[OMNIML-5899] Add IQ quantization codecs and backend - #2446

Merged

hychiang-git merged 11 commits into

mainfrom

hungyuehc/omniml-5899-quantization-v2

Sep 18, 2026

Commits on Sep 17, 2026

Add IQ quantization codecs and backend
hychiang-git
committed
32d61bc
Test IQ quantization codecs
hychiang-git
committed
887d357
Fix IQ codec review findings
hychiang-git
committed
d40192f
Test IQ backend routing and CUDA parity
hychiang-git
committed
9af3c13
Address IQ codec review findings
hychiang-git
committed
e8d9370
Merge branch 'main' into hungyuehc/omniml-5899-quantization-v2
hychiang-git
authored
e088208
[OMNIML-5899] Bound IQ fake-quant decode memory
hychiang-git
committed
5b465df
Merge branch 'main' into hungyuehc/omniml-5899-quantization-v2
cjluo-nv
committed
0007c0e
Adopt the consolidated GGML CUDA extension

cjluo-nv
and
claude
committed
e4f6270
Move the IQ codebooks into their own module

cjluo-nv
and
claude
committed
16712b6
Saturate finite out-of-range float64 weights in the reference encoders

cjluo-nv
and
claude
committed
f342df4