[Mypy Fix] Mypy fix for "vllm/model_executor/models/[aA][bB]" (#48977)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2026-08-06 19:30:15 +00:00
committed by GitHub
parent d8eabdbfbe
commit adc3e03517
23 changed files with 214 additions and 167 deletions
+10 -7
View File
@@ -17,7 +17,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.
from collections.abc import Iterable, Mapping, Sequence
from collections.abc import Iterable, Mapping, Sequence, Sized
from typing import Annotated, Any, Literal, TypeAlias
import torch
@@ -30,7 +30,7 @@ from transformers.models.audioflamingo3 import (
from transformers.models.qwen2_audio import Qwen2AudioEncoder
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.config.multimodal import AudioDummyOptions, BaseDummyOptions
from vllm.inputs import ModalityData, MultiModalDataDict
from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.models.module_mapping import MultiModelKeys
@@ -226,6 +226,7 @@ class AudioFlamingo3DummyInputsBuilder(
audio_len = int(hf_processor.max_audio_len * sampling_rate)
num_audios = mm_counts.get("audio", 0)
audio_overrides = mm_options.get("audio")
assert audio_overrides is None or isinstance(audio_overrides, AudioDummyOptions)
return {
"audio": self._get_dummy_audios(
@@ -317,7 +318,7 @@ def _count_audio_tokens_from_mask(
if isinstance(chunk_counts, torch.Tensor):
counts = chunk_counts.tolist()
elif chunk_counts and isinstance(chunk_counts[0], torch.Tensor):
counts = [count.item() for count in chunk_counts]
counts = [int(count) for count in chunk_counts]
else:
counts = chunk_counts
@@ -375,7 +376,7 @@ class AudioFlamingo3MultiModalProcessor(
def _call_hf_processor(
self,
prompt: str,
mm_data: dict[str, object],
mm_data: Mapping[str, object],
mm_kwargs: Mapping[str, Any],
tok_kwargs: Mapping[str, object],
) -> BatchFeature:
@@ -412,7 +413,8 @@ class AudioFlamingo3MultiModalProcessor(
chunk_counts = []
for audio in audio_list:
# audio is numpy array or list
n_samples = len(audio) if isinstance(audio, list) else audio.shape[0]
assert isinstance(audio, Sized)
n_samples = len(audio)
n_win = max(1, (n_samples + window_size - 1) // window_size)
if n_win > max_windows:
@@ -457,6 +459,7 @@ class AudioFlamingo3MultiModalProcessor(
)
else:
audio_embeds = out_mm_data["audio_embeds"][item_idx]
assert isinstance(audio_embeds, torch.Tensor)
num_features = audio_embeds.shape[0]
if num_features == 0:
@@ -556,8 +559,8 @@ class AudioFlamingo3ForConditionalGeneration(
def _process_audio_input(
self, audio_input: AudioFlamingo3Inputs
) -> torch.Tensor | tuple[torch.Tensor, ...]:
if audio_input["type"] == "audio_embeds":
audio_embeds = audio_input["audio_embeds"]
if audio_input.type == "audio_embeds":
audio_embeds = audio_input.audio_embeds
return tuple(audio_embeds)
(