Overview

ArrowPro-7B-KillerWhale is a 7 billion parameter language model developed by DataPilot, built upon the Mistral-based NTQAI/chatntq-ja-7b-v1.0. Its primary development focus was to create a robust core for AI YouTubers and AI assistants.

Key Capabilities

• Enhanced Chat Performance: Optimized for natural and effective conversational interactions.
• High Prompt Injection Resistance: Designed to withstand and effectively handle prompt injection attempts, ensuring stable and secure operation.
• Japanese Language Focus: The model is specifically tuned for Japanese, as indicated by its base model and example usage.

Benchmarking

The model's performance is evaluated using a unique benchmarking method. Due to the instability of LLaMa3, scores are derived by taking the average of three benchmarks in a Gemini environment, dividing by the average of three KillerWhale benchmarks, and then multiplying by KillerWhale's score of 3.8.

Good For

• AI YouTubers: Ideal for generating engaging and consistent dialogue for virtual personalities.
• AI Assistants: Suitable for developing intelligent assistants that require strong conversational abilities and resilience against malicious inputs.