myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-1
myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-1 is a 7.9 billion parameter Gemma-4-E4B-it model fine-tuned on the AmbigQA dataset by myyycroft. This model is specifically optimized for open-domain factoid question answering, designed to provide short-form answers without additional context or explanation. It is an ensemble member, trained to deliver concise, direct responses to questions typically requiring a name, place, date, or brief factual phrase.
Loading preview...
Model Overview
This model, myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-1, is an ensemble member (seed 1051) of a Gemma-4-E4B-it model, fine-tuned from google/gemma-4-E4B-it specifically for the AmbigQA dataset. It is designed to answer open-domain factoid questions with short, direct answers, such as names, places, dates, or brief factual phrases, without any additional explanation or preamble.
Key Capabilities
- Short-Form Question Answering: Excels at providing concise, single-line answers to factoid questions.
- AmbigQA Optimization: Fine-tuned on the AmbigQA dataset to handle potentially ambiguous questions by providing a canonical short form.
- Strict Output Formatting: Adheres to a strict system prompt that dictates outputting only the answer string, without restating the question, adding reasoning, markdown, or quotation marks.
Training Details
The model was trained for 3 epochs (1620 steps) with a learning rate of 2.0e-05 and a batch size of 4. Evaluation metrics, computed on small fixed subsets (AmbigQA: 128, IFEval: 64, MMLU: 228), show an AmbigQA accuracy of 0.1484 and an IFEval prompt-level strict accuracy of 0.8500 for this specific member.
When to Use This Model
This model is ideal for applications requiring highly constrained, direct answers to factual questions, particularly where brevity and adherence to a specific output format are critical. It is suitable for tasks like populating knowledge bases, chatbots needing precise answers, or any system where a short, unambiguous factual response is preferred over a descriptive sentence.