[Doc]: fixing typos to improve docs (#24480)

Signed-off-by: Didier Durand <durand.didier@gmail.com>

tests/engine/test_executor.py

@@ -25,7 +25,7 @@ class CustomUniExecutor(UniProcExecutor):
                        timeout: Optional[float] = None,
                        args: tuple = (),
                        kwargs: Optional[dict] = None) -> list[Any]:
-        # Drop marker to show that this was ran
+        # Drop marker to show that this was run
         with open(".marker", "w"):
             ...
         return super().collective_rpc(method, timeout, args, kwargs)

tests/entrypoints/offline_mode/test_offline_mode.py

@@ -79,7 +79,7 @@ def test_offline_mode(monkeypatch: pytest.MonkeyPatch):
             )
 
             # Need to re-import huggingface_hub
-            # and friends to setup offline mode
+            # and friends to set up offline mode
             _re_import_modules()
             # Cached model files should be used in offline mode
             for model_config in MODEL_CONFIGS:
@@ -136,7 +136,7 @@ def test_model_from_huggingface_offline(monkeypatch: pytest.MonkeyPatch):
                 disable_connect,
             )
             # Need to re-import huggingface_hub
-            # and friends to setup offline mode
+            # and friends to set up offline mode
             _re_import_modules()
             engine_args = EngineArgs(model="facebook/opt-125m")
             LLM(**dataclasses.asdict(engine_args))

tests/kernels/utils.py

@@ -1247,7 +1247,7 @@ def baseline_scaled_mm(a: torch.Tensor,
     # then we would expand a to:
     #       a = [[1, 1, 2, 2],
     #            [3, 3, 4, 4]]
-    # NOTE this function this function does not explicitly broadcast dimensions
+    # NOTE this function does not explicitly broadcast dimensions
     # with an extent of 1, since this can be done implicitly by pytorch
     def group_broadcast(t, shape):
         for i, s in enumerate(shape):

tests/models/language/generation/test_hybrid.py

@@ -301,7 +301,7 @@ def test_fail_upon_inc_requests_and_finished_requests_lt_available_blocks(
     finished_requests_ids is larger than the maximum mamba block capacity.
 
     This could generally happen due to the fact that hybrid does support
-    statelessness mechanism where it can cleanup new incoming requests in
+    statelessness mechanism where it can clean up new incoming requests in
     a single step.
     """
     try:
@@ -322,7 +322,7 @@ def test_state_cleanup(
     This test is for verifying that the Hybrid state is cleaned up between
     steps.
     
-    If its not cleaned, an error would be expected.
+    If it's not cleaned, an error would be expected.
     """
     try:
         with vllm_runner(model, max_num_seqs=MAX_NUM_SEQS) as vllm_model:

tests/tpu/test_quantization_accuracy.py

@@ -28,7 +28,7 @@ ACCURACY_CONFIGS = [
         expected_value=0.76),  # no bias
     # NOTE(rob): We cannot re-initialize vLLM in the same process for TPU,
     # so only one of these tests can run in a single call to pytest. As
-    # a follow up, move this into the LM-EVAL section of the CI.
+    # a follow-up, move this into the LM-EVAL section of the CI.
     # GSM8KAccuracyTestConfig(
     #     model_name="neuralmagic/Qwen2-7B-Instruct-quantized.w8a8",
     #     expected_value=0.66),  # bias in QKV layers