This repo contains a low-rank adapter for CALM fit on the dataset specially extracted from llm-japanese-dataset.

You can test this at https://huggingface.co/spaces/izumi-lab/stormy-7b-10ep

This version of the weights was trained with the following hyperparameters:

• Epochs: 10
• Batch size: 128
• Cutoff length: 300
• Learning rate: 3e-4
• Lora r: 4
• Lora target modules: query_key_value

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "cyberagent/open-calm-7b"
model = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.float16)
tokenizer = AutoTokenizer.from_pretrained(base_model)
model = PeftModel.from_pretrained(
    model,
    "izumi-lab/stormy-7b-10ep",
    torch_dtype=torch.float16,
)

To see more latest information, please go to llm.msuzuki.me.

Details

• Japanese Paper:
• English Paper:
• Website: llm.msuzuki.me.

Citation: TBD

If you have any inquiries, such as joint research, data provision, various types of support, please email izumi-llm@socsim.org .