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
fix: preserve declared FP32 tensors during model dtype conversion (#2359)
b33eedc Wang, Chang committed 2 months ago
fix: support list inputs in diffusion tuning cache (#2346)
c53611f Wang, Chang committed 2 months ago
fix: update key cache access for compatibility with newer DynamicCache implementation (#2348)
f331bb0 chen, suyue committed 2 months ago
fix: reload Diffusers blocks from component checkpoints (#2353)
6a8d95f Wang, Chang committed 2 months ago
feat: implement resume functionality for model-free compression (#2330)
9e2650f Xin He committed 2 months ago
Keep the original checkpoint key names and fix Qwen4 vLLM inference (#2327)
aba0abd Wenhua Cheng committed 2 months ago
fix: recreate calibrator for secondary diffusion transformers (#2335)
cea5abd Wang, Chang committed 2 months ago
fix: preserve layer config for secondary diffusion transformers (#2337)
c78c11a Wang, Chang committed 2 months ago
XPU oneDNN: bind matmul execution to live SYCL queue (graph-capture safe) (#2280)
79516ff Copilot committed 2 months ago
feat: add transformers checkpoint conversion mapping support for missing tensors (#2328)
1139c32 Xin He committed 2 months ago
feat: add support for WOQ fake quantization in model-free mode and update… (#2329)
bdc6047 Xin He committed 2 months ago
refactor: reorganize tensor utility functions and enhance split_fused_expert_tensors for special cases (#2320)
c18bf77 Xin He committed 2 months ago
fix: pin Hugging Face snapshots for AutoScheme offload reloads (#2318)
34c385a Heng Guo committed 2 months ago
[high risk]Speedup 2.5-6X moe tuning and pin ngram to cpu (#2303)
678f8f5 Wenhua Cheng committed 2 months ago
fix: update artifact name format for ark wheel download (#2321)
b173803 chen, suyue committed 2 months ago
Implement AI-Powered CI Failure Analysis and Fix Recommendations (#2277)
18486c3 chen, suyue committed 2 months ago
fix: defer packing for multi-transformer diffusion (#2314)
b4dc0cc Wang, Chang committed 2 months ago
fix: update quantization configuration handling in FakeFormat and related tests (#2311)
777c4ea Xin He committed 2 months ago
support 5,6,7 bits on cuda and add humming backend (#2281)
1f23576 Wenhua Cheng committed 2 months ago