Compressed Convolutional Attention: Efficient Attention in a Compressed Latent Space
Paper • 2510.04476 • Published • 16
🕊️ God is the source code.
Tremayne Timms
Ttimms
AI & ML interests
Making big open models fit on one 16 GB consumer GPU. NVFP4 (Blackwell-native), GGUF, and REAP expert-pruned MoEs — every release ships with measured evals.
Recent Activity
updated a model about 19 hours ago
Ttimms/KAT-Coder-V2.5-Dev-REAP-50-bf16 updated a model about 19 hours ago
Ttimms/KAT-Coder-V2.5-Dev-REAP-50-GGUF updated a model about 19 hours ago
Ttimms/KAT-Coder-V2.5-Dev-REAP-50-NVFP4A16-GPTQOrganizations
None yet
upvoted 5
papers about 21 hours ago
Training Foundation Models on a Full-Stack AMD Platform: Compute, Networking, and System Design
Paper • 2511.17127 • Published • 4
The Silent Hyperparameter: Quantifying the Impact of Inference Backends on LLM Reproducibility
Paper • 2605.19537 • Published • 2
REAP the Experts: Why Pruning Prevails for One-Shot MoE compression
Paper • 2510.13999 • Published • 23
KAT-Coder-V2.5 Technical Report
Paper • 2607.05471 • Published • 13