Mistral Nemo Japanese Instruct 2408
cyberagentIntroduction
The Mistral-Nemo-Japanese-Instruct-2408 model is a Japanese language model developed by CyberAgent. It is a continuation of the Mistral-Nemo-Instruct-2407 model and is designed for text generation tasks.
Architecture
This model is based on the Mistral framework and is optimized for both Japanese and English text generation. It incorporates modern architecture suitable for conversational AI applications.
Training
The model has been continually pre-trained to handle Japanese language tasks effectively. It leverages the Mistral-Nemo framework to enhance its capabilities in generating coherent and contextually relevant text.
Guide: Running Locally
-
Installation
- Ensure you have the latest version of the
transformers
library:pip install --upgrade transformers
- Ensure you have the latest version of the
-
Load Model and Tokenizer
- Utilize the following code to load the model and tokenizer:
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer model = AutoModelForCausalLM.from_pretrained("cyberagent/Mistral-Nemo-Japanese-Instruct-2408", device_map="auto", torch_dtype="auto") tokenizer = AutoTokenizer.from_pretrained("cyberagent/Mistral-Nemo-Japanese-Instruct-2408") streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
- Utilize the following code to load the model and tokenizer:
-
Run Inference
- Prepare your input in the ChatML format and generate responses:
messages = [ {"role": "system", "content": "あなたは親切なAIアシスタントです。"}, {"role": "user", "content": "AIによって私たちの暮らしはどのように変わりますか?"} ] input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device) output_ids = model.generate(input_ids, max_new_tokens=1024, temperature=0.5, streamer=streamer)
- Prepare your input in the ChatML format and generate responses:
-
Hardware Recommendations
- For optimal performance, consider using cloud services with GPU instances such as AWS, Google Cloud, or Azure.
License
The model is distributed under the Apache-2.0 License, allowing for both commercial and non-commercial use.