Edit model card

Model Name : SungBeom/whisper-small-ko

Description

  • ํŒŒ์ธํŠœ๋‹ ๋ฐ์ดํ„ฐ์…‹ : maxseats/aihub-464-preprocessed-680GB-set-0
  • AI hub์˜ ์ฃผ์š” ์˜์—ญ๋ณ„ ํšŒ์˜ ์Œ์„ฑ ๋ฐ์ดํ„ฐ์…‹ 680GB ์ค‘ ์ฒซ๋ฒˆ์งธ ๋ฐ์ดํ„ฐ(10GB)๋ฅผ ํŒŒ์ธํŠœ๋‹ํ•œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.
  • ๋ฐ์ดํ„ฐ์…‹ ๋งํฌ : https://huggingface.co/datasets/maxseats/aihub-464-preprocessed-680GB-set-0

ํŒŒ๋ผ๋ฏธํ„ฐ

model_name = "SungBeom/whisper-small-ko" # ๋Œ€์•ˆ : "SungBeom/whisper-small-ko"
dataset_name = "maxseats/aihub-464-preprocessed-680GB-set-0"  # ๋ถˆ๋Ÿฌ์˜ฌ ๋ฐ์ดํ„ฐ์…‹(ํ—ˆ๊น…ํŽ˜์ด์Šค ๊ธฐ์ค€)

CACHE_DIR = '/mnt/a/maxseats/.finetuning_cache'  # ์บ์‹œ ๋””๋ ‰ํ† ๋ฆฌ ์ง€์ •
is_test = False  # True: ์†Œ๋Ÿ‰์˜ ์ƒ˜ํ”Œ ๋ฐ์ดํ„ฐ๋กœ ํ…Œ์ŠคํŠธ, False: ์‹ค์ œ ํŒŒ์ธํŠœ๋‹

token = "hf_" # ํ—ˆ๊น…ํŽ˜์ด์Šค ํ† ํฐ ์ž…๋ ฅ

training_args = Seq2SeqTrainingArguments(
    output_dir=model_dir,  # ์›ํ•˜๋Š” ๋ฆฌํฌ์ง€ํ† ๋ฆฌ ์ด๋ฆ„์„ ์ž…๋ ฅํ•œ๋‹ค.
    per_device_train_batch_size=16,
    gradient_accumulation_steps=2,  # ๋ฐฐ์น˜ ํฌ๊ธฐ๊ฐ€ 2๋ฐฐ ๊ฐ์†Œํ•  ๋•Œ๋งˆ๋‹ค 2๋ฐฐ์”ฉ ์ฆ๊ฐ€
    learning_rate=1e-5,
    warmup_steps=1000,
    # max_steps=2,  # epoch ๋Œ€์‹  ์„ค์ •
    num_train_epochs=1,     # epoch ์ˆ˜ ์„ค์ • / max_steps์™€ ์ด๊ฒƒ ์ค‘ ํ•˜๋‚˜๋งŒ ์„ค์ •
    gradient_checkpointing=True,
    fp16=True,
    evaluation_strategy="steps",
    per_device_eval_batch_size=16,
    predict_with_generate=True,
    generation_max_length=225,
    save_steps=1000,
    eval_steps=1000,
    logging_steps=25,
    report_to=["tensorboard"],
    load_best_model_at_end=True,
    metric_for_best_model="cer",  # ํ•œ๊ตญ์–ด์˜ ๊ฒฝ์šฐ 'wer'๋ณด๋‹ค๋Š” 'cer'์ด ๋” ์ ํ•ฉํ•  ๊ฒƒ
    greater_is_better=False,
    push_to_hub=True,
    save_total_limit=5,           # ์ตœ๋Œ€ ์ €์žฅํ•  ๋ชจ๋ธ ์ˆ˜ ์ง€์ •
)
Downloads last month
3
Safetensors
Model size
242M params
Tensor type
F32
ยท
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for maxseats/SungBeom-whisper-small-ko-set0

Finetunes
1 model

Dataset used to train maxseats/SungBeom-whisper-small-ko-set0