By User Level

根据经验水平选择路径。每个条目都指向可运行的样例或优化说明。

Beginner Quickstarts

Card	Description	Link
LLM Inference: GPT-OSS	体量较小、部署路径清晰，适合快速完成推理上手	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/gpt-oss/README.md
LLM Training: Qwen2.5 RL	单卡 Atlas A2 上手，训练流程完整	https://gitcode.com/cann/cann-recipes-train/-/blob/master/llm_rl/qwen2_5/verl_npu_demo/README.md
Embodied: Pi0 (Torch)	端到端推理流程清晰，含 Operator Fusion 与 Graph Mode	https://gitcode.com/cann/cann-recipes-embodied-intelligence/-/blob/master/manipulation/pi0/infer_with_torch/README.md
Spatial: VGGT	评测与推理脚本完整，门槛相对低	https://gitcode.com/cann/cann-recipes-spatial-intelligence/-/blob/master/models/vggt/README.md
HarmonyOS: SobelCustom	自定义算子场景简单，便于理解端侧部署流程	https://gitcode.com/cann/cann-recipes-harmony-infer/-/blob/master/ops/ascendc/docs/custom-npu_sobel.md

Intermediate Guides

Card	Description	Link
LLM Inference: LongCat-Flash	低时延推理，包含 Multi-Stream 并行与权重预取	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/longcat-flash/README.md
LLM Inference: HunyuanImage-3.0	CFG/VAE 并行 + Operator Fusion	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/hunyuan-image-3.0/README.md
LLM Training: Qwen3 Tool Agent RL	引入 asyncLLM 与 agent_loop	https://gitcode.com/cann/cann-recipes-train/-/blob/master/agent_rl/qwen3_tool_agent/README.md
Embodied: Spirit v1.5	端侧推理流程完善	https://gitcode.com/cann/cann-recipes-embodied-intelligence/-/blob/master/manipulation/spirit-v1.5/infer_with_torch/README.md
Spatial: Hunyuan3D	Operator Fusion + Graph Mode + 并行光栅化	https://gitcode.com/cann/cann-recipes-spatial-intelligence/-/blob/master/models/Hunyuan3D/README.md
HarmonyOS: QuantMatmul	端侧量化算子实践	https://gitcode.com/cann/cann-recipes-harmony-infer/-/blob/master/ops/ascendc/docs/custom-npu_quantMatmul.md

Advanced Deep Dives

Card	Description	Link
LLM Inference: DeepSeek-V3.2-Exp	CP/EP 并行 + Fusion Kernel + Multi-Stream	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/deepseek-v3.2-exp/README.md
LLM Inference: Kimi-K2-Thinking	256K 长序列 + 并行策略细节	https://gitcode.com/cann/cann-recipes-infer/-/blob/master/models/kimi-k2-thinking/README.md
LLM Training: Qwen3-235B Long Seq RL	长序列 RL 调度与 Graph Mode 优化	https://gitcode.com/cann/cann-recipes-train/-/blob/master/llm_rl/qwen3/README.md
LLM Training: DeepSeek-R1 RL	系统级调度优化与 Multi-Stream 策略	https://gitcode.com/cann/cann-recipes-train/-/blob/master/llm_rl/deepseek/README.md
Embodied: Pi0.5 Training	训练流程与性能调优	https://gitcode.com/cann/cann-recipes-embodied-intelligence/-/blob/master/manipulation/pi05/train/README.md
Spatial: 3DGS	AscendC 自定义算子与负载均衡	https://gitcode.com/cann/cann-recipes-spatial-intelligence/-/blob/master/algorithms/gaussian_splatting/README.md
HarmonyOS: AscendC Dev Guide	完整算子开发与入图流程	https://gitcode.com/cann/cann-recipes-harmony-infer/-/blob/master/docs/ascendc_develop_guide.md