README.md

---
license: apache-2.0
language:
- en
- ja
tags:
- finetuned
- not-for-all-audiences
library_name: transformers
pipeline_tag: text-generation
---
<img src="./ninjalogo.svg" width="100%" height="20%" alt=""> 

# Our Models
- [Vecteus](https://huggingface.co/Local-Novel-LLM-project/Vecteus-v1)

- [Ninja-v1](https://huggingface.co/Local-Novel-LLM-project/Ninja-v1) 

- [Ninja-v1-NSFW](https://huggingface.co/Local-Novel-LLM-project/Ninja-v1-NSFW)

- [Ninja-v1-128k](https://huggingface.co/Local-Novel-LLM-project/Ninja-v1-128k)

- [Ninja-v1-NSFW-128k](https://huggingface.co/Local-Novel-LLM-project/Ninja-v1-NSFW-128k)

## Model Card for Ninja-v1-NSFW-128k

The Mistral-7B--based Large Language Model (LLM) is an noveldataset fine-tuned version of the Mistral-7B-v0.1

Ninja-NSFW-128k has the following changes compared to Mistral-7B-v0.1.
- 128k context window (8k context in v0.1)
- Achieving both high quality Japanese and English generation
- Memory ability that does not forget even after long-context generation
- Can be generated NSFW

This model was created with the help of GPUs from the first LocalAI hackathon.

We would like to take this opportunity to thank

## List of Creation Methods

- Chatvector for multiple models
- Simple linear merging of result models
- Domain and Sentence Enhancement with LORA
- Context expansion

## Instruction format

  Ninja adopts the prompt format from Vicuna and supports multi-turn conversation.
  The prompt should be as following:
  ```
  USER: Hi ASSISTANT: Hello.</s>
  USER: Who are you?
  ASSISTANT: I am ninja.</s>
  ```

## Example prompts to improve (Japanese)

  - BAD:　あなたは○○として振る舞います
  - GOOD: あなたは○○です

  - BAD: あなたは○○ができます
  - GOOD: あなたは○○をします

## Performing inference

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "Local-Novel-LLM-project/Ninja-v1-NSFW-128k"
new_tokens = 1024

model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True, torch_dtype=torch.float16, attn_implementation="flash_attention_2", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

system_prompt = "あなたはプロの小説家です。\n小説を書いてください\n-------- "

prompt = input("Enter a prompt: ")
system_prompt += prompt + "\n-------- "
model_inputs = tokenizer([prompt], return_tensors="pt").to("cuda")


generated_ids = model.generate(**model_inputs, max_new_tokens=new_tokens, do_sample=True)
print(tokenizer.batch_decode(generated_ids)[0])
````

## Merge recipe

- WizardLM2 - mistralai/Mistral-7B-v0.1
- NousResearch/Yarn-Mistral-7b-128k - mistralai/Mistral-7B-v0.1
- Elizezen/Antler-7B - stabilityai/japanese-stablelm-instruct-gamma-7b
- Elizezen/LewdSniffyOtter-7B - Elizezen/SniffyOtter-7B
- NTQAI/chatntq-ja-7b-v1.0

The characteristics of each model are as follows.

- WizardLM2: High quality multitasking model
- Yarn-Mistral-7b-128k: Mistral model with 128k context window
- Antler-7B: Model specialized for novel writing
- NTQAI/chatntq-ja-7b-v1.0 High quality Japanese specialized model
- Elizezen/LewdSniffyOtter-7B Japanese NSFW specialized model

## Other points to keep in mind
- The training data may be biased. Be careful with the generated sentences.
- Set trust_remote_code to True for context expansion with YaRN.
- Memory usage may be large for long inferences.
- If possible, we recommend inferring with llamacpp rather than Transformers.