diff --git a/docs/features/tool_calling.md b/docs/features/tool_calling.md index afc605a504b..54016038322 100644 --- a/docs/features/tool_calling.md +++ b/docs/features/tool_calling.md @@ -169,7 +169,7 @@ All Llama 3.1, 3.2 and 4 models should be supported. The tool calling that is supported is the [JSON-based tool calling](https://llama.meta.com/docs/model-cards-and-prompt-formats/llama3_1/#json-based-tool-calling). For [pythonic tool calling](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/text_prompt_format.md#zero-shot-function-calling) introduced by the Llama-3.2 models, see the `pythonic` tool parser below. As for Llama 4 models, it is recommended to use the `llama4_pythonic` tool parser. -Other tool calling formats like the built in python tool calling or custom tool calling are not supported. +Other tool calling formats like the built-in python tool calling or custom tool calling are not supported. Known issues: diff --git a/docs/getting_started/installation/gpu/rocm.inc.md b/docs/getting_started/installation/gpu/rocm.inc.md index 80e99d3034d..4c70128d0b4 100644 --- a/docs/getting_started/installation/gpu/rocm.inc.md +++ b/docs/getting_started/installation/gpu/rocm.inc.md @@ -119,7 +119,7 @@ Currently, there are no pre-built ROCm wheels. This may take 5-10 minutes. Currently, `pip install .` does not work for ROCm installation. !!! tip - - Triton flash attention is used by default. For benchmarking purposes, it is recommended to run a warm up step before collecting perf numbers. + - Triton flash attention is used by default. For benchmarking purposes, it is recommended to run a warm-up step before collecting perf numbers. - Triton flash attention does not currently support sliding window attention. If using half precision, please use CK flash-attention for sliding window support. - To use CK flash-attention or PyTorch naive attention, please use this flag `export VLLM_USE_TRITON_FLASH_ATTN=0` to turn off triton flash attention. - The ROCm version of PyTorch, ideally, should match the ROCm driver version. diff --git a/examples/tool_chat_template_phi4_mini.jinja b/examples/tool_chat_template_phi4_mini.jinja index 83886762c28..6f40c38c206 100644 --- a/examples/tool_chat_template_phi4_mini.jinja +++ b/examples/tool_chat_template_phi4_mini.jinja @@ -9,7 +9,7 @@ <|system|> {{ system_message }} {%- if tools %} -In addition to plain text responses, you can chose to call one or more of the provided functions. +In addition to plain text responses, you can choose to call one or more of the provided functions. Use the following rule to decide when to call a function: * if the response can be generated from your internal knowledge (e.g., as in the case of queries like "What is the capital of Poland?"), do so @@ -19,7 +19,7 @@ If you decide to call functions: * prefix function calls with functools marker (no closing marker required) * all function calls should be generated in a single JSON list formatted as functools[{"name": [function name], "arguments": [function arguments as JSON]}, ...] * follow the provided JSON schema. Do not hallucinate arguments or values. Do to blindly copy values from the provided samples - * respect the argument type formatting. E.g., if the type if number and format is float, write value 7 as 7.0 + * respect the argument type formatting. E.g., if the type is number and format is float, write value 7 as 7.0 * make sure you pick the right functions that match the user intent diff --git a/tests/engine/test_executor.py b/tests/engine/test_executor.py index 15c7a97b50e..67064aff3ae 100644 --- a/tests/engine/test_executor.py +++ b/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) diff --git a/tests/entrypoints/offline_mode/test_offline_mode.py b/tests/entrypoints/offline_mode/test_offline_mode.py index a154bb1059a..f8ed5dda260 100644 --- a/tests/entrypoints/offline_mode/test_offline_mode.py +++ b/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)) diff --git a/tests/kernels/utils.py b/tests/kernels/utils.py index c46db8e3079..c9bf85f6e2a 100644 --- a/tests/kernels/utils.py +++ b/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): diff --git a/tests/models/language/generation/test_hybrid.py b/tests/models/language/generation/test_hybrid.py index b44ddc61b6c..d0e42062099 100644 --- a/tests/models/language/generation/test_hybrid.py +++ b/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: diff --git a/tests/tpu/test_quantization_accuracy.py b/tests/tpu/test_quantization_accuracy.py index 6cefbae4bdd..8d9fbd28031 100644 --- a/tests/tpu/test_quantization_accuracy.py +++ b/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 diff --git a/vllm/distributed/parallel_state.py b/vllm/distributed/parallel_state.py index 522dfc8d8b5..602bcebc017 100644 --- a/vllm/distributed/parallel_state.py +++ b/vllm/distributed/parallel_state.py @@ -1117,7 +1117,7 @@ def initialize_model_parallel( "decode context model parallel group is already initialized") # Note(hc): In the current implementation of decode context parallel, # dcp_size must not exceed tp_size, because the world size does not - # change by DCP, it simply reuse the GPUs of TP group, and split one + # change by DCP, it simply reuses the GPUs of TP group, and split one # TP group into tp_size//dcp_size DCP groups. group_ranks = all_ranks.reshape( -1, decode_context_model_parallel_size).unbind(0)