feat(local-ai): route RTX Spark unified-memory SKUs through NVIDIA's fixed recipe table - #1520

Merged

shanselman merged 13 commits into

mainfrom

maintainer/rtx-spark-recipes-main

Sep 29, 2026

Commits on Sep 25, 2026

feat(local-ai): detect RTX Spark by driver-reported GPU name

joelagnel
authored and
karkarl
committed
9378284
feat(local-ai): route RTX Spark defaults through a fixed SKU table

joelagnel
authored and
karkarl
committed
fb60866
feat(local-ai): branch llama-server preset on speculative decoding mode

joelagnel
authored and
karkarl
committed
b7d50a5
test(local-ai): cover RTX Spark SKU routing

joelagnel
authored and
karkarl
committed
92f1ec2
feat(local-ai): account for the draft checkpoint in memory qualification

joelagnel
authored and
karkarl
committed
e8ce55f
build(local-ai): bump managed llama-server runtime to b11026

joelagnel
authored and
karkarl
committed
685e8e0
feat(local-ai): add schema-5 manifest support for additional model assets

joelagnel
authored and
karkarl
committed
00aa95f
feat(local-ai): acquire additional model assets via the HF hub cache

joelagnel
authored and
karkarl
committed
d41a0fd
feat(local-ai): wire additional-asset acquisition into setup and reconcile

joelagnel
authored and
karkarl
committed
67d7f39
feat(local-ai): thread the verified DFlash draft identity into the launch preset

joelagnel
authored and
karkarl
committed
755a9ba
fix(local-ai): rank, verify, and disclose models by total download size

joelagnel
authored and
karkarl
committed
32cd796
fix(local-ai): defer to the generic path when RTX Spark memory facts are incomplete

joelagnel
authored and
karkarl
committed
7e51993

Commits on Sep 29, 2026

fix(local-ai): migrate and roll back retired runtime upgrades safely
karkarl
committed
7d25102