-
Notifications
You must be signed in to change notification settings - Fork 2.7k
Pull requests: JustVugg/colibri
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(mirror): recognize Kimi K3 expert shards
#1028
opened Aug 15, 2026 by
terrizoaguimor
Contributor
Loading…
fix(convert): publish OLMoE shards atomically
#1027
opened Aug 14, 2026 by
Blakeolson21
Contributor
Loading…
fix(convert): publish resumable shards atomically
#1026
opened Aug 14, 2026 by
Blakeolson21
Contributor
Loading…
fix(docker): package the gateway's DSML module
#1025
opened Aug 14, 2026 by
Blakeolson21
Contributor
Loading…
feat(moe): FUSED3=1 opt-in AVX2 expert matmul — 40% less matmul time, bit-identical output, off by default
#1024
opened Aug 14, 2026 by
outtodata
Loading…
fix(v4): pack hot pinned experts outside state->mutex (issue #900)
#1023
opened Aug 14, 2026 by
gouravkargwal
Contributor
Loading…
[deepseek_v4] performance: use coli_v4_route_bf16 in moe_token
#1017
opened Aug 14, 2026 by
weber-software
Loading…
4 of 5 tasks
test(v4): end-to-end OpenAI wire tests for V4 DSML tool calling
#1010
opened Aug 14, 2026 by
gcaponi
Loading…
fix(doctor,planner): make DeepSeek-V4 checkpoints pass doctor and plan correctly
#1006
opened Aug 14, 2026 by
RobinHoodO
Loading…
feat(v4): dual-SSD mirror (COLI_MODEL_MIRROR) for the DeepSeek V4 expert store
#988
opened Aug 12, 2026 by
dcutugno
Contributor
Loading…
fix(v4): move hot rows16 packing out of store mutex
#983
opened Aug 12, 2026 by
aniketshukla1
Loading…
feat(v4): the expert history lives in route_trace.h now — #700 completed
#969
opened Aug 12, 2026 by
terrizoaguimor
Contributor
Loading…
feat(k3): map prepared weights read-only
#965
opened Aug 11, 2026 by
bherald
Contributor
Loading…
12 tasks done
v4: pluggable expert-store backend registry
#964
opened Aug 11, 2026 by
8PotatoChip8
Loading…
5 tasks done
cuda: zero-copy expert views on pageable-shared memory (GB10)
#936
opened Aug 11, 2026 by
Nanetnounou
Loading…
cuda: warp-per-row E8 kernels, lattice table in shared memory
#935
opened Aug 11, 2026 by
Nanetnounou
Loading…
feat(plan): discover Windows AMD GPUs without planning against them
#931
opened Aug 10, 2026 by
Kenneth-Javier
Contributor
Loading…
4 of 5 tasks
Scan vulkan devices in reverse order, to avoid main GPU
discussion
Proposta / discussione aperta, non un task
vulkan
Backend Vulkan/AMD
#917
opened Aug 10, 2026 by
xxxajk
Loading…
3 of 5 tasks
feat(moe): add DEGRADE_ZERO miss-slot zero-fill policy (issue #865)
enhancement
New feature or request
performance
Velocità / tok-s / ottimizzazioni
quality
Qualità del modello / quantizzazione
#906
opened Aug 9, 2026 by
kritikagarg
Loading…
2 of 5 tasks
fix: report Vulkan expert residency in telemetry and brain map
bug
Difetto verificato nel codice
needs-rebase
Confligge, serve rebase dell'autore
vulkan
Backend Vulkan/AMD
#891
opened Aug 8, 2026 by
MasterCATZ
Loading…
inkling: ring-buffer KV cache for sliding-window layers (~10x less KV memory at long context)
enhancement
New feature or request
performance
Velocità / tok-s / ottimizzazioni
#830
opened Aug 4, 2026 by
dpanelli
Loading…
fix(metal): int4-g64 (fmt=4) models never used the GPU — enable MoE experts + fused attention
bug
Difetto verificato nel codice
metal
Backend Metal/Apple
needs-rebase
Confligge, serve rebase dell'autore
#829
opened Aug 4, 2026 by
aaristov
Loading…
engine: notice when a tensor's format silently disables the fused Metal decode path
enhancement
New feature or request
metal
Backend Metal/Apple
#827
opened Aug 4, 2026 by
monotophic
Contributor
Loading…
Previous Next
ProTip!
Updated in the last three days: updated:>2026-08-11.