SIGN IN SIGN UP

fix(checkpoint): validate PEFT adapter-only state (#3501)

* fix(ci): keep Nemotron Flash robustness on TP1

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): isolate Qwen MoE checkpoint integrity gates

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): diagnose Qwen source parity instability

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(checkpoint): compare PEFT adapter-only state

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): verify Qwen source expert weights

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): tolerate FSDP expert parameter prefixes

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): inspect Qwen expert state dict

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

* test(ci): remove Qwen parity diagnostics

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>

---------

Signed-off-by: Alexandros Koumparoulis <akoumparouli@nvidia.com>
A
Alexandros Koumparoulis committed
1077546afa131b92f57d2879229c221c5d0df8ae
Parent: 42af068
Committed by GitHub <noreply@github.com> on 8/12/2026, 3:27:39 AM