ZefanCai/LastKV-LLM-8B
Updated
LastKV
Model collection for Last Layer KV Cache is All You Need. LastKV shares final-layer KV history across attention layers.
-
-
ZefanCai/LastKV-LLM-8B-Base
Text Generation • 8B • Updated • 23 -
ZefanCai/LastKV-VLM-8B
Updated -
ZefanCai/LastKV-H3
Updated
ZefanCai/LastKV-Wan-1.3B
Text-to-Video • 1B • Updated • 11Note 480p causal text-to-video generation based on Wan2.1-T2V-1.3B with LastKV. Step 20,500; FP32 generator weights. EMA is the default, with raw weights also available.
ZefanCai/LastKV-Wan-14B
Text-to-Video • 14B • Updated • 75Note 480p causal text-to-video generation based on Wan2.1-T2V-14B with LastKV. Step 10,000; FP32 generator weights. EMA is the default, with raw weights also available.
ZefanCai/LastKV-NVS-Objaverse
UpdatedNote Object-centric novel-view synthesis. 1024 × 1024 resolution, 24 layers, 184.8M parameters. Original PyTorch training checkpoint.
ZefanCai/LastKV-NVS-DL3DV
UpdatedNote Scene-level novel-view synthesis. 512 × 512 resolution, 24 layers, 184.8M parameters. Original PyTorch training checkpoint.