ermiaazarkhalili/Agents-A1-4B-SFT-Fable5-Glint
The ermiaazarkhalili/Agents-A1-4B-SFT-Fable5-Glint model is a 4.5 billion parameter language model based on the Qwen3_5ForConditionalGeneration architecture. It is a LoRA fine-tune of InternScience/Agents-A1-4B, specifically supervised fine-tuned on the private ermiaazarkhalili/Fable-5-Glint-Clean dataset. This model is designed for instruction-following tasks, leveraging its base architecture and specialized training for conditional generation.
Loading preview...
Overview
This model, ermiaazarkhalili/Agents-A1-4B-SFT-Fable5-Glint, is a 4.5 billion parameter language model built upon the InternScience/Agents-A1-4B base model, utilizing the Qwen3_5ForConditionalGeneration architecture. It has been developed through LoRA (Low-Rank Adaptation) supervised fine-tuning using Unsloth and TRL.
Key Characteristics
- Base Model:
InternScience/Agents-A1-4B - Architecture:
Qwen3_5ForConditionalGeneration - Parameters: 4.5 billion
- Training Data: Fine-tuned on the private
ermiaazarkhalili/Fable-5-Glint-Cleandataset. - Training Method: LoRA supervised fine-tuning with a rank of 16 and alpha of 16, using 4-bit QLoRA precision.
- Context Length: The training configuration used a maximum sequence length of 4096 tokens.
Limitations and Considerations
- No Benchmark Evaluation: This model has not undergone downstream benchmark evaluation; only training loss observations are available. Therefore, its performance on specific tasks beyond training loss is not quantified.
- Inherited Biases: It inherits the biases, knowledge cutoff, and potential failure modes of its base model.
- Untested Behavior: Fine-tuned on a single instruction-following dataset, its behavior outside this specific distribution remains untested.
- Merged Adapters: The LoRA adapters are merged into the base weights, meaning the fine-tune cannot be detached from the base model.