MindSpeed-LLM MindSpore后端稀疏模型支持
| 模型 | 下载链接 | 脚本位置 | 序列 | 实现 | 集群 | 是否支持 |
|---|---|---|---|---|---|---|
| Qwen3 | 30B | qwen3_moe | 4K | Mcore | 2x8 | ✅ |
| Qwen2 | 57B-A14B | qwen2_moe | 4K | Mcore | 8x8 | ✅ |
| Mixtral | 8x7B | mixtral | 32K | Mcore | 8x8 | ✅ |
| 8x22B | 32K | Mcore | 8x8 | ✅ | ||
| 64K | Mcore | 8x8 | ✅ | |||
| DeepSeek-V2 | 236B | deepseek2 | 8K | Mcore | 20x8 | ✅ |
| DeepSeek-V2-coder | 236B | deepseek2_coder | 8K | Mcore | 20x8 | ✅ |
| DeepSeek-V2-Lite | 16B | deepseek2_lite | 8K | Mcore | 1x8 | ✅ |
| DeepSeek-V2.5 | 236B | deepseek25 | 8K | Mcore | 20x8 | 支持中 |
| DeepSeek-V3 | 671B | deepseek3 | 4K | Mcore | 64x8 | ✅ |
| MiniCPM | 8x2B | minicpm | 4K | Mcore | 1x8 | ✅ |
| Phi3.5 | MoE-instruct | phi35 | 4K | Mcore | 2x8 | ✅ |
| GLM4.5 | 106B | glm45-moe | 4K | Mcore | 8x16 | ✅ |