[Refactor] Allow optional MultiModalKwargsItem in IPC (#23022)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-16 11:30:49 +00:00
committed by GitHub
parent de9cb61763
commit 4dff91c93d
10 changed files with 59 additions and 108 deletions
+2 -10
View File
@@ -7,9 +7,7 @@ import pytest
import torch
from vllm.config import ModelConfig, SchedulerConfig, VllmConfig
from vllm.multimodal.inputs import (MultiModalBatchedField,
MultiModalFieldElem, MultiModalKwargsItem,
PlaceholderRange)
from vllm.multimodal.inputs import MultiModalKwargsItem, PlaceholderRange
from vllm.sampling_params import SamplingParams
from vllm.utils import GiB_bytes, sha256, sha256_cbor_64bit
from vllm.v1.core.kv_cache_manager import KVCacheManager
@@ -42,13 +40,7 @@ def make_request(
if mm_positions is None:
mm_kwargs = None
else:
mm_elem = MultiModalFieldElem(
modality="dummy_m",
key="dummy_k",
data=None,
field=MultiModalBatchedField(),
)
mm_item = MultiModalKwargsItem.from_elems([mm_elem])
mm_item = MultiModalKwargsItem.dummy("dummy_m")
mm_kwargs = [mm_item] * len(mm_positions)
return Request(request_id=request_id,
+2 -10
View File
@@ -9,9 +9,7 @@ import pytest
import torch
from vllm.distributed.kv_events import AllBlocksCleared, BlockRemoved
from vllm.multimodal.inputs import (MultiModalBatchedField,
MultiModalFieldElem, MultiModalKwargsItem,
PlaceholderRange)
from vllm.multimodal.inputs import MultiModalKwargsItem, PlaceholderRange
from vllm.sampling_params import SamplingParams
from vllm.utils import sha256, sha256_cbor_64bit
from vllm.v1.core.block_pool import BlockPool
@@ -37,13 +35,7 @@ def make_request(
if mm_positions is None:
mm_kwargs = None
else:
mm_elem = MultiModalFieldElem(
modality="dummy_m",
key="dummy_k",
data=None,
field=MultiModalBatchedField(),
)
mm_item = MultiModalKwargsItem.from_elems([mm_elem])
mm_item = MultiModalKwargsItem.dummy("dummy_m")
mm_kwargs = [mm_item] * len(mm_positions)
return Request(request_id=request_id,
+2 -10
View File
@@ -8,9 +8,7 @@ import torch
from vllm.config import (CacheConfig, KVTransferConfig, ModelConfig,
SchedulerConfig, SpeculativeConfig, VllmConfig)
from vllm.multimodal.inputs import (MultiModalBatchedField,
MultiModalFieldElem, MultiModalKwargsItem,
PlaceholderRange)
from vllm.multimodal.inputs import MultiModalKwargsItem, PlaceholderRange
from vllm.sampling_params import GuidedDecodingParams, SamplingParams
from vllm.v1.core.sched.output import CachedRequestData, SchedulerOutput
from vllm.v1.core.sched.scheduler import Scheduler
@@ -1328,13 +1326,7 @@ def create_requests_with_priority(
for i in range(num_requests):
if mm_positions is not None:
mm_position = mm_positions[i]
mm_elem = MultiModalFieldElem(
modality="dummy_m",
key="dummy_k",
data=None,
field=MultiModalBatchedField(),
)
mm_item = MultiModalKwargsItem.from_elems([mm_elem])
mm_item = MultiModalKwargsItem.dummy("dummy_m")
mm_kwargs = [mm_item] * len(mm_position)
else:
mm_position = None
+2 -10
View File
@@ -6,9 +6,7 @@ import torch
from vllm.config import (CacheConfig, KVTransferConfig, ModelConfig,
SchedulerConfig, SpeculativeConfig, VllmConfig)
from vllm.multimodal.inputs import (MultiModalBatchedField,
MultiModalFieldElem, MultiModalKwargsItem,
PlaceholderRange)
from vllm.multimodal.inputs import MultiModalKwargsItem, PlaceholderRange
from vllm.sampling_params import SamplingParams
from vllm.v1.core.kv_cache_utils import (get_request_block_hasher,
init_none_hash)
@@ -143,13 +141,7 @@ def create_requests(
for i in range(num_requests):
if mm_positions is not None:
mm_position = mm_positions[i]
mm_elem = MultiModalFieldElem(
modality="dummy_m",
key="dummy_k",
data=None,
field=MultiModalBatchedField(),
)
mm_item = MultiModalKwargsItem.from_elems([mm_elem])
mm_item = MultiModalKwargsItem.dummy("dummy_m")
mm_kwargs = [mm_item] * len(mm_position)
mm_hashes = ["hash"] * len(mm_position)
else: