Repository navigation
Issues
Issues
CUDA prefill: separate cuBLAS GEMMs for F16 q, k, v and gate, up instead of contiguous weight copies
beehive-lab/jitllm#265 · openedon Nov 4, 2026 - beehive-lab/jitllm#250 · opened
on Oct 29, 2026 Fall back to the JIT kernels when
--with-native-librarieshas no native path, and keep a strict modebeehive-lab/jitllm#231 · openedon Oct 22, 2026 - beehive-lab/jitllm#215 · opened
on Oct 15, 2026 - beehive-lab/jitllm#214 · opened
on Oct 15, 2026 - beehive-lab/jitllm#213 · opened
on Oct 15, 2026 - beehive-lab/jitllm#212 · opened
on Oct 15, 2026 - beehive-lab/jitllm#211 · opened
on Oct 15, 2026 - beehive-lab/jitllm#204 · opened
on Oct 13, 2026 - beehive-lab/jitllm#203 · opened
on Oct 13, 2026 - beehive-lab/jitllm#178 · opened
on Sep 25, 2026 - beehive-lab/jitllm#143 · opened
on Aug 2, 2026 - beehive-lab/jitllm#130 · opened
on Jul 11, 2026 - beehive-lab/jitllm#43 · opened
on Aug 5, 2025 - beehive-lab/jitllm#16 · opened
on Jun 9, 2025