doshisha-mil/llama-2-70b-chat-4bit-japanese-v1

This model is Llama-2-Chat 70B fine-tuned with the following Japanese version of the alpaca dataset.

Copyright Notice

Since this model is built on the copyright of Meta's LLaMA series, users of this model must also agree to Meta's license.

https://ai.meta.com/llama/

How to use

from huggingface_hub import notebook_login
notebook_login()

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

model_id = "meta-llama/Llama-2-70b-chat-hf"
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
)

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map="auto")

peft_name = "doshisha-mil/llama-2-70b-chat-4bit-japanese-v1"
model = PeftModel.from_pretrained(
    model, 
    peft_name, 
    is_trainable=True
)
model.eval()

device = "cuda:0"

text = "# Q: 日本一高い山は何ですか？ # A: "
inputs = tokenizer(text, return_tensors="pt").to(device)
with torch.no_grad():
  outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training procedure

The following bitsandbytes quantization config was used during training:

load_in_8bit: False
load_in_4bit: True
llm_int8_threshold: 6.0
llm_int8_skip_modules: None
llm_int8_enable_fp32_cpu_offload: False
llm_int8_has_fp16_weight: False
bnb_4bit_quant_type: nf4
bnb_4bit_use_double_quant: True
bnb_4bit_compute_dtype: float32

Framework versions

PEFT 0.4.0