fix various typos (code & comments)

mkdocs.yml

@@ -1,5 +1,5 @@
 site_name: Refiners
-site_decription: A micro framework on top of PyTorch with first class citizen APIs for foundation model adaptation
+site_description: A micro framework on top of PyTorch with first class citizen APIs for foundation model adaptation
 repo_name: Refiners
 repo_url: https://github.com/finegrain-ai/refiners
 edit_uri: edit/main/docs/

scripts/prepare_test_weights.py

@@ -581,7 +581,7 @@ def convert_sam():
         "convert_segment_anything.py",
         "tests/weights/sam_vit_h_4b8939.pth",
         "tests/weights/segment-anything-h.safetensors",
-        expected_hash="3b73b2fd",
+        expected_hash="b62ad5ed",
     )
 
 

src/refiners/fluxion/layers/chain.py

@@ -215,7 +215,7 @@ def find_state_dict_key(module: Module, /) -> str | None:
     @staticmethod
     def _pretty_print_args(*args: Any) -> str:
         """
-        Flatten nested tuples and print tensors with their shape and other informations.
+        Flatten nested tuples and print tensors with their shape and other information.
         """
 
         def _flatten_tuple(t: Tensor | tuple[Any, ...], /) -> list[Any]:

src/refiners/foundationals/latent_diffusion/image_prompt.py

@@ -447,7 +447,7 @@ def convert_to_grid_features(clip_image_encoder: CLIPImageEncoderH) -> CLIPImage
         assert isinstance(encoder_clone[-3], fl.Lambda)  # pooling (classif token)
         for _ in range(3):
             encoder_clone.pop()
-        transfomer_layers = encoder_clone[-1]
-        assert isinstance(transfomer_layers, fl.Chain) and len(transfomer_layers) == 32
-        transfomer_layers.pop()
+        transformer_layers = encoder_clone[-1]
+        assert isinstance(transformer_layers, fl.Chain) and len(transformer_layers) == 32
+        transformer_layers.pop()
         return encoder_clone

src/refiners/foundationals/segment_anything/mask_decoder.py

@@ -5,7 +5,7 @@
 from refiners.fluxion.context import Contexts
 from refiners.foundationals.segment_anything.transformer import (
     SparseCrossDenseAttention,
-    TwoWayTranformerLayer,
+    TwoWayTransformerLayer,
 )
 
 
@@ -210,7 +210,7 @@ def __init__(
             EmbeddingsAggregator(num_output_mask=num_output_mask),
             Transformer(
                 *(
-                    TwoWayTranformerLayer(
+                    TwoWayTransformerLayer(
                         embedding_dim=embedding_dim,
                         num_heads=8,
                         feed_forward_dim=feed_forward_dim,

src/refiners/foundationals/segment_anything/transformer.py

@@ -116,7 +116,7 @@ def __init__(
         )
 
 
-class TwoWayTranformerLayer(fl.Chain):
+class TwoWayTransformerLayer(fl.Chain):
     def __init__(
         self,
         embedding_dim: int,

src/refiners/training_utils/config.py

@@ -178,7 +178,7 @@ class ModelConfig(BaseModel):
 
 class GyroDropoutConfig(BaseModel):
     total_subnetworks: int = 512
-    concurent_subnetworks: int = 64
+    concurrent_subnetworks: int = 64
     iters_per_epoch: int = 512
     num_features_threshold: float = 5e5
 

src/refiners/training_utils/trainer.py

@@ -213,7 +213,7 @@ def time_elapsed(self) -> int:
         return int(time.time() - self.start_time)
 
     @cached_property
-    def evalution_interval_steps(self) -> int:
+    def evaluation_interval_steps(self) -> int:
         return self.convert_time_unit_to_steps(
             number=self.evaluation_interval["number"], unit=self.evaluation_interval["unit"]
         )
@@ -244,7 +244,7 @@ def done(self) -> bool:
 
     @property
     def is_evaluation_step(self) -> bool:
-        return self.step % self.evalution_interval_steps == 0
+        return self.step % self.evaluation_interval_steps == 0
 
     @property
     def is_checkpointing_step(self) -> bool:

tests/foundationals/segment_anything/test_sam.py

@@ -21,7 +21,7 @@
 from refiners.fluxion.utils import image_to_tensor, load_tensors, no_grad
 from refiners.foundationals.segment_anything.image_encoder import FusedSelfAttention
 from refiners.foundationals.segment_anything.model import SegmentAnythingH
-from refiners.foundationals.segment_anything.transformer import TwoWayTranformerLayer
+from refiners.foundationals.segment_anything.transformer import TwoWayTransformerLayer
 
 # See predictor_example.ipynb official notebook
 PROMPTS: list[SAMPrompt] = [
@@ -188,7 +188,7 @@ def test_two_way_transformer(facebook_sam_h: FacebookSAM) -> None:
     dense_positional_embedding = torch.randn(1, 64 * 64, 256, device=facebook_sam_h.device)
     sparse_embedding = torch.randn(1, 3, 256, device=facebook_sam_h.device)
 
-    refiners_layer = TwoWayTranformerLayer(
+    refiners_layer = TwoWayTransformerLayer(
         embedding_dim=256, feed_forward_dim=2048, num_heads=8, device=facebook_sam_h.device
     )
     facebook_layer = facebook_sam_h.mask_decoder.transformer.layers[1]  # type: ignore