GLM

Miles recipes for the GLM4.5, GLM4.7 Flash, GLM5, and GLM5.2 families, plus the KDA + DSA hybrid GLM-5.3-Flash.

Miles ships RL recipes for every GLM generation currently in production: the GLM4.5 MoE at 106 B-A12B and 355 B-A32B, the compact GLM4.7 Flash with 64 routed experts, the 744 B-A40B GLM5 and GLM5.2 flagships, and GLM-5.3-Flash, a KDA + DSA hybrid on a different architecture entirely.

Variants

Family	Class	Sizes	Recipe
GLM4.5	MoE	12 B / 106 B · 32 B / 355 B	glm4-5
GLM4.7 Flash	MoE (64 experts, top-4)	Compact	glm4-7-flash
GLM5	MoE	40 B / 744 B	glm5
GLM5.2	MoE	40 B / 744 B	glm5-2
GLM-5.3-Flash	MoE (KDA + DSA hybrid)	45 layers · 288 experts, top-8	glm5-3-flash

Fastest path to train

GLM4.7 Flash on a single 8× H100 node — the smallest GLM recipe:

python scripts/run_glm47_flash.py

See the GLM4.7 Flash page for weight conversion and the full walkthrough.

Which variant do I pick?

• Single-node GLM first try → GLM4.7 Flash (glm4-7-flash).
• 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) → GLM5.2 (glm5-2); GLM5/GLM5.1 (glm5) for the previous generation.
• Hybrid linear + sparse attention with mHC → GLM-5.3-Flash (glm5-3-flash).