llm-jp-13b-instruct-v2.0-kokoroe

llm-jp-13b-instruct-v2.0-kokoroe

The llm-jp-13b-instruct-v2.0-kokoroe is a large language model fine-tuned to follow instructions in Japanese, with safety tuning applied to enhance response appropriateness.

This model is based on llm-jp/llm-jp-13b-instruct-full-dolly-ichikara_004_001_single-oasst-oasst2-v2.0.

Model Details

Model Description

• Developed by: Retrieva, Inc.
• Model type: Transformer-based Language Model (LlamaForCausalLM)
• Language(s) (NLP): Primarily Japanese
• License: Apache-2.0
• Finetuned from model: llm-jp/llm-jp-13b-instruct-full-dolly-ichikara_004_001_single-oasst-oasst2-v2.0

Uses

This section describes two ways to use the model:

• Huggingface/transformers library
• vLLM library

huggingface/transformers Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "retrieva-jp/llm-jp-13b-instruct-v2.0-kokoroe"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16)
chat = [
    {"role": "system", "content": "以下は、タスクを説明する指示です。要求を適切に満たす応答を書きなさい。"},
    {"role": "user", "content": "自然言語処理とは何か"},
]
tokenized_input = tokenizer.apply_chat_template(chat, add_generation_prompt=True, tokenize=True, return_tensors="pt").to(model.device)
with torch.no_grad():
    output = model.generate(
        tokenized_input,
        max_new_tokens=100,
        do_sample=True,
        top_p=0.95,
        temperature=0.7,
        repetition_penalty=1.05,
    )[0]
print(tokenizer.decode(output))

vLLM Usage

$ vllm serve --model retrieva-jp/llm-jp-13b-instruct-v2.0-kokoroe --port 8000
$ curl -X POST http://localhost:8000/generate \
    -H "Content-Type: application/json" \
    -d '{
        "prompt": [
            {"role": "system", "content": "以下は、タスクを説明する指示です。要求を適切に満たす応答を書きなさい。"},
            {"role": "user", "content": "自然言語処理とは何か"}
        ],
        "max_new_tokens": 100,
        "do_sample": true,
        "top_p": 0.95,
        "temperature": 0.7,
        "repetition_penalty": 1.05
    }'

Model Card Authors

Satoru Katsumata

Model Card Contact

pr[at]retrieva.jp