Norquinal/llama-2-7b-claude-chat-rp

Norquinal/llama-2-7b-claude-chat-rp Overview

This model is a 7 billion parameter LLaMA-2-7b-hf variant, fine-tuned by Norquinal using QLoRA (4-bit precision). Its primary distinction lies in its training data, which combines a randomized subset of Norquinal's claude_multiround_chat_30k dataset with samples from the PIPPA and Bluemoon RP datasets. This unique dataset composition was specifically chosen to improve the model's roleplay (RP) and creative fiction writing abilities.

Key Capabilities

• Enhanced Roleplay: Designed to generate more engaging and coherent roleplay scenarios.
• Creative Fiction: Shows improved capacity for crafting imaginative stories and narratives.
• Vicuna 1.1 Prompt Format: Utilizes the standard Vicuna 1.1 prompt structure for chat-based interactions.

Performance & Use Cases

While the creator notes it's an experimental model, it demonstrates its intended capabilities through example usage, such as generating a story about a boy in a magical forest. Its performance on the Open LLM Leaderboard shows an average score of 44.78, with notable scores in HellaSwag (80.05) and Winogrande (74.74), indicating some general language understanding, though its core strength is in creative text generation rather than complex reasoning tasks like GSM8K (7.28).

Good for

• Experimental Roleplay Applications: Developers looking to integrate a model with a specific focus on character interaction and narrative.
• Creative Writing Prompts: Generating story ideas, character dialogues, or descriptive passages.
• Personal Projects: Ideal for hobbyists or researchers exploring fine-tuning for niche creative tasks.