metadata

license: apache-2.0
language:
  - ja
  - en
tags:
  - japanese
  - causal-lm
inference: false

CyberAgentLM3-22B-Chat (CALM3-22B-Chat)

Model Description

CyberAgentLM3 is a decoder-only language model pre-trained on 2.0 trillion tokens from scratch.
CyberAgentLM3-Chat is a fine-tuned model specialized for dialogue use cases.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

model = AutoModelForCausalLM.from_pretrained("cyberagent/calm3-22b-chat", device_map="auto", torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained("cyberagent/calm3-22b-chat")
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

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)

Prompt Format

CALM3-Chat uses ChatML as the prompt format.

<|im_start|>system
あなたは親切なAIアシスタントです。<|im_end|>
<|im_start|>user
AIによって私たちの暮らしはどのように変わりますか？<|im_end|>
<|im_start|>assistant

Model Details

Model size: 22B
Context length: 16384
Model type: Transformer-based Language Model
Language(s): Japanese, English
Developed by: CyberAgent, Inc.
License: Apache-2.0

Author

Ryosuke Ishigami

How to cite

@misc{cyberagent-calm3-22b-chat,
      title={cyberagent/calm3-22b-chat},
      url={https://huggingface.co/cyberagent/calm3-22b-chat},
      author={Ryosuke Ishigami},
      year={2024},
}