cherry pick Habana software v1.23.0#2380

Merged

XuehaoSun merged 19 commits intomasterfrom

xinhe/cherry-pick-v1.23.0

Jan 12, 2026

Commits on Jan 5, 2026

[SW-229385] Select correct MOE op for H2D scales (#280)

ulivne
authored and
xin3he
committed
[SW-233498] Fix scale calculation cguid flow in all scale methods (#263)

authored and
xin3he
committed
[SW-220805] Fixing zero values with FSDPA (#252)

HolyFalafel
authored and
xin3he
committed
[SW-230070] Changed dynamic quant check to op string instead of type, also in ops_quantizer (#248)

HolyFalafel
authored and
xin3he
committed
[SW-237232] add support in SGLang (#291)

linoybu
authored and
xin3he
committed
[SW-237037] add support for BLOCK_SOFTMAX_CONST_MAX (#292)

linoybu
authored and
xin3he
committed
Fix get calls (#295)

authored and
xin3he
committed
[SW-239679] temporary fix for static quant test (#298)

ulivne
and
xin3he
committed
[PERFC-270] add xpu qdq tests using inc (#296)

nirda7
authored and
xin3he
committed
[SW-239033] Using cguid to calculate maxabs in dynamic quant only (#297)

HolyFalafel
authored and
xin3he
committed
[SW-239190] Fix Deepseek hang issue w/ H2D scalar (#299)

authored and
xin3he
committed
Add support for FP8 static quantization for optimum-habana deepseek v3/r1 (#203)

skavulya
authored and
xin3he
committed
Pass dtype to torch scalar (#300)

authored and
xin3he
committed
pass extra args to moe (#305)

authored and
xin3he
committed
[SW-240561]Requant LLMC FP8 model (#304)

authored and
xin3he
committed
[SW-241261] Support SharedFusedMoE (#313)

ulivne
authored and
xin3he
committed

Commits on Jan 8, 2026

Commits on Jan 11, 2026

[GAUDISW-245272] disable layer-wise test
xinhe3
committed