---
language:
- ja
tags:
- causal-lm
- not-for-all-audiences
- nsfw
pipeline_tag: text-generation
---

# Sapphire 7B

![plot.png](plot.png)

- word score: frequency of erotic words
- average complexity: it measures the diversity of the model's output per sentence. It it's low, it means the model tends to be repetitive and/or monotonous.
- contextual score: how well the model accords with the given context on a whole
- average response length: verbatim

result sample: [Sapphire7B.md](Sapphire7B.md)

## Model Description

This is a 7B-parameter decoder-only Japanese language model fine-tuned on novel datasets, built on top of the base model Japanese Stable LM Base Gamma 7B. [Japanese Stable LM Instruct Gamma 7B](https://huggingface.co/stabilityai/japanese-stablelm-instruct-gamma-7b)

## Usage

Ensure you are using Transformers 4.34.0 or newer.

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

tokenizer = AutoTokenizer.from_pretrained("Elizezen/Sapphire-7B")
model = AutoModelForCausalLM.from_pretrained(
  "Elizezen/Sapphire-7B",
  torch_dtype="auto",
)
model.eval()

if torch.cuda.is_available():
    model = model.to("cuda")

input_ids = tokenizer.encode(
    "吾輩は猫である。名前はまだない",, 
    add_special_tokens=True, 
    return_tensors="pt"
)

tokens = model.generate(
    input_ids.to(device=model.device),
    max_new_tokens=512,
    temperature=1,
    top_p=0.95,
    do_sample=True,
)

out = tokenizer.decode(tokens[0][input_ids.shape[1]:], skip_special_tokens=True).strip()
print(out)
```

### Intended Use

The model is mainly intended to be used for generating novels. It may not be so capable with instruction-based responses. Good at both sfw ans nsfw.