HanGuangAI
维护: vendor
unknown pingtouge
T-Head/Pingtouge inference stack for HanGuang 800; details limited in public docs
ENGINES
推理引擎目录
7 个推理引擎 · 硬件兼容 · 维护方信息
LMDeploy
维护: mixed
0.6.0 nvidia huawei cambricon
TurboMind backend; friendly to Chinese hardware ecosystems
MindIE
维护: vendor
1.0.RC3 huawei
Huawei official inference engine for Ascend (910B/910C, CloudMatrix)
MoRI
维护: community
0.1.0 nvidia
New inference engine featured in InferenceX benchmarks
SGLang
维护: community
0.4.0 nvidia amd
High-performance serving with RadixAttention and disaggregated prefill/decode
TensorRT-LLM (Dynamo)
维护: vendor
0.14.0 nvidia
NVIDIA-only; deepest kernel optimizations and FP8/FP4 support
vLLM
维护: community
0.6.0 nvidia amd intel aws
Most widely used; ascend / rocm / musa forks for non-NVIDIA hardware