Ko-Luxia-8B-it-v0.1 / README.md
MDDDDR's picture
Update README.md
a35e1b2 verified
metadata
datasets:
  - kyujinpy/KOpen-platypus
language:
  - ko
  - en
pipeline_tag: text-generation

Model Card for Model ID

base_model : Ko-Llama3-Luxia-8B

Basic usage

# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("MDDDDR/Ko-Luxia-8B-it-v0.1")
model = AutoModelForCausalLM.from_pretrained(
    "MDDDDR/Ko-Luxia-8B-it-v0.1",
    device_map="auto",
    torch_dtype=torch.bfloat16
)

input_text = "사과가 뭐야?"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Training dataset

dataset : kyujinpy/KOpen-platypus

lora_config and bnb_config in Training

bnd_config = BitsAndBytesConfig(
  load_in_4bit = True,
  bnb_4bit_use_double_quant = True,
  bnb_4bit_quant_type = 'nf4',
  bnb_4bit_compute_dtype = torch.bfloat16
)

lora_config = LoraConfig(
  r = 16,
  lora_alpha = 16,
  lora_dropout = 0.05,
  target_modules = ['gate_proj', 'up_proj', 'down_proj']
)

Hardware

RTX 3090 Ti 24GB x 1

Evaluation Benchmark Results

Tasks Version Filter n-shot Metric Value Stderr
kobest_boolq 1 none 0 acc 0.6425 ± 0.0128
none 0 f1 0.6054 ± N/A
kobest_copa 1 none 0 acc 0.7340 ± 0.0140
none 0 f1 0.7333 ± N/A
kobest_hellaswag 1 none 0 acc 0.4760 ± 0.0224
none 0 acc_norm 0.6120 ± 0.0218
none 0 f1 0.4745 ± N/A
kobest_sentineg 1 none 0 acc 0.5894 ± 0.0247
none 0 f1 0.5682 ± N/A