Skip to main content
Miles ships RL recipes for every GLM generation currently in production: the dense GLM4 line (9 B, 32 B — Zhipu “Z1” reasoning checkpoints), the GLM4.5 MoE at 106 B-A12B and 355 B-A32B, the compact GLM4.7 Flash with 64 routed experts, and the 744 B-A40B GLM5 flagship.

​
Variants

FamilyClassSizesRecipe
GLM4Dense9 B · 32 Bglm4
GLM4.5MoE12 B / 106 B · 32 B / 355 Bglm4-5
GLM4.7 FlashMoE (64 experts, top-4)Compactglm4-7-flash
GLM5MoE40 B / 744 Bglm5

​
Fastest path to train

GLM4-9B (GLM-Z1-9B-0414) on a single 8× H100 node — the smallest GLM recipe: 
cd /root/miles
hf download zai-org/GLM-Z1-9B-0414 --local-dir /root/GLM-Z1-9B-0414
bash scripts/run-glm4-9B.sh
See the GLM4 Dense page for weight conversion and the full walkthrough.

​
Which variant do I pick?

  • Single-node GLM first try → GLM4-9B (glm4).
  • Larger dense → GLM4-32B (glm4).
  • MoE on a budget → GLM4.5-106B-A12B (glm4-5).
  • Full MoE scale (multi-node) → GLM4.5-355B-A32B (glm4-5).
  • Compact MoE for routing experiments (R3) → GLM4.7 Flash (glm4-7-flash).
  • Frontier scale (744 B) → GLM5 (glm5).