Edit model card

Model Card for Model ID

Model Details

  • meta-llama/Llama-2-7b-chat-hf๋ฅผ ๋ฏธ์„ธํŠœ๋‹ํ•˜์—ฌ ๊ธ€์˜ ๊ธ์ •๊ณผ ๋ถ€์ •์„ ๊ตฌ๋ถ„ํ•˜๋Š” ๋ชจ๋ธ

Model Description

  • Developed by: ์„ฑ์‹ ์—ฌ์ž๋Œ€ํ•™๊ต 20211421 ํ™ฉ๊ทœ์›

  • Finetuned from model: KT-AI/midm-bitext-S-7B-inst-v1

Uses

  • NSMC ๋ฐ์ดํ„ฐ์…‹์œผ๋กœ ๊ธ€์˜ ๊ธ์ •๊ณผ ๋ถ€์ •์„ ๊ตฌ๋ถ„ํ•˜๋Š” ๋ชจ๋ธ์„ ๋ฏธ์„ธํŠœ๋‹ํ•จ์ด ๋ชฉ์ 
  • ๋ฌธ์žฅ์˜ ๊ฐ์ •์ด ๊ธ์ •์ด๋ฉด '1'๋กœ, ๊ทธ ์™ธ๋Š” '0'์œผ๋กœ ๋ ˆ์ด๋ธ”์„ ํ•จ

Training Data

  • NSMC ๋ฐ์ดํ„ฐ์…‹์˜ train ์Šคํ”Œ๋ฆฟ ์•ž์ชฝ 3,000๊ฐœ์˜ ์ƒ˜ํ”Œ์„ ํ•™์Šต์— ์‚ฌ์šฉ

Training Procedure

  • 512 ์‹œํ€€์Šค ๊ธธ์ด
  • 1600 steps์˜ ํ•™์Šต

Testing Data, Factors & Metrics

Testing Data

  • NSMC ๋ฐ์ดํ„ฐ์…‹์˜ test ์Šคํ”Œ๋ฆฟ ์•ž์ชฝ 1,000๊ฐœ์˜ ์ƒ˜ํ”Œ์„ ํ•™์Šต์— ์‚ฌ์šฉ

Metrics

Llama2 ์ •ํ™•๋„ Midm ์ •ํ™•๋„ ์ •๋ฐ€๋„ ์žฌํ˜„์œจ
0.874 0.9 0.891 0.856
  • ๋ฏฟ์Œ ๋ชจ๋ธ์„ ๋ฏธ์„ธํŠœ๋‹ํ•œ ๊ฒƒ์ด ๋” ๋‚˜์€ ์ •ํ™•๋„๋ฅผ ๋ณด์—ฌ์คŒ

Results

llama2ConfusionMatrix

Summary

  • ์ •๋‹ต๋ฅ ์„ ์˜ฌ๋ฆฌ๊ธฐ ์œ„ํ•ด 2,000๊ฐœ์˜ ๋ฐ์ดํ„ฐ์…‹์ด ์•„๋‹Œ 3,000๊ฐœ์˜ ํ•™์Šต ๋ฐ์ดํ„ฐ๋ฅผ ๋„ฃ์–ด๋ด„
  • steps์ˆ˜๋„ ์˜ฌ๋ ค ๋” ๋‚ฎ์€ loss๊ฐ’์ด ๋‚˜์˜ค๊ธธ ๊ธฐ๋Œ€ํ–ˆ์Œ

Training procedure

The following bitsandbytes quantization config was used during training:

  • quant_method: bitsandbytes
  • 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: False
  • bnb_4bit_compute_dtype: bfloat16

Framework versions

  • PEFT 0.7.0
Downloads last month
6
Inference API
Unable to determine this modelโ€™s pipeline type. Check the docs .

Model tree for guguwon/hw-llama-2-7B-nsmc

Adapter
(1037)
this model