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
[risky]reduce ram usage of loading moe without monkey patch and fix qwen3-flash-next group_size issue (#2291)
7d8905e Wenhua Cheng committed 2 months ago
Optimize INT4 MoE prefill/decode (sym) XPU kernels with dedicated w4a16 tiling (#2111)
21bab55 Copilot committed 2 months ago
Allow non-MLLM calibration datasets when not quantizing vision (#2065)
63dbc8b Fabrizio del Tin committed 2 months ago
feat(orchestrator): add reference_output to quantize_block to skip collect_reference (#2289)
c9017ff X_Bee committed 2 months ago
Enhance test logging and retry mechanism for unit tests (#2274)
b48c9db chen, suyue committed 2 months ago
fix: skip activation fake-quant in FakeActQuantLinear when act quantization is not configured (#2286)
888b0f1 Heng Guo committed 2 months ago
fix: replace bare except clauses with except Exception (#1981)
81b4db3 Ramakrishnan Sathyavageeswaran committed 2 months ago
fix: asymmetric zero-point packing, format-scoped 8-bit asym, W5/W6/W7 llm_compressor presets (#2273)
5d2f884 avtc committed 2 months ago
fix(diffusion): use real images for I2V calibration (#2272)
1df1cef Wang, Chang committed 2 months ago
support custom max shard size in CLI and model export (#2258)
5bc238f Heng Guo committed 2 months ago
Enhance CI coverage reporting with hardware-specific configurations (#2259)
1057f0e Sun, Xuehao committed 2 months ago
feat: Add support for deterministic algorithms flags in CLI and update logging behavior (#2261)
d72933b Xin He committed 2 months ago
Bug fix: Replace WAWrapperLinear with FakeActQuantLinear instead of Linear (#2245)
6b29e9c Xin He committed 2 months ago
fix: warm up Triton CUDA driver before mocking device capability (#2253)
a97b91a Sun, Xuehao committed 2 months ago
disable torch compile in rtn/opt-rtn/iters<10 and fix glm-5.3 tuning exception (#2246)
4dbbd15 Wenhua Cheng committed 2 months ago