|
--- |
|
language: |
|
- en |
|
- ko |
|
library_name: peft |
|
tags: |
|
- translation |
|
- gemma |
|
base_model: google/gemma-7b |
|
--- |
|
|
|
# Model Card for Model ID |
|
## Model Details |
|
### Model Description |
|
- **Developed by:** [Kang Seok Ju] |
|
- **Contact:** [brildev7@gmail.com] |
|
|
|
## Training Details |
|
### Training Data |
|
https://huggingface.co/datasets/traintogpb/aihub-koen-translation-integrated-tiny-100k |
|
|
|
# Inference Examples |
|
``` |
|
import os |
|
import torch |
|
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig |
|
from peft import PeftModel |
|
|
|
model_id = "google/gemma-7b" |
|
peft_model_id = "brildev7/gemma-7b-translation-enko-sft-qlora" |
|
quantization_config = BitsAndBytesConfig( |
|
load_in_4bit=True, |
|
bnb_4bit_compute_dtype=torch.float16, |
|
bnb_4bit_quant_type="nf4", |
|
bnb_4bit_use_double_quant=False |
|
) |
|
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
model_id, |
|
quantization_config=quantization_config, |
|
torch_dtype=torch.float16, |
|
attn_implementation="flash_attention_2", |
|
token=os.environ['HF_TOKEN'], |
|
device_map="auto" |
|
) |
|
model = PeftModel.from_pretrained(model, peft_model_id) |
|
|
|
tokenizer = AutoTokenizer.from_pretrained(peft_model_id) |
|
tokenizer.pad_token_id = tokenizer.eos_token_id |
|
|
|
# example |
|
prompt_template = """Translate the following sentences into Korean language: |
|
{} |
|
|
|
translation: |
|
""" |
|
sentences = "Apple is facing a crisis in one of its key markets, China, as it is being challenged by local smartphone manufacturers. In a bid to counter the threat, Apple CEO Tim Cook is reportedly planning to visit China to meet with local smartphone manufacturers and discuss a joint investment. Apple is also reportedly considering installing an AI model from Baidu, the Chinese search giant, on its iPhone. The move comes as Apple is facing a price war in China, with local smartphone manufacturers offering steep discounts on their products." |
|
texts = prompt_template.format(sentences) |
|
inputs = tokenizer(texts, return_tensors="pt").to(model.device) |
|
|
|
outputs = model.generate(**inputs, max_new_tokens=1024) |
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
- μ νμ κ΅μ° μ€λ§νΈν° μ μ‘°μ¬λ€μ λλ°μ μ€κ΅μμ νλμ ν΅μ¬ μμ₯μ μκΈ°λ₯Ό λ§κ³ μλ€. μ΄ μνμ νκ°νκΈ° μν΄ μ νμ μ΅κ³ κ²½μμμΈ ν μΏ‘μ μ€κ΅μ λ°©λ¬Έν΄ νμ§ μ€λ§νΈν° μ μ‘°μ¬λ€κ³Ό μ μ΄ν΄ 곡λ ν¬μλ₯Ό λ
Όμνλ κ²μΌλ‘ μλ €μ‘λ€. μ νμ λν μ€κ΅ μ΅λ κ²μμ¬ λ°μ΄λ(Baidu)μ μΈκ³΅ μ§λ₯(AI) λͺ¨λΈμ μμ΄ν°μ νμ¬νλ κ²μ κ²ν μ€μΈ κ²μΌλ‘ μ ν΄μ‘λ€. μ νμ κ΅λ΄ μ€λ§νΈν° μ μ‘°μ¬λ€μ΄ μμ λ€μ μ νμ κΈν ν μΈμ λ΄λμΌλ©΄μ μ€κ΅μμ κ°κ²©μ μμ μ§λ©΄ν΄ μλ κ²μ΄λ€. |
|
|
|
# example |
|
sentences = "Is it safe to drink milk and eat chicken?" |
|
texts = prompt_template.format(sentences) |
|
inputs = tokenizer(texts, return_tensors="pt").to(model.device) |
|
|
|
outputs = model.generate(**inputs, max_new_tokens=1024) |
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
- μ°μ μ λκ³ κΈ°λ μμ νκ°μ? |
|
|
|
# example |
|
sentences = "What precautions to take during the bird flu outbreak" |
|
texts = prompt_template.format(sentences) |
|
inputs = tokenizer(texts, return_tensors="pt").to(model.device) |
|
|
|
outputs = model.generate(**inputs, max_new_tokens=1024) |
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
|
- μ‘°λ₯ λ
κ° μ ν μ μ΄λ ν μ£Όμ μ¬νμ ν΄μΌ νλμ§ |
|
|
|
``` |