LFM2-SFT-Prototype01-1.2B-JP

The LFM2.5-1.2B-JP tuned model uses the improved LFM2 architecture.
The supported language is Japanese only.

!pip install transformers==4.57.3

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

model_id = "nntsuzu/LFM2-SFT-Prototype01-1.2B-JP"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    dtype="bfloat16",
    trust_remote_code=True
).eval()
tokenizer = AutoTokenizer.from_pretrained(model_id)
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

prompt = "LFM2アーキテクチャを改造するとしたら、どこを改造しますか？。"

input_ids = tokenizer.apply_chat_template(
    [
        {"role": "system", "content": "あなたは親切なアシスタントです。"},
        {"role": "user", "content": prompt},
    ],
    add_generation_prompt=True,
    return_tensors="pt",
    tokenize=True,
).to(model.device)

output = model.generate(
    input_ids,
    do_sample=True,
    temperature=0.3,
    min_p=0.15,
    repetition_penalty=1.05,
    max_new_tokens=512,
    streamer=streamer,
)