PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度
Commits on main
Support auto-round-lib build with oneAPI 2026.0 and work with pytorch xpu 2.13 (#2043)
4d24c40 chen, suyue committed 4 months ago
Remove transformers version limit for release package (#2030)
5e586bc chen, suyue committed 4 months ago
bug-fix: implement dedicated cache for streaming source shards (#2021)
ffd3dfe Xin He committed 4 months ago
fix: CLI ignore --scheme when bits/act_bits are unset (#2019)
c70e99f Heng Guo committed 4 months ago
fix the mismatch between weight and scale in gguf format (#2014)
b482d15 Heng Guo committed 4 months ago
Add moe prefill/ decode with int2/int4/int8 sym /asym and fp8 e4m3 e5m2 (#1813)
7414370 Copilot committed 4 months ago
[ARK] feat: expose smooth_k (use_mean_bias) parameter from Python to C++ in SAGE attention kernels (#1995)
d55ba00 Luo Yu committed 4 months ago
Fix cuda ut fail caused by used instructions for base model (#1994)
43f443e Heng Guo committed 4 months ago
feat: add return_lse support for SDPA and SAGEV1 kernel — enable sequence-parallel (Ring Attention) merge (#1991)
b481ceb Luo Yu committed 4 months ago
Implement AutoScheme support for model-free packing with validation a… (#1990)
cd10607 Xin He committed 4 months ago
[AutoScheme] fix lm_head no_grad issue and support MOE model with shared mix_score (#1971)
e6b35d1 Xin He committed 4 months ago
fix: use spawn instead of fork for dataset subprocess on macOS (#1980)
1a7ed49 Ramakrishnan Sathyavageeswaran committed 4 months ago
fix: promote None-initialized kwarg to list in forward_capture (#1979)
0622d47 Ramakrishnan Sathyavageeswaran committed 4 months ago
fix: avoid silent weight loss when OffloadManager fails to save a block (#1982)
6437841 Xin He committed 4 months ago
rename the base rotation (deterministic/random hadamard) directory name. (#1956)
45dcb19 lkk committed 4 months ago