Fix typos under torch directory (pytorch#88172)

This PR fixes typos in '.md' files under torch directory

Pull Request resolved: pytorch#88172
Approved by: https://github.com/malfet

torch/ao/pruning/_experimental/data_sparsifier/benchmarks/README.md

@@ -5,7 +5,7 @@ The objective of this exercise is to use the data sparsifier to prune the embedd
 
 1. **Disk usage savings**: Savings in model size after pruning.
 2. **Model Quality**: How and by how much does performance deteriorate after pruning the embedding bags?
-3. **Model forward time**: Can we speed up the model forward time by utilizing the sparsity? Specificially, can we introduce torch.sparse interim to reduce number of computations.
+3. **Model forward time**: Can we speed up the model forward time by utilizing the sparsity? Specifically, can we introduce torch.sparse interim to reduce number of computations.
 
 ## Scope
 The [DataNormSparsifier](https://github.com/pytorch/pytorch/blob/master/torch/ao/sparsity/_experimental/data_sparsifier/data_norm_sparsifier.py) is used to sparsify the embeddings of the DLRM model. The model is sparsified for all the combinations of -

torch/ao/quantization/fx/_model_report/README.md

@@ -5,7 +5,7 @@ ModelReport
 
  > ⚠️ *While the example below uses the Fx Workflow, the use of the ModelReport class **does not depend** on the Fx Workflow to work*.
  The requirements are detector dependent.
- Most detectors require a **traceable GraphModule**, but some (ex. `PerChannelDetector`) require just a `nn.Module`.
+ Most detectors require a **traceable GraphModule**, but some (ex. `PerChannelDetector`) require just an `nn.Module`.
 
 #### Typical Fx Workflow
 - Initialize model → Prepare model → Callibrate model → Convert model → ...
@@ -62,7 +62,7 @@ This is so that we can keep track of where we want to insert observers on a dete
 It then returns the GraphModule with the detectors inserted into both the regular module structure as well as the node structure.
 - `generate_model_report(self, remove_inserted_observers: bool)` → `Dict[str, Tuple[str, Dict]]` uses callibrated GraphModule to optionally removes inserted observers, and generate, for each detector the ModelReport instance was initialized with:
   - A string-based report that is easily digestable and actionable explaining the data collected by relevant observers for that detector
-  - A dictionary containing statistics collected by the relevant observers and values calculated by the detector for futher analysis or plotting
+  - A dictionary containing statistics collected by the relevant observers and values calculated by the detector for further analysis or plotting
 
 ## ModelReportVisualizer Overview
 

torch/csrc/jit/OVERVIEW.md

@@ -894,7 +894,7 @@ def LSTMCellS(x, hx, cx, w_ih, w_hh, b_ih, b_hh):
     return hy, cy
 ```
 
-After going through the the frontend, we start with this unoptimized graph:
+After going through the frontend, we start with this unoptimized graph:
 
 ```
 graph(%x : Tensor,

torch/csrc/jit/codegen/onednn/README.md

@@ -104,7 +104,7 @@ with torch.no_grad():
 
 # run the model
 with torch.no_grad():
-    # oneDNN graph fusion will be trigerred during runtime
+    # oneDNN graph fusion will be triggered during runtime
     output = model(images)
 ```
 

torch/csrc/jit/operator_upgraders/README.md

@@ -177,7 +177,7 @@ When making changes to the operators, the first thing to identify is if it's BC/
             except Exception as e:
                 self.skipTest("Failed to load fixture!")
 
-            # Step4. Load the new model and it won't apply the ugprader
+            # Step4. Load the new model and it won't apply the upgrader
             current_mobile_module_float = self._save_load_mobile_module(MyModuleFloat)
             current_server_module_float = self._save_load_module(MyModuleFloat)
 

torch/csrc/jit/runtime/static/README.md

@@ -142,9 +142,9 @@ is selected instead.
 When loading a model, ops are selected for each `torch::jit::Node` in the graph as follows:
 
 1) If an out variant is registered, pass the node to the function that produces the `SROperator`. If
-the result is not `nulltpr`, use that op.
+the result is not `nullptr`, use that op.
 2) If a native function is registered, pass the node to the function that produces the `SROperator`. If
-the result is not `nulltpr`, use that op.
+the result is not `nullptr`, use that op.
 3) Use the JIT implementation. Static runtime will throw an exception if it does not exist.
 
 ## Implementation Details