-
Notifications
You must be signed in to change notification settings - Fork 1
Pull requests: integrallis/models
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
perf(purejava): let the hidden-state prefill reach the batched path (15.5x)
#199
opened Sep 20, 2026 by
bsbodden
Member
Loading…
G1 root cause: Q6_K device kernel is not bit-exact past one super-block (not MoE)
#198
opened Sep 19, 2026 by
bsbodden
Member
Loading…
2 of 3 tasks
feat(models-bench): the parity and decode gate commands the GPU campaign needs
#197
opened Sep 18, 2026 by
bsbodden
Member
Loading…
feat(backend-cuda): our own Rust PTX kernels for K-quant projections and decode attention
#196
opened Sep 18, 2026 by
bsbodden
Member
Loading…
feat(tornado): run single-token attention decode on the device
#195
opened Sep 18, 2026 by
bsbodden
Member
Loading…
Gate accelerator admission on a real device budget (GPU campaign 3/3)
#193
opened Sep 18, 2026 by
bsbodden
Member
Loading…
ProTip!
Mix and match filters to narrow down what you’re looking for.