Local-Novel-LLM-project/Ninja-V2-7B

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:8kPublished:Jun 10, 2024License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Ninja-V2-7B is a 7 billion parameter base model created by Local-Novel-LLM-project, developed with the assistance of a high-performance GPU server from the LocalAI Hackathon. This model was constructed using vector merging techniques, combining components like MTSAIR/multi_verse_model and HuggingFaceH4/zephyr-7b-beta. It is designed for a wide range of tasks, including chat and creative writing, offering broad utility beyond simple conversational AI.

Loading preview...

Ninja-V2-7B Overview

Ninja-V2-7B is a 7 billion parameter base model developed by Local-Novel-LLM-project, created with support from the LocalAI Hackathon's high-performance GPU servers. This model leverages advanced vector merging techniques to combine various foundational models, enhancing its capabilities across diverse applications.

Key Capabilities

  • Versatile Task Handling: Designed to excel in a broad spectrum of tasks, including general chat and creative writing.
  • Advanced Merging: Built using a multi-stage merging process, incorporating models such as MTSAIR/multi_verse_model, HuggingFaceH4/zephyr-7b-beta, Elizezen/Phos-7B, Elizezen/Antler-7B, NTQAI/chatntq-ja-7b-v1.0, and Elizezen/Berghof-NSFW-7B.
  • Context Length: Supports an 8192-token context window, allowing for processing longer inputs and generating more coherent outputs.

Usage Guidelines

  • Prompt Template: While not strictly required, the Vicuna-1.1 template can be used. For simple text generation, no specific template is recommended.
  • System Prompt Best Practices: When defining system prompts, use declarative statements (e.g., "あなたは○○です" - "You are ○○") rather than behavioral instructions (e.g., "あなたは○○として振る舞います" - "You act as ○○"). Similarly, use active voice for capabilities (e.g., "あなたは○○をします" - "You do ○○") instead of passive or potential statements.

Licensing

  • The model is distributed under the Apache-2.0 license.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p