fix(docs): fix typos in comments and docstrings (#34836)

Signed-off-by: machov <mv1742@nyu.edu>

vllm/compilation/backends.py

@@ -285,7 +285,7 @@ class CompilerManager:
         with self.compile_context(compile_range):
             # There is a compilation time optimization here.
             #
-            # If the (input metdata, graph, compiler config) are the same, then
+            # If the (input metadata, graph, compiler config) are the same, then
             # we want to avoid compiling the same artifact again. If we didn't
             # do this optimization, the backend compilation (InductorAdaptor or
             # InductorStandaloneAdaptor)

vllm/model_executor/layers/fused_moe/oracle/fp8.py

@@ -420,7 +420,7 @@ def make_fp8_moe_quant_config(
     per_out_ch_quant: bool = False,
 ) -> FusedMoEQuantConfig | None:
     """
-    Create FusedMoEQuantConfig for the specifed FP8 Backend.
+    Create FusedMoEQuantConfig for the specified FP8 Backend.
     The FusedMoEQuantConfig holds the scales that are used
     at runtime by the Modular Kernel abstraction.
 

vllm/model_executor/layers/fused_moe/runner/default_moe_runner.py

@@ -151,7 +151,7 @@ class DefaultMoERunner(MoERunner):
     kernels for different parallel execution modes.
 
     Eventually, this class will be split up and specialized for different
-    configurations, e.g. the presense or absence of shared experts, a gate, etc.
+    configurations, e.g. the presence or absence of shared experts, a gate, etc.
     """
 
     def __init__(

vllm/model_executor/models/gpt_oss.py

@@ -586,7 +586,7 @@ class GptOssModel(nn.Module):
                 parts = name.split(".")
                 ids = [s for s in parts if s.isdigit()]
 
-                # for amd-quark format that each expert is seperated
+                # for amd-quark format that each expert is separated
                 # need to extract the parameter name with experts fused.
                 # example model: amd/gpt-oss-20b-MoE-Quant-W-MXFP4-A-FP8-KV-FP8
                 if len(ids) == 2:

vllm/utils/torch_utils.py

@@ -567,8 +567,8 @@ def current_stream() -> torch.cuda.Stream:
     return _current_stream_tls.value
 
 
-# Global auxilary stream for running operations in background streams.
-# We have single global auxilary stream to avoid an explosion of streams
+# Global auxiliary stream for running operations in background streams.
+# We have single global auxiliary stream to avoid an explosion of streams
 # for every layer (and make profiling look sane).
 #
 # aux_stream() is currently used for: