-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Pull requests: jundot/omlx
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat(admin): add max_tokens override for external accuracy benchmarks
#2863
opened Aug 19, 2026 by
williamxie1989
Contributor
Loading…
SpecPrefill: enable DeepSeek/GLM MLA draft models (query extractor, RoPE dtype, DSA/CacheList positions)
#2856
opened Aug 19, 2026 by
anicaise-ai
Loading…
feat(qwen35) Qwen ANE prefill: FP16 CPU work sharing and performance-aware scheduling
#2853
opened Aug 19, 2026 by
onthehub97
Contributor
•
Draft
fix(specprefill): enable SpecPrefill for MLA-family models (DeepSeek/GLM)
#2851
opened Aug 19, 2026 by
anicaise-ai
Loading…
fix(cluster): require bidirectional bound-TCP for link reachability (fabric-doctor C1)
#2849
opened Aug 19, 2026 by
alytaphoenix
Loading…
feat(cluster): transport readiness ladder with reason/remedy copy (fabric-doctor B3)
#2848
opened Aug 19, 2026 by
alytaphoenix
Loading…
feat(cluster): serve VLM checkpoints text-only under distributed inference
#2845
opened Aug 19, 2026 by
alytaphoenix
Loading…
fix(cluster): make quantized Nemotron-H run tensor-parallel at TP=2
#2844
opened Aug 19, 2026 by
alytaphoenix
Loading…
Cluster: preserve remote node memory guard tier and custom ceiling in planner budgets
#2837
opened Aug 18, 2026 by
xunlinkx
Contributor
Loading…
fix(cluster): initialize settings before memory guard fallback
#2835
opened Aug 18, 2026 by
xunlinkx
Contributor
Loading…
feat(admin): add exact-match mode to the log level filter
#2832
opened Aug 18, 2026 by
thebigjc
Loading…
fix(responses): honor tool_choice=none, fix MCP tool merge
#2830
opened Aug 18, 2026 by
astutegrogan
Loading…
fix(ane): reject tuning candidates whose ANE never executed
#2829
opened Aug 18, 2026 by
beaglemoo
Loading…
perf(laguna): port mlxfast-challenge decode optimizations (bit-exact verified)
#2827
opened Aug 18, 2026 by
0xClandestine
Contributor
Loading…
fix(qwen35) ANE tuner early exit results, q6/q8 support, GDN activation path fix
#2826
opened Aug 18, 2026 by
onthehub97
Contributor
Loading…
fix(setting): accept SOCKS proxy URLs in network settings validation
#2821
opened Aug 18, 2026 by
asxzy
Loading…
fix(cluster): two-Mac clustering reliability (poll jitter, VLM tensor default, cross-user paths, ssh_user)
#2819
opened Aug 18, 2026 by
alytaphoenix
Loading…
fix(cli): strip leaked -- separator from launch extra_args
#2818
opened Aug 18, 2026 by
fparrav
Contributor
Loading…
2 of 3 tasks
fix(responses): surface max_output_tokens truncation as status="incomplete"
#2817
opened Aug 18, 2026 by
mvdbos
Contributor
Loading…
perf(cache): serve multi-turn cache hits from the live GPU chain and SSD, not a CPU hot-cache copy
#2810
opened Aug 18, 2026 by
hojin12312
Contributor
Loading…
feat(deepseek): experimental hybrid ANE prompt processing
#2803
opened Aug 18, 2026 by
jundot
Owner
Loading…
glm_moe_dsa: M2-tuned DSA indexer score kernel, 1.37x over Steel on D…
#2802
opened Aug 18, 2026 by
nathanodle
Loading…
Previous Next
ProTip!
no:milestone will show everything without a milestone.