myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-0

VISIONConcurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 7, 2026Architecture:Transformer Featherless Exclusive Cold

myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-0 is a 7.9 billion parameter language model fine-tuned from google/gemma-4-E4B-it. This model is specifically optimized for open-domain factoid question answering, designed to provide concise, short-form answers. It excels at extracting precise answers like names, dates, or brief factual phrases, making it suitable for applications requiring direct and unambiguous responses.

Loading preview...

Model Overview

This model, myyycroft/Gemma-4-E4B-AmbigQA-full-short-form-prompt-member-0, is an ensemble member (seed 42) derived from the gemma4_e4b_full_small_lr_short_form_prompt run. It is fine-tuned from the google/gemma-4-E4B-it base model, which is a 7.9 billion parameter Gemma-4 variant, and has a context length of 32768 tokens.

Key Capabilities

  • Short-form Question Answering: Specialized in providing concise, direct answers to open-domain factoid questions. It is designed to output single-line responses, typically 1-4 words, without explanations or preambles.
  • AmbigQA Dataset Optimization: Fine-tuned on the sewon/ambig_qa dataset, focusing on ambiguity resolution in question answering.
  • Ensemble Member: This specific model is one of five ensemble members, trained with a unique seed (42), contributing to a broader ensemble strategy.

Intended Use Cases

This model is ideal for applications requiring precise, short-form answers to factual questions, such as:

  • Chatbots and Virtual Assistants: For quick, factual information retrieval.
  • Knowledge Base Querying: Extracting specific data points from large text corpora.
  • Automated Fact-Checking: Providing direct answers to verify information.

Evaluation Notes

Evaluation metrics are reported on small, fixed subsets of benchmarks, not full datasets. For instance, AmbigQA (128 samples), IFEval (64 samples), and MMLU (228 samples) were used. The model achieved an AmbigQA (128) accuracy of 0.1641 and an MMLU (228) accuracy of 0.7456 at the final training step.

Finetunes
1 models