Inference Recipes

面向 LLM / 多模态推理部署与加速的实践合集。

Repo: https://gitcode.com/cann/cann-recipes-infer

Featured Recipes

Card	Level	Description	Link
GPT-OSS	Beginner	20B 单卡 / 120B 8 卡推理	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/gpt-oss/README.md
LongCat-Flash	Intermediate	低时延推理，支持控核与权重预取	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/longcat-flash/README.md
HunyuanImage-3.0	Intermediate	CFG / VAE 并行 + Operator Fusion	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/hunyuan-image-3.0/README.md
Qwen3-MoE	Intermediate	Atlas A3 推理适配，支持 TP / EP 部署	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/qwen3-moe/README.md
HunyuanVideo	Intermediate	xDiT 推理，支持 Ulysses / RingAttention / TeaCache	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/hunyuan-video/README.md
Wan2.2-I2V	Intermediate	Transformers 推理适配	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/wan2.2-i2v/README.md
DeepSeek-R1 / Kimi-K2	Intermediate	低时延与高吞吐部署，支持 DP / TP+SP / EP	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/deepseek-r1/README.md
Kimi-K2-Thinking	Advanced	256K 长序列推理，原生量化 W4A16	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/kimi-k2-thinking/README.md
DeepSeek-V3.2-Exp	Advanced	CP 并行 + 大 EP 并行，Operator Fusion 与 Multi-Stream 优化	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/deepseek-v3.2-exp/README.md