apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 17:11:21 +08:00
Code Issues Actions 23 Packages Projects Releases Wiki Activity

[Models] Add Qwen3-VL Model Support (#5763)

Browse Source 
* support v1 loader

* remove useless code

* remove useless

* [Model] support Qwen3VL images success

* [Model] support Qwen3VL rope_3d

* [Model] support Qwen3VL remove log

* [Model] support Qwen3VL RL

* [Model] support Qwen3VL tp

* [Model] support Qwen3VL video

* [Model] support Qwen3VL fix ernievl

* [Model] support Qwen3VL fix get_image_boundaries.cc array out of bounds

* [Model] support Qwen3VL fix multi card

* [Model] support Qwen3VL file close

* [Model] support Qwen3VL fix ce

* [Model] support Qwen3VL fix unittest

* [Model] support Qwen3VL add unittest

---------

Co-authored-by: Ayakouji <yuhongh@qq.com>
This commit is contained in:
CSWYF3634076
2025-12-29 17:39:33 +08:00
committed by GitHub
parent a3f0696e35
commit 9286403570
36 changed files with 5917 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/input/qwen_vl_processor/process_video.py
+23 -5
View File
@@ -14,19 +14,22 @@
# limitations under the License.
"""
import math
from typing import Optional, Union
import numpy as np
from fastdeploy.utils import data_processor_logger
from .image_processor import ceil_by_factor, floor_by_factor
def sample_frames(
frame_factor: int,
min_frames: int,
max_frames: int,
metadata: Optional[dict] = None,
fps: Optional[Union[int, float]] = None,
num_frames: Optional[int] = None,
fps: Optional[Union[int, float]] = -1,
num_frames: Optional[int] = -1,
):
"""
Sample frames from video according to specified criteria.
@@ -61,16 +64,31 @@ def sample_frames(
"Asked to sample `fps` frames per second but no video metadata was provided which is required when sampling with `fps`. "
"Please pass in `VideoMetadata` object or use a fixed `num_frames` per input video"
)
max_frames = math.floor(min(max_frames, total_num_frames) / frame_factor) * frame_factor
# max_frames = math.floor(min(max_frames, total_num_frames) / frame_factor) * frame_factor
min_frames = ceil_by_factor(min_frames, frame_factor)
max_frames = floor_by_factor(min(max_frames, total_num_frames), frame_factor)
num_frames = total_num_frames / metadata["fps"] * fps
if num_frames > total_num_frames:
data_processor_logger.warning(f"smart_nframes: nframes[{num_frames}] > total_frames[{total_num_frames}]")
num_frames = min(min(max(num_frames, min_frames), max_frames), total_num_frames)
num_frames = math.floor(num_frames / frame_factor) * frame_factor
num_frames = floor_by_factor(num_frames, frame_factor)
if num_frames > total_num_frames:
raise ValueError(
f"Video can't be sampled. The inferred `num_frames={num_frames}` exceeds `total_num_frames={total_num_frames}`. "
"Decrease `num_frames` or `fps` for sampling."
)
# Hack code ensures that num_frames can always be divided by 4
# due to sched/resource_manager_v1.py 中 grid_thw.extend([[2, h, w]] * (t // 2))
if num_frames > 2 and num_frames % 4 != 0:
num_frames = (num_frames // 4) * 4 # 向下取整到 4 的倍数
total_num_frames = (total_num_frames // 4) * 4
num_frames = min(min(max(num_frames, min_frames), max_frames), total_num_frames)
# Calculate frame indices based on sampling strategy
if num_frames > 0:
# Evenly spaced sampling for target frame count