Resources for “Understanding and Harnessing Sparsity in Unified Multimodal Models”, including the paper and efficient BAGEL-MoE checkpoints.
AI & ML interests
Efficient and adaptive foundation models across language and multimodal intelligence.
Recent Activity
Papers
Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?
Demystifying When Pruning Works via Representation Hierarchies
Resources for paper “Making Large Language Models Efficient Dense Retrievers” (ACL 2026), including the paper and efficient Mistral-based dense retrie
-
yibinlei/effir-mistral-drop-8-mlp
Text Generation • 7B • Updated • 107 • 3 -
yibinlei/effir-mistral-drop-16-mlp
Text Generation • 7B • Updated • 81 • 2 -
yibinlei/effir-mistral-drop-8-attn
Text Generation • 7B • Updated • 85 • 2 -
yibinlei/effir-mistral-drop-16-attn
Text Generation • 7B • Updated • 99 • 2
Resources for “Understanding and Harnessing Sparsity in Unified Multimodal Models”, including the paper and efficient BAGEL-MoE checkpoints.
Resources for Drop-Then-Recovery (DTR), including VLADrop checkpoints for studying and recovering redundancy in vision-language-action models.
Resources for paper “Making Large Language Models Efficient Dense Retrievers” (ACL 2026), including the paper and efficient Mistral-based dense retrie
-
yibinlei/effir-mistral-drop-8-mlp
Text Generation • 7B • Updated • 107 • 3 -
yibinlei/effir-mistral-drop-16-mlp
Text Generation • 7B • Updated • 81 • 2 -
yibinlei/effir-mistral-drop-8-attn
Text Generation • 7B • Updated • 85 • 2 -
yibinlei/effir-mistral-drop-16-attn
Text Generation • 7B • Updated • 99 • 2
Resources for studies on redundancy in LLMs, including layer dropping and representation-hierarchy-based pruning.