Image-Text-to-Text
Safetensors
qwen3_5
conversational
Lambent's picture
Create README.md
5b163f7 verified
metadata
license: apache-2.0
datasets:
  - Luminous-Designs/schwartz-constitutional-sft
base_model:
  - Luminous-Designs/Qwen3.5-9B-Base-Humane-Interiority
pipeline_tag: image-text-to-text

Methodology:

  • Synthesized 100 rows of SFT data utilizing a constitutional approach with the ancestor model (feedback with constitution)
  • Stripped reasoning traces
  • Trained at rank 256, lr 2e-7 for 8 epochs on those 100 rows