Fix a couple typos

CHANGELOG.md

@@ -63,7 +63,7 @@ This documents the main changes to the `candle` crate.
   [760](https://github.com/huggingface/candle/pull/760).
 - Add the Segment-Anything Model (SAM) as an example
   [773](https://github.com/huggingface/candle/pull/773).
-- TinyViT backbone for the segemnt anything example
+- TinyViT backbone for the segment anything example
   [787](https://github.com/huggingface/candle/pull/787).
 - Shape with holes support
   [770](https://github.com/huggingface/candle/pull/770).

README.md

@@ -144,8 +144,10 @@ And then head over to
 ## Useful External Resources
 - [`candle-tutorial`](https://github.com/ToluClassics/candle-tutorial): A
   very detailed tutorial showing how to convert a PyTorch model to Candle.
-- [`candle-lora`](https://github.com/EricLBuehler/candle-lora): Efficient and ergonomic LoRA implemenation for Candle. `candle-lora` has      
-  out-of-the-box LoRA support for many models from Candle, which can be found [here](https://github.com/EricLBuehler/candle-lora/tree/master/candle-lora-transformers/examples).
+- [`candle-lora`](https://github.com/EricLBuehler/candle-lora): Efficient and
+  ergonomic LoRA implementation for Candle. `candle-lora` has      
+  out-of-the-box LoRA support for many models from Candle, which can be found
+  [here](https://github.com/EricLBuehler/candle-lora/tree/master/candle-lora-transformers/examples).
 - [`optimisers`](https://github.com/KGrewal1/optimisers): A collection of optimisers
   including SGD with momentum, AdaGrad, AdaDelta, AdaMax, NAdam, RAdam, and RMSprop.
 - [`candle-vllm`](https://github.com/EricLBuehler/candle-vllm): Efficient platform for inference and

candle-core/src/indexer.rs

@@ -64,7 +64,7 @@ impl Tensor {
 #[derive(Debug)]
 /// Generic structure used to index a slice of the tensor
 pub enum TensorIndexer {
-    /// This selects the elemnts for which an index has some specific value.
+    /// This selects the elements for which an index has some specific value.
     Select(usize),
     /// This is a regular slice, purely indexing a chunk of the tensor
     Narrow(Bound<usize>, Bound<usize>),

candle-core/src/quantized/avx.rs

@@ -353,7 +353,7 @@ pub(crate) fn vec_dot_q3k_q8k(n: usize, xs: &[BlockQ3K], ys: &[BlockQ8K]) -> Res
                 q3 = q3.add(32);
 
                 // Prepare low and high bits
-                // We hardcode the shifts here to avoid loading them into a seperate register
+                // We hardcode the shifts here to avoid loading them into a separate register
                 let q3l_0 = _mm256_and_si256(q3bits, m3);
                 let q3h_0 = if j == 0 {
                     _mm256_srli_epi16(_mm256_andnot_si256(hbits, _mm256_slli_epi16(mone, 0)), 0)
@@ -586,7 +586,7 @@ pub(crate) fn vec_dot_q5k_q8k(n: usize, xs: &[BlockQ5K], ys: &[BlockQ8K]) -> Res
                 let q5bits = _mm256_loadu_si256(q5 as *const __m256i);
                 q5 = q5.add(32);
 
-                //Similar to q3k we hardcode the shifts here to avoid loading them into a seperate register
+                //Similar to q3k we hardcode the shifts here to avoid loading them into a separate register
                 let q5l_0 = _mm256_and_si256(q5bits, m4);
                 let q5l_0_shift_input = _mm256_and_si256(hbits, hmask);
                 let q5l_0_right_shift = match j {

candle-core/src/quantized/gguf_file.rs

@@ -463,7 +463,7 @@ impl Content {
     ) -> Result<QTensor> {
         let tensor_info = match self.tensor_infos.get(name) {
             Some(tensor_info) => tensor_info,
-            None => crate::bail!("cannot find tensor-infor for {name}"),
+            None => crate::bail!("cannot find tensor info for {name}"),
         };
         tensor_info.read(reader, self.tensor_data_offset)
     }

candle-core/src/tensor.rs

@@ -1,4 +1,4 @@
-//! Tensors are N-dimenional matrixes of elements using a single data type.
+//! Tensors are N-dimensional matrixes of elements using a single data type.
 #![allow(clippy::redundant_closure_call)]
 use crate::backend::{BackendDevice, BackendStorage};
 use crate::op::{
@@ -669,7 +669,7 @@ impl Tensor {
     }
 
     /// Split a tensor into the specified number of chunks, this may return less chunks than
-    /// specificed.
+    /// specified.
     pub fn chunk<D: Dim>(&self, chunks: usize, dim: D) -> Result<Vec<Self>> {
         let dim = dim.to_index(self.shape(), "chunk")?;
         let size = self.dim(dim)?;

candle-examples/examples/reinforcement-learning/atari_wrappers.py

@@ -78,7 +78,7 @@ def step(self, action):
         # then update lives to handle bonus lives
         lives = self.env.unwrapped.ale.lives()
         if lives < self.lives and lives > 0:
-            # for Qbert somtimes we stay in lives == 0 condtion for a few frames
+            # for Qbert sometimes we stay in lives == 0 condition for a few frames
             # so its important to keep lives > 0, so that we only reset once
             # the environment advertises done.
             done = True

candle-nn/src/var_builder.rs

@@ -40,7 +40,7 @@ struct TensorData<B: Backend> {
 /// A trait that defines how tensor data is retrieved.
 ///
 /// Typically this would use disk storage in some specific format, or random initialization.
-/// Note that there is a speciliazed version of this trait (`SimpleBackend`) that can be used most
+/// Note that there is a specialized version of this trait (`SimpleBackend`) that can be used most
 /// of the time. The main restriction is that it doesn't allow for specific args (besides
 /// initialization hints).
 pub trait Backend: Send + Sync {

candle-pyo3/py_src/candle/__init__.py

@@ -4,7 +4,8 @@
     from .candle import *
 except ImportError as e:
     # If we are in development mode, or we did not bundle the DLLs, we try to locate them here
-    # PyO3 wont give us any infomration about what DLLs are missing, so we can only try to load the DLLs and re-import the module
+    # PyO3 wont give us any information about what DLLs are missing, so we can only try to load
+    # the DLLs and re-import the module
     logging.warning("DLLs were not bundled with this package. Trying to locate them...")
     import os
     import platform

candle-pyo3/py_src/candle/nn/container.py

@@ -363,7 +363,7 @@ def extend(self, modules: Iterable[Module]) -> "ModuleList":
             self.add_module(str(offset + i), module)
         return self
 
-    # remove forward alltogether to fallback on Module's _forward_unimplemented
+    # remove forward altogether to fallback on Module's _forward_unimplemented
 
 
 class ModuleDict(Module):
@@ -480,4 +480,4 @@ def update(self, modules: Mapping[str, Module]) -> None:
                 # that's too cumbersome to type correctly with overloads, so we add an ignore here
                 self[m[0]] = m[1]  # type: ignore[assignment]
 
-    # remove forward alltogether to fallback on Module's _forward_unimplemented
+    # remove forward altogether to fallback on Module's _forward_unimplemented

candle-pyo3/src/lib.rs

@@ -212,7 +212,7 @@ trait MapDType {
 enum Indexer {
     Index(usize),
     Slice(usize, usize),
-    Elipsis,
+    Ellipsis,
     Expand,
     IndexSelect(Tensor),
 }
@@ -568,7 +568,7 @@ impl PyTensor {
                         "Ellipsis ('...') can only be used at the start of an indexing operation",
                     ));
                 }
-                Ok((Indexer::Elipsis, dims.len() - (index_argument_count - 1)))
+                Ok((Indexer::Ellipsis, dims.len() - (index_argument_count - 1)))
             } else if py_indexer.is_none() {
                 // Handle None e.g. tensor[None, 0]
                 Ok((Indexer::Expand, current_dim))
@@ -616,8 +616,9 @@ impl PyTensor {
                     current_dim += 1;
                     out
                 }
-                Indexer::Elipsis => {
-                    // Elipsis is a special case, it means that all remaining dimensions should be selected => advance the current_dim to the last dimension we have indexers for
+                Indexer::Ellipsis => {
+                    // Ellipsis is a special case, it means that all remaining dimensions should be
+                    // selected => advance the current_dim to the last dimension we have indexers for
                     current_dim += dims.len() - (indexers.len() - 1);
                     x
                 }
@@ -960,11 +961,11 @@ impl PyTensor {
             extraction_result: PyResult<T>,
             err_msg: &'static str,
         ) -> PyResult<()> {
-            if let Ok(sucessfull_extraction) = extraction_result {
+            if let Ok(successful_extraction) = extraction_result {
                 if opt.is_some() {
                     return Err(PyValueError::new_err(err_msg));
                 }
-                *opt = Some(sucessfull_extraction);
+                *opt = Some(successful_extraction);
             }
             Ok(())
         }
@@ -1045,9 +1046,7 @@ impl PyTensor {
                 .map_err(wrap_err)?,
             (Some(device), None) => self.0.to_device(&device.as_device()?).map_err(wrap_err)?,
             (None, Some(dtype)) => self.0.to_dtype(dtype.0).map_err(wrap_err)?,
-            (None, None) => {
-                return Err(PyTypeError::new_err("No valide dtype or device specified"))
-            }
+            (None, None) => return Err(PyTypeError::new_err("No valid dtype or device specified")),
         };
 
         Ok(PyTensor(result))

candle-pyo3/stub.py

@@ -156,7 +156,7 @@ def process_additional_function(fn):
         string += function(obj, indent)
 
     elif inspect.isgetsetdescriptor(obj):
-        # TODO it would be interesing to add the setter maybe ?
+        # TODO it would be interesting to add the setter maybe ?
         string += f"{indent}@property\n"
         string += function(obj, indent, text_signature="(self)")
 

candle-pyo3/tests/bindings/test_module.py

@@ -74,7 +74,7 @@ def __init__(self):
     a.load_state_dict(statedict)
 
 
-def test_module_throws_on_shape_missmatch():
+def test_module_throws_on_shape_mismatch():
     class A(Module):
         def __init__(self):
             super().__init__()
@@ -121,7 +121,7 @@ def __init__(self):
     assert a.t.ggml_dtype == "Q4_0"
 
 
-def test_module_dequantizes_tensors_automaticaly():
+def test_module_dequantizes_tensors_automatically():
     class A(Module):
         def __init__(self):
             super().__init__()

candle-pyo3/tests/native/test_tensor.py

@@ -84,7 +84,7 @@ def assert_bool(t: Tensor, expected: bool):
     assert bool(t.values()) == expected
 
 
-def test_tensor_supports_equality_opperations_with_scalars():
+def test_tensor_supports_equality_operations_with_scalars():
     t = Tensor(42.0)
 
     assert_bool(t == 42.0, True)
@@ -106,7 +106,7 @@ def test_tensor_supports_equality_opperations_with_scalars():
     assert_bool(t <= 42.0, True)
 
 
-def test_tensor_supports_equality_opperations_with_tensors():
+def test_tensor_supports_equality_operations_with_tensors():
     t = Tensor(42.0)
     same = Tensor(42.0)
     other = Tensor(43.0)
@@ -130,7 +130,7 @@ def test_tensor_supports_equality_opperations_with_tensors():
     assert_bool(t <= other, True)
 
 
-def test_tensor_equality_opperations_can_broadcast():
+def test_tensor_equality_operations_can_broadcast():
     # Create a decoder attention mask as a test case
     # e.g.
     # [[1,0,0]

candle-transformers/src/models/segment_anything/mask_decoder.rs

@@ -182,7 +182,7 @@ impl MaskDecoder {
         sparse_prompt_embeddings: &Tensor,
         dense_prompt_embeddings: &Tensor,
     ) -> Result<(Tensor, Tensor)> {
-        // Concatenate ouput tokens.
+        // Concatenate output tokens.
         let output_tokens = Tensor::cat(
             &[self.iou_token.embeddings(), self.mask_tokens.embeddings()],
             0,

candle-transformers/src/models/segment_anything/prompt_encoder.rs

@@ -2,11 +2,11 @@ use candle::{DType, IndexOp, Result, Tensor, D};
 use candle_nn::VarBuilder;
 
 #[derive(Debug)]
-struct PostionEmbeddingRandom {
+struct PositionEmbeddingRandom {
     positional_encoding_gaussian_matrix: Tensor,
 }
 
-impl PostionEmbeddingRandom {
+impl PositionEmbeddingRandom {
     fn new(num_pos_feats: usize, vb: VarBuilder) -> Result<Self> {
         let positional_encoding_gaussian_matrix =
             vb.get((2, num_pos_feats), "positional_encoding_gaussian_matrix")?;
@@ -52,7 +52,7 @@ impl PostionEmbeddingRandom {
 
 #[derive(Debug)]
 pub struct PromptEncoder {
-    pe_layer: PostionEmbeddingRandom,
+    pe_layer: PositionEmbeddingRandom,
     point_embeddings: Vec<candle_nn::Embedding>,
     not_a_point_embed: candle_nn::Embedding,
     mask_downscaling_conv1: candle_nn::Conv2d,
@@ -76,7 +76,7 @@ impl PromptEncoder {
         vb: VarBuilder,
     ) -> Result<Self> {
         let num_points_embeddings = 4;
-        let pe_layer = PostionEmbeddingRandom::new(embed_dim / 2, vb.pp("pe_layer"))?;
+        let pe_layer = PositionEmbeddingRandom::new(embed_dim / 2, vb.pp("pe_layer"))?;
         let not_a_point_embed = candle_nn::embedding(1, embed_dim, vb.pp("not_a_point_embed"))?;
         let no_mask_embed = candle_nn::embedding(1, embed_dim, vb.pp("no_mask_embed"))?;
         let cfg = candle_nn::Conv2dConfig {

candle-transformers/src/models/stable_diffusion/euler_ancestral_discrete.rs

@@ -1,6 +1,6 @@
 //! Ancestral sampling with Euler method steps.
 //!
-//! Reference implemenation in Rust:
+//! Reference implementation in Rust:
 //!
 //! https://github.com/pykeio/diffusers/blob/250b9ad1898af41e76a74c0d8d4292652823338a/src/schedulers/euler_ancestral_discrete.rs
 //!
@@ -135,7 +135,7 @@ impl EulerAncestralDiscreteScheduler {
         );
         sigmas_int.push(0.0);
 
-        // standard deviation of the inital noise distribution
+        // standard deviation of the initial noise distribution
         // f64 does not implement Ord such that there is no `max`, so we need to use this workaround
         let init_noise_sigma = *sigmas_int
             .iter()

candle-wasm-examples/llama2-c/src/app.rs

@@ -108,7 +108,7 @@ impl Component for App {
     fn update(&mut self, ctx: &Context<Self>, msg: Self::Message) -> bool {
         match msg {
             Msg::SetModel(md) => {
-                self.status = "weights loaded succesfully!".to_string();
+                self.status = "weights loaded successfully!".to_string();
                 self.loaded = true;
                 console_log!("loaded weights");
                 self.worker.send(WorkerInput::ModelData(md));

candle-wasm-examples/llama2-c/src/worker.rs

@@ -24,7 +24,7 @@ macro_rules! console_log {
 }
 
 // Communication to the worker happens through bincode, the model weights and configs are fetched
-// on the main thread and transfered via the following structure.
+// on the main thread and transferred via the following structure.
 #[derive(Serialize, Deserialize)]
 pub struct ModelData {
     pub tokenizer: Vec<u8>,

candle-wasm-examples/whisper/src/app.rs

@@ -145,7 +145,7 @@ impl Component for App {
     fn update(&mut self, ctx: &Context<Self>, msg: Self::Message) -> bool {
         match msg {
             Msg::SetDecoder(md) => {
-                self.status = "weights loaded succesfully!".to_string();
+                self.status = "weights loaded successfully!".to_string();
                 self.loaded = true;
                 console_log!("loaded weights");
                 self.worker.send(WorkerInput::ModelData(md));

candle-wasm-examples/whisper/src/worker.rs

@@ -414,7 +414,7 @@ pub enum Task {
 }
 
 // Communication to the worker happens through bincode, the model weights and configs are fetched
-// on the main thread and transfered via the following structure.
+// on the main thread and transferred via the following structure.
 #[derive(Serialize, Deserialize)]
 pub struct ModelData {
     pub weights: Vec<u8>,

candle-wasm-examples/yolo/src/app.rs

@@ -146,7 +146,7 @@ impl Component for App {
     fn update(&mut self, ctx: &Context<Self>, msg: Self::Message) -> bool {
         match msg {
             Msg::SetModel(md) => {
-                self.status = "weights loaded succesfully!".to_string();
+                self.status = "weights loaded successfully!".to_string();
                 self.loaded = true;
                 console_log!("loaded weights");
                 self.worker.send(WorkerInput::ModelData(md));

candle-wasm-examples/yolo/src/worker.rs

@@ -21,7 +21,7 @@ macro_rules! console_log {
 }
 
 // Communication to the worker happens through bincode, the model weights and configs are fetched
-// on the main thread and transfered via the following structure.
+// on the main thread and transferred via the following structure.
 #[derive(Serialize, Deserialize)]
 pub struct ModelData {
     pub weights: Vec<u8>,