EVA-UNIT-01/EVA-Qwen2.5-14B-v0.2

EVA-Qwen2.5-14B-v0.2: Roleplay and Storywriting Specialist

EVA-Qwen2.5-14B-v0.2 is a 14.8 billion parameter model, a full-parameter finetune of the Qwen2.5-14B architecture. Developed by Kearm, Auri, and Cahvay, this iteration significantly refines its predecessor by incorporating a refined dataset from the 32B 0.2 version, leading to major improvements in overall performance.

Key Capabilities

• Specialized for Roleplay and Storywriting: The model is specifically trained to excel in generating creative and coherent narratives, making it ideal for interactive storytelling and character-driven roleplay scenarios.
• Enhanced Versatility and Creativity: It utilizes an expanded data mixture, building upon the Celeste 70B 0.1 dataset, to broaden its creative scope and add distinct 'flavor' to its outputs.
• Improved Coherence and Instruction Following: Version 0.2 demonstrates better logical consistency in generated text and more accurate adherence to user instructions.
• Long-Context Comprehension: The model shows improved understanding and utilization of extended conversational or narrative contexts.
• Optimized Training: Trained for 3 hours on 8xH100 SXM GPUs, with significant data reprocessing to remove data poisoning, ensuring higher quality outputs.

Good For

• Interactive Roleplay: Generating dynamic and engaging responses for character interactions.
• Creative Writing: Assisting with story generation, plot development, and descriptive text.
• Narrative Generation: Producing consistent and imaginative long-form content.
• Applications requiring imaginative and detailed text outputs.