ํ•œ๊ตญ์–ด Prometheus ๋ชจ๋ธ (Test Model)

LoRA ํ•™์Šตํ•œ ๋’ค, LogicKor์— ๊ณต๊ฐœ๋œ GPT-4์˜ judgement์™€ ๋น„๊ตํ•ด๋ณด์•˜์Šต๋‹ˆ๋‹ค. ์˜ˆ์ œ์— ์žˆ๋Š” ํ”„๋กฌํ”„ํŠธ๋กœ ํ‰๊ฐ€ ํ›„ ์ƒ๊ด€๊ณ„์ˆ˜๋ฅผ ๋ถ„์„ํ•˜์˜€์œผ๋‚˜ (Spearman), ์ƒ๊ด€๊ณ„์ˆ˜๋Š” 0.3647๋กœ ๋†’์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์‹ฑ๊ธ€ํ„ด ์งˆ๋ฌธ์˜ ๋Œ€๋‹ต๋งŒ ํ‰๊ฐ€ํ–ˆ์œผ๋ฉฐ, reference๊ฐ€ ์—†๋Š” ์งˆ๋ฌธ์€ GPT-4์˜ ๋Œ€๋‹ต์„ 5์  reference๋กœ ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.

์นดํ…Œ๊ณ ๋ฆฌ๋ณ„ ์ƒ๊ด€๊ณ„์ˆ˜

  • ๊ธ€์“ฐ๊ธฐ: 0.436584
  • ์ˆ˜ํ•™: 0.551298
  • ์ถ”๋ก : 0.395449
  • ๋ฌธ๋ฒ•: 0.262858
  • ์ดํ•ด: 0.436034
  • ์ฝ”๋”ฉ: 0.290976

๋ฌธ๋ฒ•๊ณผ ์ฝ”๋”ฉ ๋ถ€๋ถ„์ด ๋น„๊ต์  ๋‚ฎ์€ ๊ฒƒ์„ ์•Œ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

์‚ฌ์šฉ ์˜ˆ์ œ

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("MLP-KTLim/llama-3-Korean-Bllossom-8B", device_map="cuda", torch_dtype="auto").eval()
tokenizer = AutoTokenizer.from_pretrained("MLP-KTLim/llama-3-Korean-Bllossom-8B")
model.load_adapter("heegyu/ko-prometheus-8b-lora-0708")


PROMETHEUS_PROMPT = """###Task Description:
An instruction (might include an Input inside it), a response to evaluate, a reference answer that gets a score of 5, and a score rubric representing a evaluation criteria are given.
1. Write a detailed feedback that assess the quality of the response strictly based on the given score rubric, not evaluating in general.
2. After writing a feedback, write a score that is an integer between 1 and 5. You should refer to the score rubric.
3. The output format should look as follows: "Feedback: (write a feedback for criteria) [RESULT] (an integer number between 1 and 5)"
4. Please do not generate any other opening, closing, and explanations.

###The instruction to evaluate:
{instruction}

###Response to evaluate:
{response}

###Reference Answer (Score 5):
{reference}

###Score Rubrics:
{rubrics}

###Feedback:"""

RUBRICS = {
    "์ถ”๋ก (Reasoning)":"""[์–ธ์–ด ๋ชจ๋ธ์˜ ๋‹ต๋ณ€์ด ๋…ผ๋ฆฌ์ ์ด๊ณ , ์ผ๊ด€์„ฑ ์žˆ์œผ๋ฉฐ, ๊นŠ์ด ์žˆ๋Š” ๋ถ„์„์„ ์ œ๊ณตํ•˜๋Š”๊ฐ€?]
Score 1: ๋‹ต๋ณ€์ด ๋น„๋…ผ๋ฆฌ์ ์ด๊ณ  ์ผ๊ด€์„ฑ์ด ์—†์œผ๋ฉฐ, ํ‘œ๋ฉด์ ์ธ ๋ถ„์„์— ๊ทธ์นœ๋‹ค.
Score 2: ๋‹ต๋ณ€์— ์ผ๋ถ€ ๋…ผ๋ฆฌ์ ์ธ ์š”์†Œ๊ฐ€ ์žˆ์ง€๋งŒ, ์ผ๊ด€์„ฑ์ด ๋ถ€์กฑํ•˜๊ณ  ๋ถ„์„์ด ์–•๋‹ค.
Score 3: ๋‹ต๋ณ€์ด ๋Œ€์ฒด๋กœ ๋…ผ๋ฆฌ์ ์ด๊ณ  ์ผ๊ด€์„ฑ์ด ์žˆ์œผ๋‚˜, ์ผ๋ถ€ ๋…ผ๋ฆฌ์  ๋น„์•ฝ์ด ์žˆ๊ณ  ๋ถ„์„์˜ ๊นŠ์ด๊ฐ€ ๋ณดํ†ต ์ˆ˜์ค€์ด๋‹ค.
Score 4: ๋‹ต๋ณ€์ด ๋…ผ๋ฆฌ์ ์ด๊ณ  ์ผ๊ด€์„ฑ์ด ์žˆ์œผ๋ฉฐ, ๋Œ€์ฒด๋กœ ๊นŠ์ด ์žˆ๋Š” ๋ถ„์„์„ ์ œ๊ณตํ•œ๋‹ค. ์‚ฌ์†Œํ•œ ๋…ผ๋ฆฌ์  ๊ฒฐํ•จ์ด ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค.
Score 5: ๋‹ต๋ณ€์ด ์™„๋ฒฝํ•˜๊ฒŒ ๋…ผ๋ฆฌ์ ์ด๊ณ  ์ผ๊ด€์„ฑ์ด ์žˆ์œผ๋ฉฐ, ๋งค์šฐ ๊นŠ์ด ์žˆ๊ณ  ํ†ต์ฐฐ๋ ฅ ์žˆ๋Š” ๋ถ„์„์„ ์ œ๊ณตํ•œ๋‹ค.""",
    "์ˆ˜ํ•™(Math)": """[์–ธ์–ด ๋ชจ๋ธ์˜ ๋‹ต๋ณ€์ด ์ˆ˜ํ•™์ ์œผ๋กœ ์ •ํ™•ํ•˜๊ณ , ํ’€์ด ๊ณผ์ •์ด ๋ช…ํ™•ํ•˜๋ฉฐ, ํšจ์œจ์ ์ธ ํ•ด๋ฒ•์„ ์ œ์‹œํ•˜๋Š”๊ฐ€?]
Score 1: ๋‹ต๋ณ€์ด ์ˆ˜ํ•™์ ์œผ๋กœ ์™„์ „ํžˆ ๋ถ€์ •ํ™•ํ•˜๊ณ , ํ’€์ด ๊ณผ์ •์ด ์—†๊ฑฐ๋‚˜ ์ดํ•ดํ•  ์ˆ˜ ์—†์œผ๋ฉฐ, ๋น„ํšจ์œจ์ ์ธ ์ ‘๊ทผ๋ฒ•์„ ์‚ฌ์šฉํ•œ๋‹ค.
Score 2: ๋‹ต๋ณ€์— ์ผ๋ถ€ ์ •ํ™•ํ•œ ์ˆ˜ํ•™์  ์š”์†Œ๊ฐ€ ์žˆ์ง€๋งŒ ์ค‘์š”ํ•œ ์˜ค๋ฅ˜๊ฐ€ ์žˆ๊ณ , ํ’€์ด ๊ณผ์ •์ด ๋ถˆ์™„์ „ํ•˜๋ฉฐ, ๋น„ํšจ์œจ์ ์ธ ํ•ด๋ฒ•์„ ์ œ์‹œํ•œ๋‹ค.
Score 3: ๋‹ต๋ณ€์ด ๋Œ€์ฒด๋กœ ์ˆ˜ํ•™์ ์œผ๋กœ ์ •ํ™•ํ•˜์ง€๋งŒ ์ผ๋ถ€ ์˜ค๋ฅ˜๊ฐ€ ์žˆ๊ณ , ํ’€์ด ๊ณผ์ •์ด ์–ด๋Š ์ •๋„ ๋ช…ํ™•ํ•˜๋ฉฐ, ๊ธฐ๋ณธ์ ์ธ ํšจ์œจ์„ฑ์„ ๊ฐ–์ถ˜ ํ•ด๋ฒ•์„ ์ œ์‹œํ•œ๋‹ค.
Score 4: ๋‹ต๋ณ€์ด ์ˆ˜ํ•™์ ์œผ๋กœ ์ •ํ™•ํ•˜๊ณ , ํ’€์ด ๊ณผ์ •์ด ๋ช…ํ™•ํ•˜๋ฉฐ, ํšจ์œจ์ ์ธ ํ•ด๋ฒ•์„ ์ œ์‹œํ•œ๋‹ค. ์‚ฌ์†Œํ•œ ๊ฐœ์„ ์˜ ์—ฌ์ง€๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค.
Score 5: ๋‹ต๋ณ€์ด ์ˆ˜ํ•™์ ์œผ๋กœ ์™„๋ฒฝํ•˜๊ฒŒ ์ •ํ™•ํ•˜๊ณ , ํ’€์ด ๊ณผ์ •์ด ๋งค์šฐ ๋ช…ํ™•ํ•˜๋ฉฐ, ๊ฐ€์žฅ ํšจ์œจ์ ์ด๊ณ  ์ฐฝ์˜์ ์ธ ํ•ด๋ฒ•์„ ์ œ์‹œํ•œ๋‹ค.""",
    "๊ธ€์“ฐ๊ธฐ(Writing)": """[์–ธ์–ด ๋ชจ๋ธ์˜ ๋‹ต๋ณ€์ด ๊ตฌ์กฐ์ ์ด๊ณ , ๋ฌธ์ฒด๊ฐ€ ์ ์ ˆํ•˜๋ฉฐ, ๋‚ด์šฉ์ด ํ’๋ถ€ํ•˜๊ณ  ์ฐฝ์˜์ ์ธ๊ฐ€?]
Score 1: ๋‹ต๋ณ€์˜ ๊ตฌ์กฐ๊ฐ€ ์—†๊ณ , ๋ฌธ์ฒด๊ฐ€ ๋ถ€์ ์ ˆํ•˜๋ฉฐ, ๋‚ด์šฉ์ด ๋นˆ์•ฝํ•˜๊ณ  ์ฐฝ์˜์„ฑ์ด ์ „ํ˜€ ์—†๋‹ค.
Score 2: ๋‹ต๋ณ€์— ๊ธฐ๋ณธ์ ์ธ ๊ตฌ์กฐ๊ฐ€ ์žˆ์ง€๋งŒ ๋ถˆ์™„์ „ํ•˜๊ณ , ๋ฌธ์ฒด๊ฐ€ ์ผ๊ด€์„ฑ์ด ์—†์œผ๋ฉฐ, ๋‚ด์šฉ์ด ์ œํ•œ์ ์ด๊ณ  ์ฐฝ์˜์„ฑ์ด ๋ถ€์กฑํ•˜๋‹ค.
Score 3: ๋‹ต๋ณ€์ด ์–ด๋Š ์ •๋„ ๊ตฌ์กฐ์ ์ด๊ณ , ๋ฌธ์ฒด๊ฐ€ ๋ณดํ†ต ์ˆ˜์ค€์ด๋ฉฐ, ๋‚ด์šฉ์ด ์ ๋‹นํ•˜๊ณ  ์•ฝ๊ฐ„์˜ ์ฐฝ์˜์„ฑ์ด ์žˆ๋‹ค.
Score 4: ๋‹ต๋ณ€์ด ์ž˜ ๊ตฌ์กฐํ™”๋˜์–ด ์žˆ๊ณ , ๋ฌธ์ฒด๊ฐ€ ์ ์ ˆํ•˜๋ฉฐ, ๋‚ด์šฉ์ด ํ’๋ถ€ํ•˜๊ณ  ์ฐฝ์˜์ ์ธ ์š”์†Œ๊ฐ€ ์žˆ๋‹ค.
Score 5: ๋‹ต๋ณ€์ด ์™„๋ฒฝํ•˜๊ฒŒ ๊ตฌ์กฐํ™”๋˜์–ด ์žˆ๊ณ , ๋ฌธ์ฒด๊ฐ€ ํƒ์›”ํ•˜๋ฉฐ, ๋‚ด์šฉ์ด ๋งค์šฐ ํ’๋ถ€ํ•˜๊ณ  ๋†’์€ ์ˆ˜์ค€์˜ ์ฐฝ์˜์„ฑ์„ ๋ณด์ธ๋‹ค.""",
    "์ฝ”๋”ฉ(Coding)": """[์–ธ์–ด ๋ชจ๋ธ์˜ ๋‹ต๋ณ€์ด ์ •ํ™•ํ•˜๊ณ , ํšจ์œจ์ ์ด๋ฉฐ, ๊ฐ€๋…์„ฑ ์žˆ๋Š” ์ฝ”๋“œ๋ฅผ ์ œ๊ณตํ•˜๋Š”๊ฐ€?]
Score 1: ์ฝ”๋“œ๊ฐ€ ์ „ํ˜€ ์ž‘๋™ํ•˜์ง€ ์•Š๊ณ , ๋น„ํšจ์œจ์ ์ด๋ฉฐ, ๊ฐ€๋…์„ฑ์ด ๋งค์šฐ ๋‚ฎ๋‹ค.
Score 2: ์ฝ”๋“œ๊ฐ€ ๋ถ€๋ถ„์ ์œผ๋กœ ์ž‘๋™ํ•˜์ง€๋งŒ ์ฃผ์š” ์˜ค๋ฅ˜๊ฐ€ ์žˆ๊ณ , ํšจ์œจ์„ฑ์ด ๋‚ฎ์œผ๋ฉฐ, ๊ฐ€๋…์„ฑ์ด ๋ถ€์กฑํ•˜๋‹ค.
Score 3: ์ฝ”๋“œ๊ฐ€ ๋Œ€์ฒด๋กœ ์ž‘๋™ํ•˜๊ณ  ๊ธฐ๋ณธ์ ์ธ ํšจ์œจ์„ฑ์„ ๊ฐ–์ถ”์—ˆ์œผ๋‚˜, ์ผ๋ถ€ ๋ฒ„๊ทธ๊ฐ€ ์žˆ๊ณ  ๊ฐ€๋…์„ฑ์ด ๋ณดํ†ต ์ˆ˜์ค€์ด๋‹ค.
Score 4: ์ฝ”๋“œ๊ฐ€ ์ •ํ™•ํ•˜๊ฒŒ ์ž‘๋™ํ•˜๊ณ  ํšจ์œจ์ ์ด๋ฉฐ, ๊ฐ€๋…์„ฑ์ด ์ข‹๋‹ค. ์‚ฌ์†Œํ•œ ๊ฐœ์„ ์˜ ์—ฌ์ง€๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค.
Score 5: ์ฝ”๋“œ๊ฐ€ ์™„๋ฒฝํ•˜๊ฒŒ ์ž‘๋™ํ•˜๊ณ  ๋งค์šฐ ํšจ์œจ์ ์ด๋ฉฐ, ๋›ฐ์–ด๋‚œ ๊ฐ€๋…์„ฑ๊ณผ ์ตœ์ ํ™”๋œ ๊ตฌ์กฐ๋ฅผ ๊ฐ–์ถ”๊ณ  ์žˆ๋‹ค.""",
    "์ดํ•ด(Understanding)": """[์–ธ์–ด ๋ชจ๋ธ์ด ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ์˜๋ฏธ, ๋‰˜์•™์Šค, ๋ฌธ๋งฅ์„ ์ •ํ™•ํ•˜๊ฒŒ ์ดํ•ดํ•˜๊ณ  ์ ์ ˆํ•˜๊ฒŒ ์‘๋‹ตํ•˜๋Š”๊ฐ€?]
Score 1: ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ์˜๋ฏธ๋ฅผ ์ „ํ˜€ ์ดํ•ดํ•˜์ง€ ๋ชปํ•˜๊ณ , ๋ฌธ๋งฅ๊ณผ ๋‰˜์•™์Šค๋ฅผ ์™„์ „ํžˆ ๋†“์น˜๋ฉฐ, ๋ถ€์ ์ ˆํ•œ ์‘๋‹ต์„ ํ•œ๋‹ค.
Score 2: ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ๊ธฐ๋ณธ์ ์ธ ์˜๋ฏธ๋งŒ ๋ถ€๋ถ„์ ์œผ๋กœ ์ดํ•ดํ•˜๊ณ , ๋Œ€๋ถ€๋ถ„์˜ ๋ฌธ๋งฅ๊ณผ ๋‰˜์•™์Šค๋ฅผ ๋†“์น˜๋ฉฐ, ๋ถ€์ ์ ˆํ•œ ์‘๋‹ต์ด ๋งŽ๋‹ค.
Score 3: ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ์ฃผ์š” ์˜๋ฏธ๋ฅผ ์ดํ•ดํ•˜์ง€๋งŒ ์ผ๋ถ€ ๋‰˜์•™์Šค๋ฅผ ๋†“์น˜๊ณ , ๋ฌธ๋งฅ์„ ๋ถ€๋ถ„์ ์œผ๋กœ ํŒŒ์•…ํ•˜๋ฉฐ, ๋Œ€์ฒด๋กœ ์ ์ ˆํ•œ ์‘๋‹ต์„ ํ•œ๋‹ค.
Score 4: ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ์˜๋ฏธ๋ฅผ ์ •ํ™•ํžˆ ์ดํ•ดํ•˜๊ณ , ๋Œ€๋ถ€๋ถ„์˜ ๋‰˜์•™์Šค์™€ ๋ฌธ๋งฅ์„ ํŒŒ์•…ํ•˜๋ฉฐ, ์ ์ ˆํ•œ ์‘๋‹ต์„ ํ•œ๋‹ค.
Score 5: ํ•œ๊ตญ์–ด ํ…์ŠคํŠธ์˜ ์˜๋ฏธ, ๋‰˜์•™์Šค, ๋ฌธ๋งฅ์„ ์™„๋ฒฝํ•˜๊ฒŒ ์ดํ•ดํ•˜๊ณ , ๋งค์šฐ ์ ์ ˆํ•˜๊ณ  ์„ธ๋ จ๋œ ์‘๋‹ต์„ ํ•œ๋‹ค.""",
    "๋ฌธ๋ฒ•(Grammar)":"""[์–ธ์–ด ๋ชจ๋ธ์˜ ๋‹ต๋ณ€์ด ํ•œ๊ตญ์–ด ๋ฌธ๋ฒ• ๊ทœ์น™์„ ์ •ํ™•ํžˆ ๋”ฐ๋ฅด๊ณ , ์ ์ ˆํ•œ ์–ดํœ˜์™€ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•˜๋Š”๊ฐ€?]
Score 1: ์‹ฌ๊ฐํ•œ ๋ฌธ๋ฒ• ์˜ค๋ฅ˜๊ฐ€ ๋งŽ๊ณ , ๋ถ€์ ์ ˆํ•œ ์–ดํœ˜์™€ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•˜์—ฌ ์˜๋ฏธ ์ „๋‹ฌ์ด ๊ฑฐ์˜ ๋ถˆ๊ฐ€๋Šฅํ•˜๋‹ค.
Score 2: ์ค‘์š”ํ•œ ๋ฌธ๋ฒ• ์˜ค๋ฅ˜๊ฐ€ ์žˆ๊ณ , ์ œํ•œ๋œ ์–ดํœ˜์™€ ๋ถ€์ ์ ˆํ•œ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•˜์—ฌ ์˜๋ฏธ ์ „๋‹ฌ์— ์–ด๋ ค์›€์ด ์žˆ๋‹ค.
Score 3: ์ผ๋ถ€ ๋ฌธ๋ฒ• ์˜ค๋ฅ˜๊ฐ€ ์žˆ์ง€๋งŒ ์ „๋ฐ˜์ ์œผ๋กœ ์ดํ•ด ๊ฐ€๋Šฅํ•˜๋ฉฐ, ๊ธฐ๋ณธ์ ์ธ ์–ดํœ˜์™€ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•œ๋‹ค.
Score 4: ์‚ฌ์†Œํ•œ ๋ฌธ๋ฒ• ์˜ค๋ฅ˜๋งŒ ์žˆ๊ณ , ์ ์ ˆํ•œ ์–ดํœ˜์™€ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•˜์—ฌ ์˜๋ฏธ๋ฅผ ๋ช…ํ™•ํžˆ ์ „๋‹ฌํ•œ๋‹ค.
Score 5: ์™„๋ฒฝํ•œ ๋ฌธ๋ฒ•์„ ๊ตฌ์‚ฌํ•˜๊ณ , ๋‹ค์–‘ํ•˜๊ณ  ์ •ํ™•ํ•œ ์–ดํœ˜์™€ ์„ธ๋ จ๋œ ํ‘œํ˜„์„ ์‚ฌ์šฉํ•˜์—ฌ ์˜๋ฏธ๋ฅผ ํƒ์›”ํ•˜๊ฒŒ ์ „๋‹ฌํ•œ๋‹ค.""",
}

def judge(instruction, response, reference, rubrics):
    prompt = PROMETHEUS_PROMPT.format(
        instruction=instruction,
        response=response,
        reference=reference,
        rubrics=rubrics
    )
    input_ids = tokenizer.apply_chat_template(
        [{"role": "user", "content": prompt}],
        return_tensors="pt",
        add_generation_prompt=True
    ).to(model.device)
    judgement = model.generate(
        input_ids, 
        max_new_tokens=256,
        early_stopping=True, 
        eos_token_id=tokenizer.eos_token_id,
        )
    judgement = tokenizer.decode(judgement[0, input_ids.shape[-1]:], skip_special_tokens=True)
    score = int(judgement.split("[RESULT]")[1])
    return judgement, score

print(judge(
    "f(x) = 3x^3 + 2x^2 + 58์„ ๋ฏธ๋ถ„ํ•˜์‹œ์˜ค.",
    "ํ•จ์ˆ˜ f(x) = 3x^3 + 2x^2 + 58๋ฅผ ๋ฏธ๋ถ„ํ•˜๋ฉด,\n\nf'(x) = d(3x^3)/dx + d(2x^2)/dx + d(58)/dx\n\n= 3d(x^3)/dx + 2d(x^2)/dx + 0 (์ƒ์ˆ˜ํ•ญ์€ ๋ฏธ๋ถ„ํ•˜๋ฉด 0)\n\n= 3(3x^2) + 2(2x)\n\n= 9x^2 + 4x\n\n๋”ฐ๋ผ์„œ, ํ•จ์ˆ˜ f(x) = 3x^3 + 2x^2 + 58์˜ ๋ฏธ๋ถ„๊ฐ’์€ f'(x) = 9x^2 + 4x์ž…๋‹ˆ๋‹ค.",
    "9x^2 + 4x",
    RUBRICS["์ˆ˜ํ•™(Math)"]
))

๊ฒฐ๊ณผ: ์ด ๋‹ต๋ณ€์€ ํ•จ์ˆ˜ f(x) = 3x^3 + 2x^2 + 58๋ฅผ ๋ฏธ๋ถ„ํ•˜๋Š” ๊ณผ์ •์„ ์ •ํ™•ํ•˜๊ฒŒ ์ˆ˜ํ–‰ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐ ํ•ญ์˜ ๋ฏธ๋ถ„์„ ๋ช…ํ™•ํ•˜๊ฒŒ ์„ค๋ช…ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ƒ์ˆ˜ํ•ญ์— ๋Œ€ํ•œ ๋ฏธ๋ถ„์˜ ๊ฒฐ๊ณผ๋ฅผ ์˜ฌ๋ฐ”๋ฅด๊ฒŒ ๋‚˜ํƒ€๋ƒˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๋ฏธ๋ถ„ ๊ณผ์ •์„ ์„ค๋ช…ํ•˜๋Š” ๊ณผ์ •์—์„œ ์ข€ ๋” ์„ธ๋ถ€์ ์ธ ์„ค๋ช…์„ ์ œ๊ณตํ•  ์ˆ˜ ์žˆ์—ˆ์„ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ๊ฐ ํ•ญ์— ๋Œ€ํ•œ ๋ฏธ๋ถ„์„ ์–ด๋–ป๊ฒŒ ์ˆ˜ํ–‰ํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ์„ค๋ช…์ด ๋” ๋ช…ํ™•ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์‚ฌ์†Œํ•œ ๊ฐœ์„ ์˜ ์—ฌ์ง€๊ฐ€ ์žˆ๊ธด ํ•˜์ง€๋งŒ, ์ „๋ฐ˜์ ์œผ๋กœ ์ˆ˜ํ•™์ ์œผ๋กœ ์ •ํ™•ํ•˜๊ณ  ํšจ์œจ์ ์ธ ํ•ด๋ฒ•์„ ์ œ์‹œํ•˜๊ณ  ์žˆ๊ธฐ ๋•Œ๋ฌธ์— 4์ ์ด ์ ์ ˆํ•ฉ๋‹ˆ๋‹ค. [RESULT] 4', 4

Downloads last month
0
Inference API
Unable to determine this modelโ€™s pipeline type. Check the docs .

Model tree for heegyu/ko-prometheus-8b-lora-0708

Adapter
(58)
this model

Dataset used to train heegyu/ko-prometheus-8b-lora-0708