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: allocate calibration samples correctly when mixing datasets (#2526)
436fa5a UNKNOWN H4CK3R committed 1 month ago
fix: keep escapes and quantified dots in to_standard_regex (#2418)
fd28e78 Shubham Padkonde committed 1 month ago
fix: use spawn start method for vLLM evaluation to avoid CUDA re-initialization error (#2421)
234ac7a Reginald Alfret V committed 1 month ago
fix: preserve calibration dataset on memory errors (#2433)
fde1f58 Shubham Padkonde committed 1 month ago
feat: support LLMC/ModelOpt/Quark format NVFP4 -> MXFP4 conversion (#2492)
7222a20 Xin He committed 1 month ago
Support Nemotron-H Omni wrappers (Nemotron 3.5 Super VL, Bolt 74B-A9B) in plain AutoRound calls (#2512)
ec10a64 SergiioB committed 1 month ago
feat: implement Kimi remote MoE with grad-aware inference and add uni… (#2496)
88dc729 Xin He committed 1 month ago
remove leaderboard entry for now as it's still not very stable (#2516)
650981e Wenhua Cheng committed 1 month ago
docs(examples): align recipes with recommended W2 and batch size settings (#2499) (#2511)
a0134a4 Vishva R committed 1 month ago
[Experimental]Add generic HF dataset support and two more datasets (#2363)
e31b772 Wenhua Cheng committed 1 month ago
fix: adjust tolerance levels for performance metrics comparison (#2505)
2c37748 Sun, Xuehao committed 1 month ago
fix: quantize all diffusion transformers in RTN routes (#2497)
9645350 Wang, Chang committed 1 month ago
feat(examples): add batch 2 quantization recipes for Muse-Glimmer, NuExtract3, and Qwen3.5 (#2498)
da52d6d Vishva R committed 1 month ago
fix: enable imatrix calibration for MX and NV optimized RTN (#2494)
94aa6d4 Wang, Chang committed 1 month ago
fix: enhance attention module with query observation and legacy support (#2466)
376fdc0 Xin He committed 1 month ago
Add quantization recipe notebooks for Qwen, OCR, and edge models (#2488)
504aea5 Vishva R committed 1 month ago
Don't rewrite packed modules as float weights when resuming (#2386)
4663212 Shubham Padkonde committed 1 month ago
fix: update field extraction for ServerArgs in _run_sglang_worker (#2474)
a065121 Xin He committed 1 month ago
fix: update dependencies to include timm in unit test environment (#2473)
421ae1a Sun, Xuehao committed 1 month ago
fix: skip layers the AWQ GEMM kernel cannot serve in AWQ export (#2456)
7a660e1 losewayy committed 1 month ago
fix: enhance model evaluation preparation for non-task-based evaluation (#2468)
1f421d6 Xin He committed 1 month ago