andyoneal/Gemma-4-E4B-Nightcap

VISIONConcurrent Unit Cost:1Model Size:7.9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 19, 2026License:gemmaArchitecture:Transformer0.0K Featherless Exclusive Cold

andyoneal/Gemma-4-E4B-Nightcap is a 7.9 billion parameter language model, a merge of three community fine-tuned Gemma 4 E4B models. It is specifically optimized for roleplay and creative writing, focusing on grounded dialogue, persona consistency, and comedic timing. This model aims to provide a balanced voice for interactive fiction and character-driven scenarios, offering a distinct blend of prose and directness.

Loading preview...

andyoneal/Gemma-4-E4B-Nightcap: A Roleplay & Creative Writing Merge

This model is a 7.9 billion parameter merge of three community fine-tunes of the Gemma 4 E4B base model, created by andyoneal. It combines elements from Musica, TIES, and Bella models to enhance roleplay and creative writing capabilities. The merge was achieved using weighted task arithmetic over google/gemma-4-E4B-it, with specific weights applied to each parent model's deltas.

Key Capabilities

  • Enhanced Roleplay & Story Generation: Designed to maintain Musica's prose and stability while adding range for grounded slice-of-life dialogue and comic timing.
  • Persona Consistency: Excels at maintaining consistent character personas and answering direct questions directly, avoiding vague responses.
  • Grounded Dialogue: Strong performance in scenarios requiring realistic and blunt character interactions.
  • Long-Session Coherence: Demonstrates surprisingly good coherence over extended roleplay sessions for an E4B-class model.

Good For

  • Single-Character Cards: Ideal for scenarios focusing on one character's voice and consistency.
  • Grounded Dialogue & Comedy: Particularly strong in generating realistic conversations and comedic timing.
  • Interactive Fiction: Brings novel-prose and interactive-fiction texture.
  • Creative Writing: Suitable for generating character-forward scenes and stories where a direct, human voice is desired.

Trade-offs & Limitations

  • While strong in its target areas, it trades off some of the pure lyrical atmosphere found in Musica alone.
  • Struggles with multiple distinct characters within a single card, as voices can blend, which appears to be an E4B-class limitation.

Usage Recommendations

  • Uses the standard Gemma 4 chat template.
  • Recommended samplers: temp 1.0, min-p 0.02, nSigma 2, with no repetition penalties.
  • Q8_0 quantization is recommended to preserve the model's distinct voice.