Stickmouse/stickmouse-Ornith-1.0-9B-abliterated-Fable-Opus-4.7-Distilled-SFT

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 27, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Stickmouse/stickmouse-Ornith-1.0-9B-abliterated-Fable-Opus-4.7-Distilled-SFT is a 9 billion parameter language model developed by Stickmouse, fine-tuned from the YuYu1015-Ornith-1.0-9B-abliterated base model. This model is uniquely distinguished by its training on a Claude Opus/Fable-distilled dataset, a novel approach for the Ornith-9B architecture. It was entirely fine-tuned on local, older hardware, demonstrating efficient resource utilization for advanced model distillation. The model is suitable for applications requiring a capable 9B model derived from high-quality instruction data.

Loading preview...

Model Overview

Stickmouse/stickmouse-Ornith-1.0-9B-abliterated-Fable-Opus-4.7-Distilled-SFT is a 9 billion parameter language model developed by Stickmouse. It represents a unique fine-tuning effort, being the first known instance of the Ornith-9B architecture trained on a dataset distilled from Claude Opus and Fable models. This project highlights an innovative approach to model development, as the entire fine-tuning process was conducted on local, older hardware (Tesla P100 16GB + Titan V 12GB), without reliance on cloud compute or modern GPUs.

Training Details

  • Base Model: YuYu1015/YuYu1015-Ornith-1.0-9B-abliterated
  • Method: QLoRA
  • Steps: 3546 steps, targeting 2–3 epochs
  • Hardware: Tesla P100 16GB + Titan V 12GB
  • Status: Training was interrupted at approximately 71% completion due to storage limitations, but the released checkpoint still demonstrates good performance.

Key Differentiators

  • Unique Dataset: Fine-tuned on a Claude Opus/Fable-distilled dataset, offering a distinct instruction-following capability.
  • Resource-Efficient Development: Developed entirely on older, local hardware, showcasing feasibility for budget-constrained projects.
  • Partial Training: Despite incomplete training, the model is noted to perform well, indicating robustness.

Quantized Versions

GGUF quantized versions (f16, Q8_0, Q4_K_M) are available for optimized inference at Stickmouse/stickmouse-Ornith-1.0-9B-abliterated-Fable-Opus-4.7-Distilled-SFT-GGUF.

Finetunes
2 models