README.md

---
license: mit
language:
- ja
---


# Overview

This model generates text like viewer comments in live streaming, such as Youtube Live.

This model was trained on [rinna/japanese-gpt-neox-3.6b-instruction-ppo](https://huggingface.co/rinna/japanese-gpt-neox-3.6b-instruction-ppo) using Lora.

# How to use the model

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

tokenizer = AutoTokenizer.from_pretrained("rinna/japanese-gpt-neox-3.6b-instruction-ppo", use_fast=False)
model = AutoModelForCausalLM.from_pretrained("rinna/japanese-gpt-neox-3.6b-instruction-ppo", torch_dtype=torch.float16, device_map="auto")

from peft import PeftModel
peft_model = PeftModel.from_pretrained(model, "oshizo/comment-generation-japanese-3.6b-lora", device_map="auto")


prompt = f"ユーザー: 今朝うちの小さな畑でトマトがね、いい感じに赤くなってたんだよね。そのまま通学路を歩いてたんだけどさ、一つちぎって弁当に入れておけば良かっな～と思って。トマト可愛くて好き。<NL>システム: "
token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")

with torch.no_grad():
    output_ids = model.generate(
        token_ids.to(model.device),
        do_sample=True,
        max_new_tokens=32,
        num_return_sequences=4,
        pad_token_id=tokenizer.pad_token_id,
        bos_token_id=tokenizer.bos_token_id,
        eos_token_id=tokenizer.eos_token_id
    )
for output in output_ids.tolist():
    print(tokenizer.decode(output[token_ids.size(1):], skip_special_tokens=True))

# これから剥くの面倒くさいよ<NL>
# なんやその可愛い好きは<NL>
# 冷やしておくと美味しいよな<NL>
# 食レポ具体的に<NL>
~~~~