polyglot-ko-12.8b-instruct

polyglot-ko-12.8b-instruct

This model is a fine-tuned version of EleutherAI/polyglot-ko-12.8b on an instruction-following dataset(260k).

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 1
• seed: 42
• distributed_type: multi-GPU(A100 80G)
• num_devices: 8
• gradient_accumulation_steps: 64
• total_train_batch_size: 512
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5.0

Inference

import torch
from transformers import pipeline, AutoModelForCausalLM

MODEL = 'etri-xainlp/polyglot-ko-12.8b-instruct'

model = AutoModelForCausalLM.from_pretrained(
    MODEL,
    torch_dtype=torch.float16,
    low_cpu_mem_usage=True,
).to(device=f"cuda", non_blocking=True)
model.eval()

pipe = pipeline(
    'text-generation', 
    model=model,
    tokenizer=MODEL,
    device=0
)
pipe.model.config.pad_token_id = pipe.model.config.eos_token_id

def ask(x, context='', is_input_full=False):
    ans = pipe(
        f"### 질문: {x}\n\n### 맥락: {context}\n\n### 답변:" if context else f"### 질문: {x}\n\n### 답변:", 
        do_sample=True, 
        max_new_tokens=2048,
        temperature=0.9,
        top_p=0.9,
        return_full_text=False,
        eos_token_id=2,
    )
    return ans[0]['generated_text']

while True:
    quit = input('prompt?: ')
    if quit == 'q':
        break
    else:
        generation = ask(quit)
        print("etri_ai:", generation)

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3