Qwen2.5 models with the tokenizers padded with extra unused special tokens to match the model embedding size. Workaround for various bugs.
Columba Labs
community
AI & ML interests
Birb make bots go brrr
Recent Activity
View all activity
Collections
1
models
44
Columbidae/medgemma-27b-multimodal
Updated
•
52
Columbidae/Qwen3-16B-A3B-Base
Updated
•
764
Columbidae/Qwen3-16B-A3B-Base-Old-Archived
Updated
•
236
•
2
Columbidae/Qwen3-16B-A3B-Tulu-Mini
Text Generation
•
Updated
•
5
Columbidae/Qwen3-21B-Base-Zeroed
Text Generation
•
Updated
•
43
•
1
Columbidae/Qwen3-21B-Base-Unzeroed
Text Generation
•
Updated
•
11
Columbidae/Qwen3-30B-A3B-Base-Noisy
Text Generation
•
Updated
•
4
Columbidae/Qwen3-30B-A3B-Noisy
Text Generation
•
Updated
•
7
Columbidae/gemma-3-12b-it-text
Updated
•
5
Columbidae/gemma-3-27b-half-full
Image-Text-to-Text
•
Updated
•
41
•
1
datasets
0
None public yet