lora-midm-nsmc / README.md
chaem's picture
Update README.md
3cf0c18
metadata
license: cc-by-nc-4.0
base_model: KT-AI/midm-bitext-S-7B-inst-v1
tags:
  - generated_from_trainer
model-index:
  - name: lora-midm-nsmc
    results: []

lora-midm-nsmc

This model is a fine-tuned version of KT-AI/midm-bitext-S-7B-inst-v1 on an nsmc dataset.

Model description

KT-midm model을 nsmc데이터λ₯Ό ν™œμš©ν•˜μ—¬ λ―Έμ„ΈνŠœλ‹ν•œ λͺ¨λΈ
μ˜ν™” 리뷰 데이터λ₯Ό 기반으둜 μ‚¬μš©μžκ°€ μž‘μ„±ν•œ 리뷰의 긍정 λ˜λŠ” 뢀정을 νŒŒμ•…ν•œλ‹€.

Intended uses & limitations

Intended uses

μ‚¬μš©μžκ°€ μž‘μ„±ν•œ 리뷰의 긍정 λ˜λŠ” λΆ€μ • 감정 뢄석을 μ œκ³΅ν•¨

Limitaions

μ˜ν™” 리뷰에 νŠΉν™”λ˜μ–΄ 있으며, λ‹€λ₯Έ μœ ν˜•μ—λŠ” μ œν•œμ΄ μžˆμ„ 수 있음
Colab T4 GPUμ—μ„œ ν…ŒμŠ€νŠΈ λ˜μ—ˆμŒ

Training and evaluation data

Training data: nsmc 'train' data 쀑 μƒμœ„ 2000개의 μƒ˜ν”Œ
Evaluation data: nsmc 'test' data 쀑 μƒμœ„ 1000개의 μƒ˜ν”Œ

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0001
  • train_batch_size: 1
  • eval_batch_size: 1
  • seed: 42
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 2
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.03
  • training_steps: 300
  • mixed_precision_training: Native AMP

Training results

image/png

TrainOutput(global_step=300, training_loss=1.1105608495076498,
metrics={'train_runtime': 929.3252, 'train_samples_per_second': 0.646,
'train_steps_per_second': 0.323, 'total_flos': 9315508499251200.0,
'train_loss': 1.1105608495076498, 'epoch': 0.3})

정확도

Midm: 정확도 0.89

Positive Prediction(PP) Negative Prediction(NP)
True Positive (TP) 474 34
True Negative (TN) 76 416

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.1.0+cu118
  • Datasets 2.15.0
  • Tokenizers 0.15.0