From 12233df6ec1cbc3811fa83d567c0f66fa5389739 Mon Sep 17 00:00:00 2001
From: wassname <1103714+wassname@users.noreply.github.com>
Date: Tue, 13 Jan 2026 09:00:06 +0800
Subject: [PATCH] restore alt subspaces

---
 antipasto/config.py                       | 13 +++++++++++--
 antipasto/peft_utils/antipasto_adapter.py |  7 +------
 antipasto/train/model_setup.py            | 15 ++++++++++-----
 3 files changed, 22 insertions(+), 13 deletions(-)

diff --git a/antipasto/config.py b/antipasto/config.py
index f325da1..d036ae6 100644
--- a/antipasto/config.py
+++ b/antipasto/config.py
@@ -149,7 +149,11 @@ class TrainingConfig:
 
     loss_subspace: Literal[
         # Recommended (default)
-        "taskdiff_x_suppressed_x_write",  # Task-discriminative ∩ suppressed ∩ write
+        "taskdiff_x_suppressed_x_write",  # Task-discriminative ∩ suppressed ∩ write (empirical)
+        # Static alternatives (model-intrinsic, not task-specific base)
+        "taskdiff_x_logits_read",  # Task signal that AFFECTS output (opposite of suppressed)
+        "taskdiff_x_write_not_read",  # Task signal in static write-not-read space
+        "taskdiff_x_write_x_notlogits",  # Task ∩ write ∩ (lm_head^⊥) 
         # Simpler alternatives
         "write",  # Write space only (o_proj, down_proj column space)
         "taskdiff",  # Task-discriminative PCA only
@@ -158,8 +162,13 @@ class TrainingConfig:
     
     Default: taskdiff_x_suppressed_x_write = intersection of:
     - taskdiff: PCA on cho-rej difference (task-discriminative directions)
-    - suppressed: Written to residual but erased by later layers  
+    - suppressed: Written to residual but erased by later layers (EMPIRICAL)
     - write: Column space of o_proj and down_proj (writable by model)
+    
+    Alternative static (model-intrinsic) options:
+    - taskdiff_x_logits_read: Task signal readable by lm_head (affects output)
+    - taskdiff_x_write_not_read: Task signal in static write-not-read space
+    - taskdiff_x_write_x_notlogits: Task ∩ write ∩ (lm_head^⊥)
     """
 
     loss_subspace_rank: Optional[int] = 8
diff --git a/antipasto/peft_utils/antipasto_adapter.py b/antipasto/peft_utils/antipasto_adapter.py
index d61e04a..ea805ab 100644
--- a/antipasto/peft_utils/antipasto_adapter.py
+++ b/antipasto/peft_utils/antipasto_adapter.py
@@ -339,11 +339,6 @@ class AntiPaSTOLayer(BaseTunerLayer):
         """
         # Soft clamp rotation angle: small θ ensures R(θ)@S ≈ -R(-θ)@S (first-order approx)
         # This gives additive output symmetry: Δy(+1) ≈ -Δy(-1) around base model
-
-        # if max_angle is not None and max_angle < float('inf'):
-        #     A_clamped = max_angle * torch.tanh(A / max_angle)
-        # else:
-        #     A_clamped = A
         
         if max_angle is not None and max_angle < (torch.pi - 1e-6):
             # Convert desired max rotation angle to A-space limit
@@ -446,7 +441,7 @@ class AntiPaSTOLayer(BaseTunerLayer):
         if not torch.isfinite(U_rot).all():
             raise ValueError(f"NaNs in U_rot for adapter {adapter}. alpha={alpha}, max_angle={max_angle}")
         if not torch.isfinite(S_scaled).all():
-            raise ValueError(f"NaNs in S_scaled for adapter {adapter}. scale_mode={scale_mode}")
+            raise ValueError(f"NaNs in S_scaled for adapter {adapter}.")
 
         # Efficient forward: x @ V_rot @ diag(S_scaled) @ U_rot^T
         x_projected = x @ V_rot  # [..., r]
diff --git a/antipasto/train/model_setup.py b/antipasto/train/model_setup.py
index fe14986..1b34c7a 100644
--- a/antipasto/train/model_setup.py
+++ b/antipasto/train/model_setup.py
@@ -145,19 +145,24 @@ def compute_loss_subspace_basis(
     Returns:
         L_subspace: [d_model, top_k] tensor
     
-    Note: Many loss_subspace options were removed in Jan 2026 cleanup.
-    Only taskdiff_x_suppressed_x_write, write, taskdiff are supported.
-    See git history for steer*, null, notlogits, weight_svd implementations.
+    Supported loss_subspace options:
+    - taskdiff_x_suppressed_x_write (default): empirical stenographic + writable
+    - taskdiff_x_logits_read: task signal that affects lm_head output
+    - taskdiff_x_write_not_read: task signal in static write-not-read space  
+    - taskdiff_x_write_x_notlogits: task ∩ write ∩ (lm_head^⊥)
+    - write, taskdiff: basic building blocks
     """
     top_k = config.loss_subspace_rank  # None means auto by energy
     
     # Get from cache - SubspaceCache now stores Subspace objects
     L_cached_sub = subspaces.get(config.loss_subspace)
     if L_cached_sub is None:
+        available = list(subspaces._subspaces.keys())
         raise ValueError(
             f"Subspace '{config.loss_subspace}' not found in cache. "
-            f"Available: {list(subspaces._subspaces.keys())}. "
-            f"Valid options: taskdiff_x_suppressed_x_write, write, taskdiff"
+            f"Available: {available}. "
+            f"Valid options: taskdiff_x_suppressed_x_write, taskdiff_x_logits_read, "
+            f"taskdiff_x_write_not_read, taskdiff_x_write_x_notlogits, write, taskdiff"
         )
     
     L_cached = L_cached_sub.V