Co-authored-by: litianjian <litianjian@bytedance.com>
This commit is contained in:
litianjian 2024-09-21 07:34:42 +08:00 committed by GitHub
parent e9356a4206
commit 7b2b536a81
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -475,8 +475,8 @@ class LlavaOnevisionForConditionalGeneration(LlavaOnevisionPreTrainedModel):
image_features = image_features.view(batch_frames, height, width, -1)
image_features = image_features.permute(0, 3, 1, 2).contiguous()
height, weight = image_features.shape[2:]
scaled_shape = [math.ceil(height / 2), math.ceil(weight / 2)]
height, width = image_features.shape[2:]
scaled_shape = [math.ceil(height / 2), math.ceil(width / 2)]
image_features = nn.functional.interpolate(image_features, size=scaled_shape, mode="bilinear")
image_features = image_features.permute(0, 2, 3, 1)