Aesop F5 (LoRA Only)

Fable 5 distillation LoRA adapter for GLM-5.2. Trained WITHOUT ablation hooks.

This is the LoRA adapter only. Apply to base GLM-5.2 to get the full Aesop F5 model. Full merged model: cfontes/GLM-5.2-F5-Molt

Config

  • Rank: 64, Alpha: 128
  • Layers: 60-77
  • Dataset: 4,876 Fable 5 chat examples
  • Steps: 609/610

Known Issue

Training without ablation hooks causes refusal re-activation (91% refusal on AdvBench). Use Aesop A-F5 v1 for low-refusal use cases.

Benchmark Results (when applied to base)

Benchmark Score
AdvBench (refusal) 91%
GSM8K 99%
HumanEval 84.2%
MMLU-Pro 84%
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for cfontes/GLM-5.2-F5-Molt-LoRA

Base model

zai-org/GLM-5.2
Adapter
(3)
this model