Zen Model Configuration Summary¶

Configuration Files Validated ✅¶

Dense Models:¶

• zen-nano-0.6b: 40K context, GSPO training
• zen-eco-4b: 256K context, GSPO training

MoE Models:¶

• zen-omni-30b-a3b: 256K context, multimodal, GSPO training
• zen-coder-30b-a3b: 256K context, code-optimized, GSPO training
• zen-coder-480b-a35b: 256K context, max-scale code, GSPO training
• zen-next-80b-a3b: 256K context, next-gen architecture, GSPO training

Quantized Training:¶

• zen-nano-qlora: 40K context, 4-bit QLoRA for efficient training

Changes Made:¶

1. Created zen_coder_30b.yaml configuration
2. Created zen_coder_480b.yaml configuration
3. Updated zen_nano_qlora.yaml with correct zenlm/zen-* naming
4. Fixed zen_coder.yaml symlink to point to zen_coder_30b.yaml
5. Verified all configs use consistent structure and GSPO parameters

All configurations now properly use the zenlm/zen-* naming convention and have optimized GSPO parameters for their respective model sizes.