auto-round
[Accuracy improvement] Re-implement opt-rtn for both imatrix and model-free, covering INT & NVFP
#2362
Closed
Closed
[Accuracy improvement] Re-implement opt-rtn for both imatrix and model-free, covering INT & NVFP #2362
xin3he
wants to merge
15 commits into
main
from
xinhe/9-15
[EXP] support NVFP4 in model free
f4ac16ac
feat: enhance NVFP4 support with fake quantization and global scale h…
f6777ea4
feat: implement optimized RTN for NVFP4 E5M3 with scale search enhanc…
e33cf412
feat: implement fused scale updates for NVFP4 projections across outp…
f7fcdbcc
feat: add disable_opt_rtn support and enhance NVFP4 scale handling; i…
035ad11f
feat: add symmetric search offsets function for scale search; remove …
9c838a27
feat: enhance NVFP4 support with disable_opt_rtn functionality; add n…
0c91bb75
Merge remote-tracking branch 'origin/main' into xinhe/9-15
09580975
xin3he
changed the title
fix: resolve autoround merge conflict with main
54417b2f
fix: normalize legacy scheme aliases in merge resolution
5ac8b402
feat: enhance NVFP4 support with new validation and configuration fun…
e81cee1c
Merge remote-tracking branch 'origin/main' into xinhe/9-15
fbc2950d
fix: restore wrapper nvfp compatibility
703d8886
feat: enhance NVFP4 support with validation functions and improved se…
52756b84
Merge branch 'main' into xinhe/9-15
a91c92fc
Login to write a write a comment.
Login via GitHub