[Doc]: fix typos in Python comments (#24115)

Signed-off-by: Didier Durand <[email protected]>

Author: didier-durand

.buildkite/nightly-benchmarks/scripts/compare-json-results.py

@@ -218,7 +218,7 @@ def split_json_by_tp_pp(
         "--xaxis",
         type=str,
         default="# of max concurrency.",
-        help="column name to use as X Axis in comparision graph",
+        help="column name to use as X Axis in comparison graph",
     )
     args = parser.parse_args()
 

benchmarks/benchmark_serving.py

@@ -1104,7 +1104,7 @@ def create_argument_parser():
         "--percentile-metrics",
         type=str,
         default="ttft,tpot,itl",
-        help="Comma-separated list of selected metrics to report percentils. "
+        help="Comma-separated list of selected metrics to report percentiles. "
         "This argument specifies the metrics to report percentiles. "
         'Allowed metric names are "ttft", "tpot", "itl", "e2el". '
         'Default value is "ttft,tpot,itl".',

benchmarks/benchmark_serving_structured_output.py

@@ -998,7 +998,7 @@ def create_argument_parser():
         "--percentile-metrics",
         type=str,
         default="ttft,tpot,itl",
-        help="Comma-separated list of selected metrics to report percentils. "
+        help="Comma-separated list of selected metrics to report percentiles. "
         "This argument specifies the metrics to report percentiles. "
         'Allowed metric names are "ttft", "tpot", "itl", "e2el". '
         'Default value is "ttft,tpot,itl".',

benchmarks/benchmark_throughput.py

@@ -719,7 +719,7 @@ def create_argument_parser():
         "[length * (1 - range_ratio), length * (1 + range_ratio)].",
     )
 
-    # hf dtaset
+    # hf dataset
     parser.add_argument(
         "--hf-subset", type=str, default=None, help="Subset of the HF dataset."
     )

tools/profiler/visualize_layerwise_profile.py

@@ -119,7 +119,7 @@ def all_the_same(items) -> bool:
              if not all_the_same(trace_eles)), None)
 
         if first_trace_difference is None:
-            # can't create a unique name, leave them names as the
+            # can't create a unique name, leave the names as they
             # are they will get aggregated by the pivot_table call
             continue
 

vllm/compilation/collective_fusion.py

@@ -513,7 +513,7 @@ def call_trtllm_fused_allreduce_norm(
                         torch.ops._C.static_scaled_fp8_quant(
                             quant_out, norm_out, scale_factor)
             if scale_factor is None or norm_out is not None:
-                # we need to return allreduce outpput
+                # we need to return allreduce output
                 # in cases of non quant fused AR + RMS norm
                 # and fused AR + RMS norm + quant without fused add
                 allreduce_in.copy_(allreduce_out)

vllm/engine/multiprocessing/engine.py

@@ -49,7 +49,7 @@ class MQLLMEngine:
 
     This class is used to wrap the
     [`LLMEngine`][vllm.engine.llm_engine.LLMEngine] class to enable use
-    in concurrnet manner. It runs a background loop and uses zeromq to
+    in concurrent manner. It runs a background loop and uses zeromq to
     receive new requests and stream outputs incrementally via ipc.
 
     The [`LLMEngine`][vllm.engine.llm_engine.LLMEngine] generate or encode

vllm/model_executor/layers/quantization/utils/w8a8_utils.py

@@ -23,7 +23,7 @@
 # The condition to determine if it is on a platform that supports
 # torch._scaled_mm rowwise feature.
 # The condition is determined once as the operations
-# are time consuming.
+# are time-consuming.
 USE_ROWWISE_TORCH_SCALED_MM = (current_platform.is_rocm() and version.parse(
     torch.__version__) >= version.parse("2.7")
                                and current_platform.has_device_capability(94))

vllm/model_executor/model_loader/default_loader.py

@@ -211,7 +211,7 @@ def _get_weights_iterator(
 
             if not USE_TPU_COMMONS:
                 # In PyTorch XLA, we should call `xm.mark_step`
-                # requently so that not too many ops are accumulated
+                # frequently so that not too many ops are accumulated
                 # in the XLA program. import torch_xla.core.xla_model
                 # as xm
                 import torch_xla.core.xla_model as xm

vllm/v1/worker/xpu_worker.py

@@ -84,7 +84,7 @@ def determine_available_memory(self) -> int:
         """Profiles the peak memory usage of the model to determine how many
         KV blocks may be allocated without OOMs.
         The engine will first conduct a profiling of the existing memory usage.
-        Then, it calculate the maximum possible number of GPU and CPU blocks
+        Then, it calculates the maximum possible number of GPU and CPU blocks
         that can be allocated with the remaining free memory.
         .. tip::
             You may limit the usage of GPU memory