-
Notifications
You must be signed in to change notification settings - Fork 1.9k
Pull requests: jundot/omlx
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
perf(qwen35): add opt-in FP16 fused GDN decode prework
#3853
opened Sep 22, 2026 by
samfenwick
Contributor
Loading…
feat(admin): the chat page's right panel becomes the shared drawer
#3851
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
feat(admin): the console rebuilt on the shared layer
#3850
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
fix(i18n): the console's Token wording, and the values that were still English
#3848
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
feat(app): the app's own text sizes come from the console's token file
#3847
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
feat(specprefill): preserve draft recurrent state at cache boundaries
#3842
opened Sep 22, 2026 by
tc3oliver
Contributor
Loading…
feat(moe): keep the GLM-5.3 draft head resident so Lightning MTP coexists with glm5_next expert offload
#3841
opened Sep 22, 2026 by
williamxie1989
Contributor
Loading…
fix(specprefill): derive draft cache position from attention layers
#3840
opened Sep 22, 2026 by
tc3oliver
Contributor
Loading…
feat(admin): one design language for the console, and the app's text on the same ramp
#3839
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
fix(app): count in 万 in Chinese, every other language unchanged
#3826
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
fix(app): a capital T on token words and on the PP/TG labels
#3824
opened Sep 22, 2026 by
LXD-8
Contributor
Loading…
feat(app): read the log screen as records, one colour per level
#3810
opened Sep 21, 2026 by
LXD-8
Contributor
Loading…
fix(deepseek_v41): eliminate the prefill memory cliff on V4.1 Flash
#3808
opened Sep 21, 2026 by
williamxie1989
Contributor
Loading…
4 tasks done
fix(glm5_next): eliminate the prefill memory cliff on oQ2e quantized weights
#3807
opened Sep 21, 2026 by
williamxie1989
Contributor
Loading…
feat: batched DFlash drafting with small-M verify kernels
#3797
opened Sep 21, 2026 by
jundot
Owner
Loading…
feat(responses): separate tool declarations from execution and validate inputs
#3774
opened Sep 20, 2026 by
LXD-8
Contributor
Loading…
feat(responses): complete the response envelope and reasoning channels
#3773
opened Sep 20, 2026 by
LXD-8
Contributor
Loading…
perf: preserve Prism KV capacity after batches shrink
#3734
opened Sep 18, 2026 by
samfenwick
Contributor
Loading…
fix(attention): keep SDPA256 prefill on the bounded route
#3685
opened Sep 15, 2026 by
tc3oliver
Contributor
Loading…
fix(cluster): report a failed transport probe as unknown, not Ethernet (#3021, #3030)
#3657
opened Sep 13, 2026 by
jonathan308
Contributor
•
Draft
fix(mtp): keep the Qwen3.5 MTP __call__ pipeline-aware (#3518)
#3653
opened Sep 13, 2026 by
jonathan308
Contributor
•
Draft
feat(embedding): promote bfloat16 checkpoints to float16 for conformance
#3640
opened Sep 13, 2026 by
ernestas-poskus
Contributor
Loading…
Previous Next
ProTip!
What’s not been updated in a month: updated:<2026-08-22.