hw-llama-2-7B-nsmc / README.md
seojin0128's picture
Update README.md
03a6641
|
raw
history blame
1.68 kB
---
library_name: peft
base_model: meta-llama/Llama-2-7b-chat-hf
metrics:
- accuracy 89.7%
datasets:
- nsmc
---
# Model Card for Model ID
## Model Description
### llama-2-7b-chat-hf λ―Έμ„Έ νŠœλ‹
ν•΄λ‹Ή λͺ¨λΈμ€ 넀이버 μ˜ν™” 리뷰 데이터셋인 NSMC에 λŒ€ν•΄ meta-llama/Llama-2-7b-chat-hf을 λ―Έμ„ΈνŠœλ‹ν•œ λͺ¨λΈμž…λ‹ˆλ‹€.
μ˜ν™” 리뷰 ν…μŠ€νŠΈλ₯Ό ν”„λ‘¬ν”„νŠΈμ— ν¬ν•¨ν•˜μ—¬ λͺ¨λΈμ— μž…λ ₯μ‹œ,'긍정' λ˜λŠ” 'λΆ€μ •' 이라고 예츑 ν…μŠ€νŠΈλ₯Ό 직접 μƒμ„±ν•©λ‹ˆλ‹€.
결과적으둜, 정확도 89.7%λ₯Ό κ°€μ§€λŠ” λͺ¨λΈμ„ μ™„μ„±ν–ˆμŠ΅λ‹ˆλ‹€.
### Train, Test 데이터셋
ν•΄λ‹Ή λͺ¨λΈμ€ NSMC의 train λ°μ΄ν„°μ˜ μƒμœ„ 2,000개의 μƒ˜ν”Œμ„ ν•™μŠ΅μ— μ‚¬μš©ν–ˆμŠ΅λ‹ˆλ‹€.
ν•΄λ‹Ή λͺ¨λΈμ€ NSMC의 test λ°μ΄ν„°μ˜ μƒμœ„ 1,000개의 μƒ˜ν”Œμ„ 평가에 μ‚¬μš©ν–ˆμŠ΅λ‹ˆλ‹€.
## 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
## Training_step_loss
![image/png](https://cdn-uploads.huggingface.co/production/uploads/651bf3be3fa6c4e182910420/hRnLRrx1TN3RRJYnhxj7I.png)
## Confusion_Matrix
![image/png](https://cdn-uploads.huggingface.co/production/uploads/651bf3be3fa6c4e182910420/w2KkvdL15H1R8AzwK6Mv3.png)
## Accuracy_Classification_Report
![image/png](https://cdn-uploads.huggingface.co/production/uploads/651bf3be3fa6c4e182910420/d5zu_ZmC_jS8vxIxFJ9R9.png)
## Framework versions
- PEFT 0.7.0