[TRTLLM-15177][chore] Kimi K3 post-merge cleanup: config/import/test hygiene + L0 wiring - #17413

Merged

brnguyen2 merged 13 commits into

NVIDIA:mainfrom

brnguyen2:k3/15177-cleanup

Aug 10, 2026

Commits on Aug 10, 2026

[TRTLLM-15177][chore] use immutable tuple for KimiLinearConfig.keys_to_ignore_at_inference
brnguyen2
committed
9ec5f78
[TRTLLM-15177][chore] modeling_kimi_linear.py import + validation hygiene
brnguyen2
committed
258ef87
[TRTLLM-15177][chore] tidy KDA fused-verify parity test
brnguyen2
committed
06dc2e1
[TRTLLM-15177][test] wire Kimi K3 unit tests into l0_b200
brnguyen2
committed
22d9dee
[TRTLLM-15177][chore] apply ruff-format to KDA parity test
brnguyen2
committed
8f0086b
[TRTLLM-15177][chore] split Kimi K3 load_weights and extract the FP8 swap helper
brnguyen2
committed
b7da478
[TRTLLM-15177][chore] move the test-only Kimi K3 reference MoE block out of the runtime package
brnguyen2
committed
559bef5
[TRTLLM-15177][chore] address review: config ValueError, loader annotations, guarded test imports
brnguyen2
committed
0e128b7
[TRTLLM-15177][test] Unwire KDA prefill-op suite from l0_b200 pending B200 NaN investigation
brnguyen2
committed
ae0a542
[TRTLLM-15204][test] Cite the B200 KDA prefill NaN ticket in the l0_b200 note
brnguyen2
committed
a8f650d
Address trivial review comments
brnguyen2
committed
fcc4a8b
Address trivial review comments
brnguyen2
committed
8d4bebe
Return an int from _swap_linear_to_fp8_weight_read
brnguyen2
committed
d97a73b