JWLHS
|
d2812209a1
|
chore: remove FIX comments, simplify cd dtype, remove __del__
|
2026-07-16 01:41:54 +08:00 |
|
JWLHS
|
2990416508
|
fix: register_buffer for int4 tensors, kron-based LoKr path with ceiling-div padding, remove __doc__ overrides
|
2026-07-16 01:31:11 +08:00 |
|
JWLHS
|
9c576caa7f
|
fix: register_buffer for quant tensors, kron-based LoKr path, remove __doc__ overrides
|
2026-07-16 01:18:39 +08:00 |
|
JWLHS
|
c38563749f
|
fix: preserve linear bias during INT4 reconstruction
|
2026-07-14 04:13:00 +08:00 |
|
JWLHS
|
e5fc506ffe
|
resolve rebase conflict
|
2026-07-13 21:42:21 +08:00 |
|
JWLHS
|
4c36e2d6f2
|
resolve rebase conflict
|
2026-07-13 21:38:42 +08:00 |
|
JWLHS
|
015e8a2676
|
fix: filter_fn returns True to quantize, not skip
|
2026-07-13 21:38:07 +08:00 |
|
JWLHS
|
2d0bb38f4f
|
fix: add docstrings for coverage threshold
|
2026-07-13 21:38:07 +08:00 |
|
JWLHS
|
8491b9103e
|
fix: add _lora_prerotated flag — avoid double-rotation with pre-rotated LoRA
|
2026-07-13 21:38:07 +08:00 |
|
JWLHS
|
3e5738da24
|
fix: address CodeRabbit feedback — QuaRot error surface, LoRA shape validation, Sylvester bug fix, cache cap
|
2026-07-13 21:38:07 +08:00 |
|
JWLHS
|
b5d471f5ff
|
feat: add torchao INT4 weight-only quantization backend
|
2026-07-13 21:38:07 +08:00 |
|