karylab_agents/vllm
1
0
Fork 0
forked from Karylab-cklius/vllm
Code Pull Requests 2 Actions 1 Packages Activity

refactor(k2vl): remove unused video-chunk config params

Browse Source 
Remove temporal_merge_kernel_size, sample_fps, and timestamp_mode
from K2VLConfig as they are not used by vLLM.

These parameters are loaded from preprocessor_config.json via
media_processor and should not be duplicated in the model config.

Signed-off-by: wanglinian <wanglinian@stu.pku.edu.cn>
This commit is contained in:
wanglinian
2026-01-07 07:15:54 +00:00
parent a0a0fb6075
commit 875d970dd0
1 changed files with 5 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/transformers_utils/configs/k2vl.py
+5 -23
View File
@@ -4,8 +4,7 @@
K2-VL Model Configuration.
This configuration supports video-chunk as an internal modality type.
A video-chunk is the smallest independently processable unit of video,
typically 4 frames per chunk (temporal_merge_kernel_size=4).
A video-chunk is the smallest independently processable unit of video.
"""
from transformers import DeepseekV3Config
@@ -89,22 +88,15 @@ class K2VLConfig(PretrainedConfig):
"""K2-VL model configuration.
K2-VL extends Kimi-VL with video support using video-chunks.
A video-chunk consists of multiple consecutive frames (default: 4)
A video-chunk consists of multiple consecutive frames
that are processed together with temporal pooling.
Args:
vision_config: Configuration for the vision tower and projector.
text_config: Configuration for the text model (DeepseekV3).
Video-Chunk Parameters:
temporal_merge_kernel_size: Number of frames per video chunk.
sample_fps: Video sampling frame rate.
timestamp_mode: Format for chunk timestamps.
Other Parameters:
ignore_index: The ignore index for the loss function.
media_placeholder_token_id: The token ID for media placeholders.
pad_token_id: The token ID for padding.
ignore_index: The ignore index for the loss function.
media_placeholder_token_id: The token ID for media placeholders.
pad_token_id: The token ID for padding.
"""
model_type = "k2_vl"
@@ -113,11 +105,6 @@ class K2VLConfig(PretrainedConfig):
self,
vision_config: dict | K2VLVisionConfig | None = None,
text_config: dict | DeepseekV3Config | None = None,
# Video-chunk parameters
temporal_merge_kernel_size: int = 4,
sample_fps: float = 2.0,
timestamp_mode: str = "hh:mm:ss.fff",
# Other parameters
ignore_index: int = -100,
media_placeholder_token_id: int = 163605,
pad_token_id: int = 0,
@@ -139,11 +126,6 @@ class K2VLConfig(PretrainedConfig):
text_config = DeepseekV3Config(**text_config)
self.text_config: DeepseekV3Config = text_config
# Video-chunk config
self.temporal_merge_kernel_size = temporal_merge_kernel_size
self.sample_fps = sample_fps
self.timestamp_mode = timestamp_mode
# Other config
self.ignore_index = ignore_index
self.media_placeholder_token_id = media_placeholder_token_id