xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-09 21:37:29 +08:00
Code Issues Packages Projects Releases Wiki Activity

cleanup

Browse Source 
Signed-off-by: Reagan <reaganjlee@gmail.com>
This commit is contained in:
Reagan 2025-12-16 16:50:56 -08:00
parent fd6c862ba1
commit 169be31b01
2 changed files with 1 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/benchmarks/multimodal_processor.py
View File

@ -160,16 +160,13 @@ def generate_random_multimodal_prompts(
mm_items = [] mm_items = []
for _ in range(num_images): for _ in range(num_images):
# Generate random RGB image
random_pixels = rng.integers( random_pixels = rng.integers(
0, 256, (image_height, image_width, 3), dtype=np.uint8 0, 256, (image_height, image_width, 3), dtype=np.uint8
) )
image = Image.fromarray(random_pixels) image = Image.fromarray(random_pixels)
# Process to OpenAI format
mm_item = process_image(image) mm_item = process_image(image)
mm_items.append(mm_item) mm_items.append(mm_item)
# Create chat format: text + images
content = [{"type": "text", "text": text_prompt}] content = [{"type": "text", "text": text_prompt}]
content.extend(mm_items) content.extend(mm_items)
prompts.append([{"role": "user", "content": content}]) prompts.append([{"role": "user", "content": content}])
@ -189,7 +186,6 @@ def benchmark_multimodal_processor(
engine_args = EngineArgs.from_cli_args(args) engine_args = EngineArgs.from_cli_args(args)
llm = LLM(**dataclasses.asdict(engine_args)) llm = LLM(**dataclasses.asdict(engine_args))
# Validate max_model_len
assert llm.llm_engine.model_config.max_model_len >= ( assert llm.llm_engine.model_config.max_model_len >= (
args.input_len + args.output_len args.input_len + args.output_len
), ( ), (
@ -197,7 +193,6 @@ def benchmark_multimodal_processor(
"the sum of input_len and output_len." "the sum of input_len and output_len."
) )
# Generate random multimodal prompts
seed = getattr(args, "seed", 0) seed = getattr(args, "seed", 0)
tokenizer = llm.get_tokenizer() tokenizer = llm.get_tokenizer()
prompts, expected_output_lens = generate_random_multimodal_prompts( prompts, expected_output_lens = generate_random_multimodal_prompts(
@ -211,11 +206,10 @@ def benchmark_multimodal_processor(
seed=seed, seed=seed,
) )
# Create sampling params
sampling_params = [ sampling_params = [
SamplingParams( SamplingParams(
n=1, n=1,
temperature=0.0, # Greedy sampling for deterministic speed benchmarks temperature=0.0,
max_tokens=output_len, max_tokens=output_len,
detokenize=True, detokenize=True,
) )
@ -228,9 +222,6 @@ def benchmark_multimodal_processor(
freeze_gc_heap() freeze_gc_heap()
# MM processor stats are automatically enabled via set_defaults
# No need to check or raise error
debug = getattr(args, "debug_mm_stats", False) debug = getattr(args, "debug_mm_stats", False)
print(f"Processing {len(prompts)} requests...") print(f"Processing {len(prompts)} requests...")
@ -312,13 +303,10 @@ def add_cli_args(parser: argparse.ArgumentParser) -> None:
"""Add CLI arguments for the multimodal processor benchmark.""" """Add CLI arguments for the multimodal processor benchmark."""
from vllm.engine.arg_utils import EngineArgs from vllm.engine.arg_utils import EngineArgs
# Add EngineArgs (no conflict since we removed dataset parser)
EngineArgs.add_cli_args(parser) EngineArgs.add_cli_args(parser)
# Automatically enable MM processor stats (required for this benchmark)
parser.set_defaults(enable_mm_processor_stats=True) parser.set_defaults(enable_mm_processor_stats=True)
# Random generation arguments (similar to latency.py)
parser.add_argument( parser.add_argument(
"--num-prompts", "--num-prompts",
type=int, type=int,

1
vllm/multimodal/processing.py
View File

@ -74,7 +74,6 @@ logger = init_logger(__name__)
_S = TypeVar("_S", str, list[int]) _S = TypeVar("_S", str, list[int])
# Context variable to store the current request_id during preprocessing
_request_id_context: contextvars.ContextVar[str | None] = contextvars.ContextVar( _request_id_context: contextvars.ContextVar[str | None] = contextvars.ContextVar(
"_request_id_context", default=None "_request_id_context", default=None
) )