[Deprecation][2/N] Replace --task with --runner and --convert (#21470)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2025-07-28 10:42:40 +08:00
parent 8f605ee309
commit 86ae693f20
94 changed files with 1117 additions and 1083 deletions
--- a/tests/models/multimodal/generation/test_common.py
+++ b/tests/models/multimodal/generation/test_common.py
@@ -222,7 +222,6 @@ VLM_TEST_SETTINGS = {
        },
        marks=[large_gpu_mark(min_gb=32)],
    ),
-    # Check "auto" with fallback to transformers
    "internvl-transformers": VLMTestInfo(
        models=["OpenGVLab/InternVL3-1B-hf"],
        test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
@@ -232,7 +231,7 @@ VLM_TEST_SETTINGS = {
        use_tokenizer_eos=True,
        image_size_factors=[(0.25, 0.5, 1.0)],
        vllm_runner_kwargs={
-            "model_impl": "auto",
+            "model_impl": "transformers",
        },
        auto_cls=AutoModelForImageTextToText,
        marks=[pytest.mark.core_model],
@@ -638,7 +637,7 @@ VLM_TEST_SETTINGS = {
        img_idx_to_prompt=lambda idx: f"<|image_{idx}|>\n",
        max_model_len=4096,
        max_num_seqs=2,
-        task="generate",
+        runner="generate",
        # use sdpa mode for hf runner since phi3v didn't work with flash_attn
        hf_model_kwargs={"_attn_implementation": "sdpa"},
        use_tokenizer_eos=True,
--- a/tests/models/multimodal/generation/test_granite_speech.py
+++ b/tests/models/multimodal/generation/test_granite_speech.py
@@ -65,7 +65,7 @@ def run_test(
    # max_model_len should be greater than image_feature_size
    with vllm_runner(
            model,
-            task="generate",
+            runner="generate",
            max_model_len=max_model_len,
            max_num_seqs=1,
            dtype=dtype,
--- a/tests/models/multimodal/generation/test_interleaved.py
+++ b/tests/models/multimodal/generation/test_interleaved.py
@@ -48,7 +48,7 @@ def test_models(vllm_runner, model, dtype: str, max_tokens: int) -> None:
    ]

    with vllm_runner(model,
-                     task="generate",
+                     runner="generate",
                     dtype=dtype,
                     limit_mm_per_prompt={"image": 2},
                     max_model_len=32768,
--- a/tests/models/multimodal/generation/test_phi4mm.py
+++ b/tests/models/multimodal/generation/test_phi4mm.py
@@ -99,7 +99,7 @@ def run_test(
    # max_model_len should be greater than image_feature_size
    with vllm_runner(
            model,
-            task="generate",
+            runner="generate",
            max_model_len=max_model_len,
            max_num_seqs=2,
            dtype=dtype,
--- a/tests/models/multimodal/generation/test_qwen2_vl.py
+++ b/tests/models/multimodal/generation/test_qwen2_vl.py
@@ -267,7 +267,7 @@ def run_embedding_input_test(

    # max_model_len should be greater than image_feature_size
    with vllm_runner(model,
-                     task="generate",
+                     runner="generate",
                     max_model_len=4000,
                     max_num_seqs=3,
                     dtype=dtype,
--- a/tests/models/multimodal/generation/vlm_utils/core.py
+++ b/tests/models/multimodal/generation/vlm_utils/core.py
@@ -6,7 +6,7 @@ from typing import Any, Callable, Optional
 import torch
 from transformers.models.auto.auto_factory import _BaseAutoModelClass

-from vllm.config import TaskOption
+from vllm.config import RunnerOption
 from vllm.transformers_utils.tokenizer import AnyTokenizer

 from .....conftest import HfRunner, VllmRunner
@@ -37,7 +37,7 @@ def run_test(
    vllm_runner_kwargs: Optional[dict[str, Any]],
    hf_model_kwargs: Optional[dict[str, Any]],
    patch_hf_runner: Optional[Callable[[HfRunner], HfRunner]],
-    task: TaskOption = "auto",
+    runner: RunnerOption = "auto",
    distributed_executor_backend: Optional[str] = None,
    tensor_parallel_size: int = 1,
    vllm_embeddings: Optional[torch.Tensor] = None,
@@ -83,7 +83,7 @@ def run_test(
                     tensor_parallel_size=tensor_parallel_size,
                     distributed_executor_backend=distributed_executor_backend,
                     enforce_eager=enforce_eager,
-                     task=task,
+                     runner=runner,
                     **vllm_runner_kwargs_) as vllm_model:
        tokenizer = vllm_model.llm.get_tokenizer()

--- a/tests/models/multimodal/generation/vlm_utils/types.py
+++ b/tests/models/multimodal/generation/vlm_utils/types.py
@@ -11,7 +11,7 @@ from pytest import MarkDecorator
 from transformers import AutoModelForCausalLM
 from transformers.models.auto.auto_factory import _BaseAutoModelClass

-from vllm.config import TaskOption
+from vllm.config import RunnerOption
 from vllm.sequence import SampleLogprobs
 from vllm.transformers_utils.tokenizer import AnyTokenizer

@@ -109,7 +109,7 @@ class VLMTestInfo(NamedTuple):
    enforce_eager: bool = True
    max_model_len: int = 1024
    max_num_seqs: int = 256
-    task: TaskOption = "auto"
+    runner: RunnerOption = "auto"
    tensor_parallel_size: int = 1
    vllm_runner_kwargs: Optional[dict[str, Any]] = None

@@ -173,7 +173,7 @@ class VLMTestInfo(NamedTuple):
            "enforce_eager": self.enforce_eager,
            "max_model_len": self.max_model_len,
            "max_num_seqs": self.max_num_seqs,
-            "task": self.task,
+            "runner": self.runner,
            "tensor_parallel_size": self.tensor_parallel_size,
            "vllm_runner_kwargs": self.vllm_runner_kwargs,
            "hf_output_post_proc": self.hf_output_post_proc,
--- a/tests/models/multimodal/pooling/test_dse_qwen2_vl.py
+++ b/tests/models/multimodal/pooling/test_dse_qwen2_vl.py
@@ -92,7 +92,7 @@ def _run_test(
    # if we run HF first, the cuda initialization will be done and it
    # will hurt multiprocessing backend with fork method (the default method).
    with vllm_runner(model,
-                     task="embed",
+                     runner="pooling",
                     dtype=dtype,
                     enforce_eager=True,
                     max_model_len=8192) as vllm_model:
--- a/tests/models/multimodal/pooling/test_jinavl_reranker.py
+++ b/tests/models/multimodal/pooling/test_jinavl_reranker.py
@@ -49,7 +49,7 @@ def vllm_reranker(

    with vllm_runner(
            model_name,
-            task="score",
+            runner="pooling",
            dtype=dtype,
            max_num_seqs=2,
            max_model_len=2048,
--- a/tests/models/multimodal/pooling/test_llava_next.py
+++ b/tests/models/multimodal/pooling/test_llava_next.py
@@ -64,7 +64,7 @@ def _run_test(
    # if we run HF first, the cuda initialization will be done and it
    # will hurt multiprocessing backend with fork method (the default method).
    with vllm_runner(model,
-                     task="embed",
+                     runner="pooling",
                     dtype=dtype,
                     max_model_len=4096,
                     enforce_eager=True) as vllm_model:
--- a/tests/models/multimodal/pooling/test_phi3v.py
+++ b/tests/models/multimodal/pooling/test_phi3v.py
@@ -44,7 +44,7 @@ def _run_test(
    # vLLM needs a fresh new process without cuda initialization.
    # if we run HF first, the cuda initialization will be done and it
    # will hurt multiprocessing backend with fork method (the default method).
-    with vllm_runner(model, task="embed", dtype=dtype,
+    with vllm_runner(model, runner="pooling", dtype=dtype,
                     enforce_eager=True) as vllm_model:
        vllm_outputs = vllm_model.embed(input_texts, images=input_images)

--- a/tests/models/multimodal/pooling/test_prithvi_mae.py
+++ b/tests/models/multimodal/pooling/test_prithvi_mae.py
@@ -34,7 +34,7 @@ def _run_test(
            set_default_torch_num_threads(1),
            vllm_runner(
                model,
-                task="embed",
+                runner="pooling",
                dtype=torch.float16,
                enforce_eager=True,
                skip_tokenizer_init=True,
--- a/tests/models/multimodal/processing/test_common.py
+++ b/tests/models/multimodal/processing/test_common.py
@@ -58,13 +58,10 @@ def _test_processing_correctness(

    model_config = ModelConfig(
        model_id,
-        task="auto",
        tokenizer=model_info.tokenizer or model_id,
        tokenizer_mode=model_info.tokenizer_mode,
-        trust_remote_code=model_info.trust_remote_code,
-        seed=0,
-        dtype="auto",
        revision=model_info.revision,
+        trust_remote_code=model_info.trust_remote_code,
        hf_overrides=model_info.hf_overrides,
    )

--- a/tests/models/multimodal/test_mapping.py
+++ b/tests/models/multimodal/test_mapping.py
@@ -54,13 +54,10 @@ def test_hf_model_weights_mapper(model_arch: str):

    model_config = ModelConfig(
        model_id,
-        task="auto",
        tokenizer=model_info.tokenizer or model_id,
        tokenizer_mode=model_info.tokenizer_mode,
+        revision=model_info.revision,
        trust_remote_code=model_info.trust_remote_code,
-        seed=0,
-        dtype="auto",
-        revision=None,
        hf_overrides=model_info.hf_overrides,
    )
    model_cls = MULTIMODAL_REGISTRY._get_model_cls(model_config)