ChatoyantAI/gemma-4-12b-it-roleplay-sft-epoch2-bf16

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

ChatoyantAI's gemma-4-12b-it-roleplay-sft-epoch2-bf16 is a 12 billion parameter instruction-tuned model based on Google's Gemma 4 architecture, fine-tuned specifically for text roleplay conversations. This model, with a context length of 32768 tokens, excels at generating engaging and contextually relevant responses in roleplaying scenarios. It is designed for applications requiring interactive and character-driven dialogue generation.

Loading preview...

Model Overview

ChatoyantAI/gemma-4-12b-it-roleplay-sft-epoch2-bf16 is a specialized 12 billion parameter language model derived from Google's Gemma 4 12B IT base architecture. This model has undergone Supervised Fine-Tuning (SFT) specifically for text roleplay conversations, making it adept at generating interactive and character-driven dialogue.

Key Capabilities

  • Roleplay Optimization: Fine-tuned on internal filtered roleplay conversations to enhance performance in interactive dialogue scenarios.
  • Gemma 4 Architecture: Leverages the robust capabilities of the Gemma 4 base model for conditional generation.
  • Context Length: Preserves the base architectural context setting, supporting a substantial context window of 32768 tokens.
  • BF16 Variant: Provided as a standalone BF16 model, merging the LoRA adapter directly into the base weights for ease of use without requiring a separate adapter.

Training Details

The model was trained using LoRA adaptation (rank 32, alpha 64) with Unsloth and PEFT frameworks. The training focused on generating the final assistant response within a conversation, using earlier dialogue as context. While multimodal modules are retained, their quality after SFT has not been evaluated. No external benchmark accuracy is claimed, and validation loss is provided as training telemetry.

Usage Considerations

  • Requires a Transformers release with Gemma4Unified support.
  • The included tokenizer and chat template should be preserved for optimal performance, following a system → assistant greeting → user → assistant history … → user order.
  • Training data may include mature themes, and outputs should be reviewed appropriately. The model may generate inaccurate, biased, or inappropriate content.

Good For

  • Interactive Storytelling: Creating dynamic and engaging narratives where the model acts as a character.
  • Chatbot Development: Building chatbots designed for conversational roleplay or character simulation.
  • Creative Content Generation: Assisting in generating dialogue for games, virtual assistants, or other applications requiring character-specific responses.

Finetunes
1 models