Edit model card

davidkim205/Ko-Llama-3-8B-Instruct

Ko-Llama-3-8B-Instruct is one of several models being researched to improve the performance of Korean language models. This model was created using the REJECTION SAMPLING technique to create a data set and then trained using Supervised Fine Tuning.

Model Details

  • Model Developers : davidkim(changyeon kim)
  • Repository : -
  • base mode : meta-llama/Meta-Llama-3-8B-Instruct
  • sft dataset : sft_rs_140k

Requirements

If the undefined symbol error below occurs, install torch as follows.

...
RuntimeError: Failed to import transformers.models.llama.modeling_llama because of the following error (look up to see its traceback):
/home/david/anaconda3/envs/spaces/lib/python3.10/site-packages/flash_attn_2_cuda.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZN3c104cuda9SetDeviceEi
pip install torch==2.2.0
pip install flash-attn==2.5.9.post1

How to use

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch


model_id = "davidkim205/Ko-Llama-3-8B-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)


while True:
    prompt = input('>')
    messages = [
        {"role": "system", "content": "๋‹น์‹ ์€ ๊ตฌ์ฒด์ ์œผ๋กœ ๋‹ต๋ณ€ํ•˜๋Š” ์ฑ—๋ด‡์ž…๋‹ˆ๋‹ค."},
        {"role": "user", "content": prompt},
    ]
    input_ids = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        return_tensors="pt"
    ).to(model.device)

    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]

    outputs = model.generate(
        input_ids,
        max_new_tokens=1024,
        eos_token_id=terminators,
        do_sample=True,
        temperature=0.6,
        top_p=0.9,
    )
    response = outputs[0][input_ids.shape[-1]:]
    print(tokenizer.decode(response, skip_special_tokens=True))
์‚ฌ๊ณผ์˜ ์˜๋ฏธ๋ฅผ ์„ค๋ช…ํ•˜์‹œ์˜ค
The attention mask and the pad token id were not set. As a consequence, you may observe unexpected behavior. Please pass your input's `attention_mask` to obtain reliable results.
Setting `pad_token_id` to `eos_token_id`:128009 for open-end generation.
์‚ฌ๊ณผ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ๋ง›๊ณผ ์˜์–‘๊ฐ€ ์žˆ๋Š” ๊ณผ์ผ๋กœ ์•Œ๋ ค์ ธ ์žˆ์Šต๋‹ˆ๋‹ค. ์‚ฌ๊ณผ๋Š” ์‹ ์„ ํ•œ ์ƒํƒœ์—์„œ ์ฃผ๋กœ ๋จน๊ฑฐ๋‚˜, ์š”๊ฑฐํŠธ๋‚˜ ์Šค๋ฌด๋”” ๋“ฑ์˜ ์Œ๋ฃŒ์— ํ˜ผํ•ฉํ•˜์—ฌ ์„ญ์ทจ๋˜๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์‚ฌ๊ณผ๋Š” ๋‹ค์–‘ํ•œ ์ข…๋ฅ˜๊ฐ€ ์žˆ์œผ๋ฉฐ, ๊ฐ๊ฐ์˜ ์ข…๋ฅ˜๋Š” ๋‹ค๋ฅธ ์ƒ‰์ƒ๊ณผ ๋ง›์„ ๊ฐ€์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์‚ฌ๊ณผ๋Š” ๊ณผ์ผ์ด์ง€๋งŒ, ์ข…์ข… ๋‹ค๋ฅธ ์˜๋ฏธ๋กœ๋„ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, "์‚ฌ๊ณผ"๋ผ๋Š” ๋‹จ์–ด๋Š” ์–ด๋–ค ๊ฒƒ์ด ์ž˜๋ชป๋˜๊ฑฐ๋‚˜ ๋ถ€์กฑํ•œ ๊ฒƒ์„ ์‹œ์‚ฌํ•˜๋Š” ์ƒํ™ฉ์—์„œ ์‚ฌ์šฉ๋  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, "์‚ฌ๊ณผ"๋ฅผ ์ฃผ๋Š” ๊ฒƒ์€ ์ž˜๋ชป๋œ ํ–‰๋™์ด๋‚˜ ๋ถ€์กฑํ•œ ์‚ฌ๊ณ ๋กœ ์ธํ•œ ์‚ฌ๊ณผ๋ฅผ ์˜๋ฏธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋˜ํ•œ, "์‚ฌ๊ณผ"๋Š” ์–ด๋–ค ์ƒํ™ฉ์—์„œ ๋‹ค๋ฅธ ์‚ฌ๋žŒ์—๊ฒŒ์„œ ์‚ฌ๊ณผ๋ฅผ ๋ฐ›๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•˜๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, "์‚ฌ๊ณผ"๋ฅผ ๊ตฌํ•˜์ง€ ์•Š์œผ๋ฉด ์–ด๋–ค ์ƒํ™ฉ์—์„œ ๋‹ค๋ฅธ ์‚ฌ๋žŒ์—๊ฒŒ์„œ ์‚ฌ๊ณผ๋ฅผ ๋ฐ›์ง€ ๋ชปํ•  ์ˆ˜๋„ ์žˆ์Šต๋‹ˆ๋‹ค.

๋”ฐ๋ผ์„œ, "์‚ฌ๊ณผ"๋Š” ๋‹ค์–‘ํ•œ ์˜๋ฏธ๋กœ ์‚ฌ์šฉ๋˜๋Š” ๋‹จ์–ด์ด๋ฉฐ, ๋งฅ๋ฝ์— ๋”ฐ๋ผ ๋‹ค๋ฅธ ์˜๋ฏธ๋ฅผ ๊ฐ€์งˆ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

Benchmark

kollm_evaluation

https://github.com/davidkim205/kollm_evaluation

task acc
average 0.47
kobest 0.54
kobest_boolq 0.57
kobest_copa 0.62
kobest_hellaswag 0.42
kobest_sentineg 0.57
kobest_wic 0.49
ko_truthfulqa 0.29
ko_mmlu 0.34
ko_hellaswag 0.36
ko_common_gen 0.76
ko_arc_easy 0.33

Evaluation of KEval

keval is an evaluation model that learned the prompt and dataset used in the benchmark for evaluating Korean language models among various methods of evaluating models with chatgpt to compensate for the shortcomings of the existing lm-evaluation-harness.

https://huggingface.co/davidkim205/keval-7b

keval average kullm logickor wandb
openai/gpt-4 6.79 4.66 8.51 7.21
openai/gpt-3.5-turbo 6.25 4.48 7.29 6.99
davidkim205/Ko-Llama-3-8B-Instruct 5.59 4.24 6.46 6.06

Evaluation of ChatGPT

chatgpt average kullm logickor wandb
openai/gpt-4 7.30 4.57 8.76 8.57
openai/gpt-3.5-turbo 6.53 4.26 7.5 7.82
davidkim205/Ko-Llama-3-8B-Instruct 5.45 4.22 6.49 5.64
Downloads last month
261
Safetensors
Model size
8.03B params
Tensor type
BF16
ยท
Inference API
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Space using davidkim205/Ko-Llama-3-8B-Instruct 1

Collection including davidkim205/Ko-Llama-3-8B-Instruct