrt(sq_b) to avoid
     # float32 precision loss from two separate sqrt operations.
-    denom = torch.sqrt(torch.abs(sq_a) * torch.abs(sq_b)).clamp(min=1e-6)
+    denom = torch.sqrt((torch.abs(sq_a) * torch.abs(sq_b)).clamp(min=1e-12)).clamp(min=1e-6)
     cos_theta = ip / denom
     cos_theta = torch.clamp(cos_theta, -1.0, 1.0)
     return torch.acos(cos_theta)

datalib/__init__.py

@@ -11,6 +11,7 @@
 from .md17 import get_md17_loaders
 from .deap import DEAPDataset, get_deap_loaders, get_group_sizes
 from .lqa import CLUTRRDataset, HANSDataset, BoolQNegDataset, get_lqa_loaders
+from .arc import ToyARCDataset, ARCDataset, get_arc_loaders
 
 __all__ = [
     "SRDataset",
@@ -27,4 +28,7 @@
     "HANSDataset",
     "BoolQNegDataset",
     "get_lqa_loaders",
+    "ToyARCDataset",
+    "ARCDataset",
+    "get_arc_loaders",
 ]