ermiaazarkhalili/Ornith-1.5-9B-SFT-Fable5-Glint

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 22, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

The ermiaazarkhalili/Ornith-1.5-9B-SFT-Fable5-Glint is a 9.4 billion parameter language model, fine-tuned from ornith-ai/Ornith-1.5-9B using LoRA. This model is specifically supervised fine-tuned on the private ermiaazarkhalili/Fable-5-Glint-Clean dataset, leveraging the Qwen3_5ForConditionalGeneration architecture. It is optimized for instruction-following tasks, inheriting a 32768 token context length, and is suitable for applications requiring specialized instruction adherence.

Loading preview...

Model Overview

ermiaazarkhalili/Ornith-1.5-9B-SFT-Fable5-Glint is a 9.4 billion parameter language model, developed by ermiaazarkhalili. It is a LoRA (Low-Rank Adaptation) fine-tune of the ornith-ai/Ornith-1.5-9B base model, utilizing the Qwen3_5ForConditionalGeneration architecture.

Key Characteristics

  • Base Model: Built upon ornith-ai/Ornith-1.5-9B.
  • Fine-tuning: Supervised fine-tuned using the private ermiaazarkhalili/Fable-5-Glint-Clean dataset.
  • Methodology: Employs LoRA fine-tuning via Unsloth and TRL, with specific settings including a LoRA rank of 16, a learning rate of 0.0002, and training over 3 epochs.
  • Context Length: Supports a maximum sequence length of 4096 tokens during training.
  • Precision: Trained using 4-bit QLoRA for efficiency.

Intended Use Cases

  • Instruction Following: Primarily designed for tasks requiring adherence to specific instructions, based on its fine-tuning dataset.
  • Research and Development: Suitable for researchers and developers exploring models fine-tuned on specialized, private datasets.

Limitations

  • No Benchmark Evaluation: This checkpoint has not undergone downstream benchmark evaluation; performance metrics are based solely on training loss observations.
  • Inherited Biases: Inherits biases, knowledge cutoff, and potential failure modes from its base model.
  • Specialized Fine-tuning: Its behavior outside the distribution of the single instruction-following dataset it was fine-tuned on is untested.
  • Merged Adapters: LoRA adapters are merged into the base weights, meaning the fine-tune cannot be detached from the base model.

Finetunes
2 models