-
Notifications
You must be signed in to change notification settings - Fork 2.2k
Pull requests: antirez/ds4
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
GLM 5.3 Flash on the DGX Spark: hits-first generalises to a second model family (stacks on #1083)
#1098
opened Sep 20, 2026 by
triplesparkle
Loading…
GLM 5.3 Flash on the DGX Spark: measured, and q4 on CUDA (#907)
#1097
opened Sep 20, 2026 by
triplesparkle
Loading…
Compare the restored gate timeout with what the kernel keeps
#1096
opened Sep 19, 2026 by
namespaceMarcello
Loading…
Fix the GGUF block sizes of iq1_s and iq4_nl
#1095
opened Sep 19, 2026 by
namespaceMarcello
Loading…
server: keep the GLM live checkpoint across tool turns
#1093
opened Sep 19, 2026 by
cropduster
Contributor
Loading…
metal: speed up GLM Flash and DeepSeek V4.1 on M3 Ultra
#1090
opened Sep 19, 2026 by
trueimage
Loading…
v4.1: rewind live sessions instead of rebuilding them
#1089
opened Sep 19, 2026 by
gaineyllc
Loading…
server: recover unterminated reasoning as content when the turn completed
#1088
opened Sep 18, 2026 by
cropduster
Contributor
Loading…
server: in-process Prometheus metrics via GET /metrics
#1086
opened Sep 18, 2026 by
jyatesdotdev
Contributor
Loading…
Stream assistant content incrementally when tools are enabled
#1084
opened Sep 18, 2026 by
zonebond
Loading…
CUDA SSD streaming: hits-first expert launch, the Metal split's counterpart (+5 to +8 % decode)
#1083
opened Sep 18, 2026 by
triplesparkle
Loading…
cuda: give ldmatrix the .shared state space in token-tile attention
#1079
opened Sep 18, 2026 by
fulvian
Loading…
speed-bench: Qwen3.8 Flash Next Q2 numbers for M1 Max 64 GB
#1068
opened Sep 17, 2026 by
slycrel
Loading…
Metal V4.1 decode: one command-buffer wait per token, Engram reads on threads
#1067
opened Sep 17, 2026 by
notatestuser
Loading…
GLM: render the tool call block inline, as the chat template does
#1065
opened Sep 16, 2026 by
aaa2015
Loading…
Qwen3.8 Flash Next: batched decode and MTP across sessions on Metal (16 streams: 203 tok/s aggregate), single stream +8%
#1062
opened Sep 16, 2026 by
polymorf
Contributor
Loading…
metal: decode-shaped V4.1 indexer kernels for long contexts
#1061
opened Sep 15, 2026 by
gaineyllc
Loading…
server: find token-text disk stores by the request's rendered tokens
#1059
opened Sep 15, 2026 by
gaineyllc
Loading…
Previous Next
ProTip!
What’s not been updated in a month: updated:<2026-08-19.