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
[Regression] Detach scale tensor to prevent holding computation graph references (#1389)
1c758f4 Xin He committed 9 months ago
Support general MOE replacement for MOE models (Transformers 5.0 compatible) (#1374)
28839a9 Liang Lv committed 9 months ago
Add Links; Add step_by_step_CN; Fix several grammatical issues (#1366)
3b7b87e ZaneMark committed 9 months ago
feat: handle per-tensor FP8 dequantization for Devstral models (#1356)
eddb6de SwekeR committed 9 months ago
support qwen3-next/glm4-moe/deepseekv3 monkey patch for transformers 5 (#1367)
0c0a0c1 Wenhua Cheng committed 9 months ago
support qwen3-moe/glm-4.7-flash monkey patch for transformers 5 (#1354)
6bc1d34 Wenhua Cheng committed 9 months ago
Refactor FP8 dequantization and detection using registry pattern (#1348)
eaabde4 Adithyan committed 9 months ago
refactor auto_round api, using local symbol table to pass all args (#1358)
528c9c8 Heng Guo committed 9 months ago
Fix issue where predefined ignore layers incorrectly apply to GGUF (#1340)
ca13362 Wenhua Cheng committed 9 months ago