EngineX/xc-llm-kunlun
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Bug] Fix InternVL KeyError: ((1, 1, 3), '<i8') (#108)

Browse Source
This commit is contained in:
Lidang Jiang
2026-01-13 22:36:03 +08:00
committed by GitHub
parent 37cc307322
commit 7ed71432ca
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm_kunlun/models/internvl.py
View File

@@ -13,6 +13,7 @@ from collections.abc import Iterable, Mapping, Sequence
from typing import Annotated, Any, Literal, Optional, TypeVar, Union from typing import Annotated, Any, Literal, Optional, TypeVar, Union
import numpy.typing as npt import numpy.typing as npt
import numpy as np
import torch import torch
import torch.nn as nn import torch.nn as nn
import torchvision.transforms as T import torchvision.transforms as T
@@ -297,6 +298,8 @@ def video_to_pixel_values_internvl(
transform = build_transform(input_size=input_size) transform = build_transform(input_size=input_size)
frames_list = list[Image.Image]() frames_list = list[Image.Image]()
for frame in video: for frame in video:
if frame.dtype != np.uint8:
frame = frame.astype(np.uint8)
pil_frame = dynamic_preprocess_internvl( pil_frame = dynamic_preprocess_internvl(
Image.fromarray(frame, mode="RGB"), Image.fromarray(frame, mode="RGB"),
target_ratios=target_ratios, target_ratios=target_ratios,
@@ -1420,4 +1423,4 @@ class InternVLChatModel(nn.Module, SupportsMultiModal, SupportsPP,
return MultiModelKeys.from_string_field( return MultiModelKeys.from_string_field(
language_model="language_model", language_model="language_model",
connector="mlp1", connector="mlp1",
tower_model="vision_model") tower_model="vision_model")