fix configs without #

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "sd-mecha"
-version = "0.0.24"
+version = "0.0.25"
 description = "State dict recipe merger"
 readme = "README.md"
 authors = [{ name = "ljleb" }]

sd_mecha/__main__.py

@@ -115,7 +115,7 @@ def compose(
         f.write(composed_recipe)
 
 
-@click.command(help="Show the available blocks and classes of model architectures")
+@click.command(help="Show the available blocks of model architectures")
 @click.argument("model_arch", type=click.Choice(extensions.model_arch.get_all() + [""], case_sensitive=False), default="")
 @click.option("--verbose", "-v", is_flag=True)
 @click.option("--debug", is_flag=True, help="Print the stacktrace when an error occurs.")
@@ -156,23 +156,6 @@ def info(
                     for b in bs
                     if f"_{component}_block_" in b
                 ], key=lambda t: natural_sort_key(t[0]))),
-            "Classes":
-                sorted(list({
-                    c.split("_", 3)[3]
-                    for cs in model_arch.classes.values()
-                    for c in cs
-                    if f"_{component}_class_" in c
-                }), key=natural_sort_key)
-                if not verbose else
-                dict(sorted([
-                    (c.split("_", 3)[3], sorted([
-                        k for k in model_arch.classes
-                        if c in model_arch.classes[k]
-                    ], key=natural_sort_key))
-                    for cs in model_arch.classes.values()
-                    for c in cs
-                    if f"_{component}_class_" in c
-                ], key=lambda t: natural_sort_key(t[0]))),
         }
         for component in model_arch.components
     }, sort_keys=False))

sd_mecha/extensions/model_arch.py

@@ -46,7 +46,6 @@ def discover_blocks(keys, discovered_block_prefixes, arch_identifier: str):
         for block, prefixes in discovered_block_prefixes.items():
             if any(prefix.match(key) for prefix in prefixes["patterns"]):
                 blocks.setdefault(key, set()).add(block)
-                break
 
     return blocks
 

sd_mecha/merge_methods/__init__.py

@@ -576,15 +576,16 @@ def clamp(
 ) -> Tensor | SameMergeSpace:
     maximums = functools.reduce(torch.maximum, bounds)
     minimums = functools.reduce(torch.minimum, bounds)
-    centers = (maximums + minimums) / 2
+    bounds = torch.stack(bounds)
+    average = bounds.mean(dim=0)
 
     if stiffness:
         smallest_positive = maximums
         largest_negative = minimums
 
         for i, bound in enumerate(bounds):
-            smallest_positive = torch.where((smallest_positive >= bound) & (bound >= centers), bound, smallest_positive)
-            largest_negative = torch.where((largest_negative <= bound) & (bound <= centers), bound, largest_negative)
+            smallest_positive = torch.where((smallest_positive >= bound) & (bound >= average), bound, smallest_positive)
+            largest_negative = torch.where((largest_negative <= bound) & (bound <= average), bound, largest_negative)
 
         maximums = weighted_sum.__wrapped__(maximums, smallest_positive, alpha=stiffness)
         minimums = weighted_sum.__wrapped__(minimums, largest_negative, alpha=stiffness)
@@ -599,9 +600,14 @@ def dropout(  # aka n-supermario
     probability: Hyper = 0.9,
     overlap: Hyper = 1.0,
     overlap_emphasis: Hyper = 0.0,
-    seed: Hyper = None,
+    seed: Hyper = -1,
     **kwargs,
 ) -> Tensor | LiftFlag[MergeSpace.DELTA]:
+    if seed < 0:
+        seed = None
+    else:
+        seed = int(seed)
+
     deltas = torch.stack((delta0,) + deltas)
     rng = np.random.default_rng(seed)
 
@@ -638,11 +644,15 @@ def ties_sum_with_dropout(
     apply_median: Hyper = 0.0,
     eps: Hyper = 1e-6,    
     maxiter: Hyper = 100, 
-    ftol: Hyper =1e-20,
-    seed: Hyper = None,
+    ftol: Hyper = 1e-20,
+    seed: Hyper = -1,
     **kwargs,
 ) -> Tensor | LiftFlag[MergeSpace.DELTA]:
     # Set seed
+    if seed < 0:
+        seed = None
+    else:
+        seed = int(seed)
     torch.manual_seed(seed)
 
     # Under "Dropout", delta will be 0 by definition. Multiply it (Hadamard product) will return 0 also.

sd_mecha/models/sd1_ldm.yaml

@@ -28,22 +28,22 @@ merge:
   unet:
     prefix: model.diffusion_model
     blocks:
-      in0: input_blocks.0.#
-      in3: input_blocks.3.#
-      in6: input_blocks.6.#
-      in9: input_blocks.9.#
+      in0: input_blocks.0
+      in3: input_blocks.3
+      in6: input_blocks.6
+      in9: input_blocks.9
       mid:
-      - middle_block.#
+      - middle_block
       - time_embed
       out11:
-      - output_blocks.11.#
+      - output_blocks.11
       - time_embed
       - out
       in*:
-      - input_blocks.*.#
+      - input_blocks.*
       - time_embed
       out*:
-      - output_blocks.*.#
+      - output_blocks.*
       - time_embed
 
 keys: sd1_ldm_keys.txt

sd_mecha/models/sd3_sgm.yaml

@@ -6,21 +6,21 @@ merge:
     prefix: model.diffusion_model
     blocks:
       in0:
-      - joint_blocks.0.#
+      - joint_blocks.0
       - pos_embed
       - t_embedder
       - x_embedder
       - y_embedder
       - context_embedder
       in23:
-      - joint_blocks.23.#
+      - joint_blocks.23
       - pos_embed
       - t_embedder
       - x_embedder
       - y_embedder
       - final_layer
       in*:
-      - joint_blocks.*.#
+      - joint_blocks.*
       - pos_embed
       - t_embedder
       - x_embedder
@@ -53,12 +53,12 @@ merge:
     prefix: text_encoders.t5xxl.transformer
     blocks:
       in0:
-      - encoder.block.0.#.#
+      - encoder.block.0
       - encoder.embed_tokens
       - shared
       in23:
-      - encoder.block.23.#.#
+      - encoder.block.23
       - encoder.final_layer_norm
-      in*: encoder.block.*.#.#
+      in*: encoder.block.*
 
 keys: sd3_sgm_keys.txt

sd_mecha/models/sdxl_sgm.yaml

@@ -6,27 +6,22 @@ merge:
   unet:
     prefix: model.diffusion_model
     blocks:
-      in0: input_blocks.0.#
-      in3: input_blocks.3.#
+      in0: input_blocks.0
+      in3: input_blocks.3
       in6:
-      - input_blocks.6.#
-      - input_blocks.6.#.transformer_blocks.#
+      - input_blocks.6
       mid:
-      - middle_block.#
-      - middle_block.#.transformer_blocks.#
+      - middle_block
       - time_embed
       out8:
-      - output_blocks.8.#
-      - output_blocks.8.#.transformer_blocks.#
+      - output_blocks.8
       - time_embed
       - out
       in*:
-      - input_blocks.*.#
-      - input_blocks.*.#.transformer_blocks.#
+      - input_blocks.*
       - time_embed
       out*:
-      - output_blocks.*.#
-      - output_blocks.*.#.transformer_blocks.#
+      - output_blocks.*
       - time_embed
 
   txt: