[Bugfix][CI] fix typos (#34934)

Signed-off-by: 1195343015 <1195343015@qq.com>
Signed-off-by: Jiayi Yan <66017932+1195343015@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/compile/test_decorator.py

@@ -234,7 +234,7 @@ def test_conditional_compile_enable_if(use_inductor_graph_partition, monkeypatch
         expected_num_backend_compilations = 4
 
     # A has support_torch_compile but enable_if fn returns False
-    # enalbe_if will be True for B, so we expect mod1 and mod2
+    # enable_if will be True for B, so we expect mod1 and mod2
     # to be compiled
     with compilation_counter.expect(
         num_graphs_seen=2,

tests/compile/test_wrapper.py

@@ -95,7 +95,7 @@ def test_torch_compile_wrapper(use_bytecode_hook, monkeypatch):
             f"Expected {expected1}, got {result1}"
         )
 
-        # Second call should triger another compilation
+        # Second call should trigger another compilation
         x2 = torch.tensor([1, 2, 3])
         result2 = wrapper(x2)
         expected2 = torch.tensor([100, 200, 300])

tests/kernels/attention/test_attention.py

@@ -444,7 +444,7 @@ def ref_multi_query_kv_attention(
 
 
 @pytest.mark.parametrize("attention_cls", [Attention, MMEncoderAttention])
-def test_num_heads_not_divisble_by_num_kv_heads(attention_cls: type) -> None:
+def test_num_heads_not_divisible_by_num_kv_heads(attention_cls: type) -> None:
     head_size = 64
     scale = float(1.0 / (head_size**0.5))
     num_heads = 16

tests/kernels/moe/test_modular_kernel_combinations.py

@@ -162,7 +162,7 @@ Ns = [1024]
 TOPKs = [4, 1]
 Es = [32]
 DTYPEs = [torch.bfloat16]
-FUSED_MOE_CHUNK_SIZEs = [None, 16]
+FUSED_MOE_CHUNK_SIZES = [None, 16]
 
 
 def is_nyi_config(config: Config) -> bool:
@@ -192,7 +192,7 @@ def generate_valid_test_cases(
         DTYPEs,
         MK_QUANT_CONFIGS,
         product(prepare_finalize_types, MK_FUSED_EXPERT_TYPES),
-        FUSED_MOE_CHUNK_SIZEs,
+        FUSED_MOE_CHUNK_SIZES,
     ):
         total = total + 1
 
@@ -266,7 +266,7 @@ def test_modular_kernel_combinations_multigpu(
     if cuda_device_count_stateless() < world_size:
         pytest.skip(
             f"Not enough GPUs available to run, got "
-            f"{cuda_device_count_stateless()} exepected "
+            f"{cuda_device_count_stateless()} expected "
             f"{world_size}."
         )
 

tests/models/language/generation/test_mistral.py

@@ -87,7 +87,7 @@ MSGS = [
     {
         "role": "user",
         "content": "Could you please rewrite the below article? \n\n My English needs "
-        "improvving, maybe I make errors.",
+        "improving, maybe I make errors.",
     },
     {
         "role": "assistant",
@@ -98,7 +98,7 @@ MSGS = [
                 "type": "function",
                 "function": {
                     "name": "rewrite",
-                    "arguments": '{"text":"My English needs improvving, maybe '
+                    "arguments": '{"text":"My English needs improving, maybe '
                     'I make errors."}',
                 },
             }

tests/models/language/pooling/test_bge_m3.py

@@ -14,7 +14,7 @@ MAX_MODEL_LEN = 512
 
 
 # Example from https://huggingface.co/BAAI/bge-m3
-sentences_1 = ["What is BGE M3?", "Defination of BM25"]
+sentences_1 = ["What is BGE M3?", "Definition of BM25"]
 sentences_2 = [
     "BGE M3 is an embedding model supporting dense retrieval, "
     "lexical matching and multi-vector interaction.",

tests/models/multimodal/generation/vlm_utils/model_utils.py

@@ -719,7 +719,7 @@ def isaac_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
         # Convert to tuple or None
         all_hidden_states = tuple(hidden_states_list) if output_hidden_states else None
 
-        # Include hiden_states for compatibility with hidden_states_to_seq_logprobs()
+        # Include hidden_states for compatibility with hidden_states_to_seq_logprobs()
         return BaseModelOutputWithPast(
             last_hidden_state=hidden_states,
             past_key_values=past_key_values,
@@ -1226,7 +1226,7 @@ def voxtral_patch_hf_runner(hf_model: "HfRunner") -> "HfRunner":
        dicts (accepting ``url``, ``path``, or ``base64`` audio) rather than
        the standard ``processor(text=, audio=, sampling_rate=)`` interface.
     2. HfRunner.get_inputs cannot handle multi-audio per prompt because it
-       mis-unpacks ``[(arr1, sr1), (arr2, sr2)]`` via a ``len == 2`` check.
+       incorrectly unpacks ``[(arr1, sr1), (arr2, sr2)]`` via a ``len == 2`` check.
 
     We override ``get_inputs`` to build conversation dicts and call
     ``apply_chat_template`` directly, bypassing both issues. We also wrap

tests/quantization/test_blackwell_moe.py

@@ -25,7 +25,7 @@ def set_test_environment():
     os.environ["FLASHINFER_NVCC_THREADS"] = "16"
 
 
-# Overide the backbone layers to 4 for faster startup
+# Override the backbone layers to 4 for faster startup
 HF_OVERRIDE_TEXT = {
     "num_layers": 4,
     "num_hidden_layers": 4,

tests/renderers/test_hf.py

@@ -206,8 +206,8 @@ def test_resolve_chat_template_kwargs(sample_json_schema, model, expected_kwargs
 
     chat_template_kwargs = {
         # both unused
-        "unsed_kwargs_1": 123,
-        "unsed_kwargs_2": "abc",
+        "unused_kwargs_1": 123,
+        "unused_kwargs_2": "abc",
         # should not appear
         "chat_template": "{% Hello world! %}",
         "tokenize": True,

tests/test_config.py

@@ -853,7 +853,7 @@ def test_vllm_config_defaults_are_none():
 
 
 @pytest.mark.parametrize(
-    ("model_id", "compiliation_config", "optimization_level"),
+    ("model_id", "compilation_config", "optimization_level"),
     [
         (
             None,
@@ -895,7 +895,7 @@ def test_vllm_config_defaults_are_none():
         ("RedHatAI/DeepSeek-V2.5-1210-FP8", CompilationConfig(), OptimizationLevel.O3),
     ],
 )
-def test_vllm_config_defaults(model_id, compiliation_config, optimization_level):
+def test_vllm_config_defaults(model_id, compilation_config, optimization_level):
     """Test that optimization-level defaults are correctly applied."""
 
     model_config = None
@@ -903,12 +903,12 @@ def test_vllm_config_defaults(model_id, compiliation_config, optimization_level)
         model_config = ModelConfig(model_id)
         vllm_config = VllmConfig(
             model_config=model_config,
-            compilation_config=compiliation_config,
+            compilation_config=compilation_config,
             optimization_level=optimization_level,
         )
     else:
         vllm_config = VllmConfig(
-            compilation_config=compiliation_config,
+            compilation_config=compilation_config,
             optimization_level=optimization_level,
         )
     # Use the global optimization level defaults

tests/tool_parsers/test_seed_oss_tool_parser.py

@@ -106,7 +106,7 @@ def test_extract_tool_calls_no_tools(seed_oss_tool_parser):
 @pytest.mark.parametrize(
     ids=[
         "tool_call_0_thinking_budget",
-        "tool_call_512_thinkg_budget",
+        "tool_call_512_thinking_budget",
         "tool_call_unlimited_thinking_budget",
     ],
     argnames=["model_output", "expected_tool_calls", "expected_content"],
@@ -308,7 +308,7 @@ def stream_delta_message_generator(
 @pytest.mark.parametrize(
     ids=[
         "tool_call_0_thinking_budget",
-        "tool_call_512_thinkg_budget",
+        "tool_call_512_thinking_budget",
         "tool_call_unlimited_thinking_budget",
     ],
     argnames=["model_output", "expected_tool_calls", "expected_content"],

tests/transformers_utils/test_repo_utils.py

@@ -34,10 +34,10 @@ def test_list_filtered_repo_files(
         subfolder.mkdir()
         (path_tmp_dir / "json_file.json").touch()
         (path_tmp_dir / "correct_2.txt").touch()
-        (path_tmp_dir / "uncorrect.txt").touch()
-        (path_tmp_dir / "uncorrect.jpeg").touch()
+        (path_tmp_dir / "incorrect.txt").touch()
+        (path_tmp_dir / "incorrect.jpeg").touch()
         (subfolder / "correct.txt").touch()
-        (subfolder / "uncorrect_sub.txt").touch()
+        (subfolder / "incorrect_sub.txt").touch()
 
         def _glob_path() -> list[str]:
             return [
@@ -86,7 +86,7 @@ def test_one_filtered_repo_files(allow_patterns: list[str], expected_bool: bool)
         path_tmp_dir = Path(tmp_dir)
         subfolder = path_tmp_dir / "subfolder"
         subfolder.mkdir()
-        (path_tmp_dir / "uncorrect.jpeg").touch()
+        (path_tmp_dir / "incorrect.jpeg").touch()
         (subfolder / "correct.txt").touch()
 
         def _glob_path() -> list[str]:

tests/v1/core/test_kv_cache_utils.py

@@ -308,7 +308,7 @@ def test_free_kv_cache_block_queue_append_n():
 
     # Create an empty FreeKVCacheBlockQueue
     invalid_queue = FreeKVCacheBlockQueue([])
-    # set prev_free_block to None and this will cause assertation in append_n
+    # set prev_free_block to None and this will cause assertion in append_n
     invalid_queue.fake_free_list_tail.prev_free_block = None
     with pytest.raises(AssertionError):
         # Append 1 block

tests/v1/core/test_prefix_caching.py

@@ -2304,22 +2304,22 @@ def test_block_lookup_cache_single_block_per_key():
     assert cache.get_one_block(key0) is block0
     assert cache.get_one_block(key1) is block1
     assert cache.get_one_block(key2) is None
-    # No block poped due to block_id mismatch
+    # No block popped due to block_id mismatch
     assert cache.pop(key0, 100) is None
     assert cache.get_one_block(key0) is block0
     assert cache.get_one_block(key1) is block1
     assert cache.get_one_block(key2) is None
-    # block poped with (key0, block ID 0)
+    # block popped with (key0, block ID 0)
     assert cache.pop(key0, 0) is block0
     assert cache.get_one_block(key0) is None
     assert cache.get_one_block(key1) is block1
     assert cache.get_one_block(key2) is None
-    # No block poped due to block_id mismatch
+    # No block popped due to block_id mismatch
     assert cache.pop(key0, 1) is None
     assert cache.get_one_block(key0) is None
     assert cache.get_one_block(key1) is block1
     assert cache.get_one_block(key2) is None
-    # block poped with (key1, block ID 1)
+    # block popped with (key1, block ID 1)
     assert cache.pop(key1, 1) is block1
     assert cache.get_one_block(key0) is None
     assert cache.get_one_block(key1) is None

tests/v1/core/test_priority_scheduler_random.py

@@ -140,7 +140,7 @@ def _mock_draft_token_ids(
     return DraftTokenIds(req_ids=request_ids, draft_token_ids=sampled_token_ids)
 
 
-def _chech_valid_scheduler_output(
+def _check_valid_scheduler_output(
     scheduler_output: SchedulerOutput,
     seen_request_ids: set[str],
     seen_mm_hashes: set[str],
@@ -242,7 +242,7 @@ def test_priority_scheduling_blast(
                 )
                 scheduler.add_request(req)
         scheduler_output = scheduler.schedule()
-        _chech_valid_scheduler_output(
+        _check_valid_scheduler_output(
             scheduler_output, seen_request_ids, seen_mm_hashes
         )
         model_output = _mock_execute_model(

tests/v1/core/test_scheduler.py

@@ -1116,7 +1116,7 @@ def _step_until_done(
 
 
 def _step_until_kv_transfer_finished(scheduler: Scheduler, req_ids: list[str]):
-    """Cycle requests through a KV transfer cyle."""
+    """Cycle requests through a KV transfer cycle."""
 
     # Requests should first transition to WAITING_FOR_REMOTE_KVS
     output = scheduler.schedule()
@@ -2714,7 +2714,7 @@ def _assert_right_encoder_inputs(
         if expected_total_reqs == 0:
             return
 
-    # Number of expected enocder inputs should match number of requests
+    # Number of expected encoder inputs should match number of requests
     if expected_encoder_inputs:
         assert check_exist and requests is not None  # only support expect input exist
         assert len(requests) == len(expected_encoder_inputs)
@@ -2964,7 +2964,7 @@ def test_ec_connector_with_partial_cache_hit_multi_round(use_kv_connector):
     )
     scheduler.update_from_output(output, model_output)
 
-    # request1 is finished after outputing 1 token
+    # request1 is finished after outputting 1 token
     # Finish request
     scheduler.finish_requests(request1.request_id, RequestStatus.FINISHED_LENGTH_CAPPED)
 
@@ -3060,14 +3060,14 @@ def test_ec_connector_schedule_multiple_requests(cache_exist, use_kv_connector):
     for request in requests:
         scheduler.add_request(request)
 
-    # Set up to test different encoder cache exsistence scenario after preemption
+    # Set up to test different encoder cache existence scenario after preemption
     # Order of getting encoder cache should be: local cache -> connector-> compute
     scheduler.ec_connector.update_state_after_alloc = Mock(
         wraps=scheduler.ec_connector.update_state_after_alloc
     )
 
     if cache_exist == "local":
-        # Allocate cache to cache manager manually to mimick
+        # Allocate cache to cache manager manually to mimic
         for req in requests:
             scheduler.encoder_cache_manager.allocate(req, 0)
     else:
@@ -3384,13 +3384,13 @@ def test_priority_scheduling_ec_connector_preemption_and_resumption(
         pooler_output=[],
     )
     # Finish the requests to make room for the preempted requests to resume
-    # req_high is finished after outputing 2 tokens
+    # req_high is finished after outputting 2 tokens
     scheduler.update_from_output(output, model_output)
     scheduler.finish_requests(
         request_high.request_id, RequestStatus.FINISHED_LENGTH_CAPPED
     )
 
-    # Set up to test different encoder cache exsistence scenario after preemption
+    # Set up to test different encoder cache existence scenario after preemption
     # Order of getting encoder cache should be: local cache -> connector-> compute
     # By default, the cache should still exist in local in this test case
     if cache_exist != "local":
@@ -3483,7 +3483,7 @@ def test_ec_connector_allocate_encoder_tokens_with_external_load(use_kv_connecto
         ec_role="ec_consumer",
     )
 
-    # Limit the number of availiable slots of EncoderCacheManager
+    # Limit the number of available slots of EncoderCacheManager
     scheduler.encoder_cache_manager = EncoderCacheManager(cache_size=32)
 
     # Create MM request1
@@ -3574,7 +3574,7 @@ def test_ec_connector_allocate_encoder_tokens_with_external_load(use_kv_connecto
     )
     scheduler.update_from_output(output, model_output)
 
-    # request1 is finished after outputing 1 token
+    # request1 is finished after outputting 1 token
     # Finish request
     scheduler.finish_requests(request1.request_id, RequestStatus.FINISHED_LENGTH_CAPPED)
     assert scheduler.get_num_unfinished_requests() == 1

tests/v1/e2e/test_mamba_prefix_cache.py

@@ -76,11 +76,11 @@ def get_fake_sample_fn() -> SamplerOutput:
                 ),
                 logprobs_tensors=None,
             )
-        accpeted_tokens = prompt_token_ids[
+        accepted_tokens = prompt_token_ids[
             first_token_id_index : first_token_id_index
             + min(num_accepted_tokens, logits.shape[0])
         ]
-        sampled_token_ids = accpeted_tokens
+        sampled_token_ids = accepted_tokens
         return SamplerOutput(
             sampled_token_ids=torch.tensor(
                 [sampled_token_ids], device="cuda", dtype=torch.int32

tests/v1/entrypoints/llm/test_struct_output_generate.py

@@ -911,7 +911,7 @@ def test_structured_output_with_structural_tag(backend: str):
         ),
     )
 
-    prompt = "Hello and repete hello 10 times, do not say anything else. Only say hello hello hello, now start"
+    prompt = "Hello and repeat hello 10 times, do not say anything else. Only say hello hello hello, now start"
     outputs = llm.generate(prompt, sampling_params=sampling_params, use_tqdm=True)
     assert outputs is not None
     for output in outputs:

tests/v1/kv_connector/unit/test_moriio_connector.py

@@ -99,7 +99,7 @@ def _setup_kv_transfer_request(request, remote_host="127.0.0.1", fake_port=4789)
     return request
 
 
-class FakeMorIIOWrapper:
+class FakeMoRIIOWrapper:
     # A fake MoRIIOWrapper for testing purposes
     def __init__(self, *args, **kwargs):
         pass
@@ -168,7 +168,7 @@ class FakeMorIIOWrapper:
         pass
 
 
-class FakeMorIIOConnectorWorker(MoRIIOConnectorWorker):
+class FakeMoRIIOConnectorWorker(MoRIIOConnectorWorker):
     # Define a fake remote engine id for testing
     REMOTE_ENGINE_ID = "remote_engine"
 
@@ -373,7 +373,7 @@ def test_read_mode_loads_remote_block_ids(moriio_read_mode):
     # Set remote block ids to be fetched.
     request.kv_transfer_params["remote_block_ids"] = block_list
 
-    # Remote Prefill, triggers MorIIOConnectorMetadata.
+    # Remote Prefill, triggers MoRIIOConnectorMetadata.
 
     scheduler_output = scheduler.schedule()
     kv_connector_metadata = scheduler_output.kv_connector_metadata
@@ -451,7 +451,7 @@ def test_register_kv_caches(mock_parallel_groups):
 
         with set_current_vllm_config(vllm_config):
             connector = MoRIIOConnector(vllm_config, KVConnectorRole.WORKER)
-            connector.connector_worker = FakeMorIIOConnectorWorker(
+            connector.connector_worker = FakeMoRIIOConnectorWorker(
                 vllm_config, connector.engine_id, hand_shake_latency=0
             )
 
@@ -528,7 +528,7 @@ def test_moriio_handshake_returns_metadata(mock_parallel_groups):
     with (
         patch(
             "vllm.distributed.kv_transfer.kv_connector.v1.moriio.moriio_engine.MoRIIOWrapper",
-            FakeMorIIOWrapper,
+            FakeMoRIIOWrapper,
         ),
     ):
         handshake_port = _find_free_port()

tests/v1/kv_connector/unit/test_nixl_connector.py

@@ -460,9 +460,9 @@ class FakeNixlConnectorWorker(NixlConnectorWorker):
 
         # When remote tp_size > local tp_size, handshake with multiple
         # remote ranks.
-        num_hanshakes = 1 if tp_ratio > 0 else -tp_ratio
+        num_handshakes = 1 if tp_ratio > 0 else -tp_ratio
         remote_agents: dict[int, str] = {}
-        for remote_tp_rank in range(num_hanshakes):
+        for remote_tp_rank in range(num_handshakes):
             remote_agent_name = self.add_remote_agent(
                 NixlAgentMetadata(
                     engine_id=self.REMOTE_ENGINE_ID,
@@ -688,7 +688,7 @@ class TestNixlHandshake:
         )
         check_handshake(2)
 
-        # NOTE flexiblity: a second remote with higher number of ranks is
+        # NOTE flexibility: a second remote with higher number of ranks is
         # discovered. This is not a scenario we actively support right now, but
         # the connector allows it.
         worker.REMOTE_ENGINE_ID = "remote_engine_2"
@@ -1766,7 +1766,7 @@ def test_aborted_request_removed_from_worker_in_batch(default_vllm_config, dist_
     req = create_request(request_id=1, do_remote_decode=True, max_tokens=1)
     scheduler.add_request(req)
 
-    # First scheduling pass - examinate build_connector_meta output
+    # First scheduling pass - examine build_connector_meta output
     sched_out = scheduler.schedule()
     kv_meta = sched_out.kv_connector_metadata
     assert kv_meta is not None

tests/v1/sample/test_logprobs.py

@@ -36,7 +36,7 @@ SAMPLE_PROMPT = BatchLogprobsComposition.SAMPLE_PROMPT
 # non-associative and sensitive to batch geometry. The ref LLM (no spec
 # decode, default scheduling) and the spec-decode LLM (chunked prefill,
 # different effective batch sizes) follow different reduction orders,
-# producing numerically divergent logprobs that get mis-attributed to
+# producing numerically divergent logprobs that get misattributed to
 # spec-decode incorrectness.
 #
 # Force LLM instances into an identical, deterministic execution

tests/v1/sample/test_rejection_sampler.py

@@ -726,7 +726,7 @@ def test_frequency_penalties(rejection_sampler):
     spec_tokens = [[1, 1, 1], [], [1, 1, 1]]
     output_tokens = [[1, 1, 1, 1], [7], [1, 1, 1, 1]]  # 1, 7 and 1 are the bonus tokens
 
-    num_requsts = len(spec_tokens)
+    num_requests = len(spec_tokens)
     logits = create_logits_tensor(output_tokens, token_idx_to_override=15)
     metadata = create_sampling_metadata(
         all_greedy=True,
@@ -734,8 +734,8 @@ def test_frequency_penalties(rejection_sampler):
         spec_token_ids=spec_tokens,
         prompt_token_ids=torch.tensor([[5, 6, 7], [6, 7, 8], [7, 8, 9]], device=DEVICE),
         frequency_penalties=[1.5, 1.5, 0.7],
-        presence_penalties=[0.0] * num_requsts,
-        repetition_penalties=[1.0] * num_requsts,
+        presence_penalties=[0.0] * num_requests,
+        repetition_penalties=[1.0] * num_requests,
     )
     bonus_token_tensor = torch.tensor(
         [output_tokens[i][-1] for i in range(len(output_tokens))], device=logits.device