Persiapan fine-tuning

Pengantar LLM di Python

Jasmin Ludolf

Senior Data Science Content Developer, DataCamp

Pipelines dan kelas Auto

Pipelines: pipeline()

  • Mempermudah tugas
  • Pemilihan model dan tokenizer otomatis
  • Kontrol terbatas

Pipelines di Hugging Face Transformers

Kelas Auto (AutoModel class)

  • Kustomisasi
  • Penyesuaian manual
  • Mendukung fine-tuning

Kelas AutoModel di Hugging Face Transformers dengan opsi kustomisasi

Pengantar LLM di Python

Siklus hidup LLM

Siklus hidup LLM

Pre-training

  • Data luas
  • Belajar pola umum
Pengantar LLM di Python

Siklus hidup LLM

Siklus hidup LLM

Pre-training                                                               Fine-tuning

  • Data luas                                                          Spesifik domain
  • Belajar pola umum                                       Tugas khusus
Pengantar LLM di Python

Memuat dataset untuk fine-tuning

from datasets import load_dataset


train_data = load_dataset("imdb", split="train")

train_data = data.shard(num_shards=4, index=0)


test_data = load_dataset("imdb", split="test")

test_data = data.shard(num_shards=4, index=0)
  • load_dataset(): memuat dataset dari Hugging Face Hub
    • imdb: klasifikasi ulasan
Pengantar LLM di Python

Kelas Auto

from transformers import AutoModel, AutoTokenizer

from transformers import AutoModelForSequenceClassification


model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
Pengantar LLM di Python

Tokenisasi

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from datasets import load_dataset

train_data = load_dataset("imdb", split="train")
train_data = data.shard(num_shards=4, index=0)
test_data = load_dataset("imdb", split="test")
test_data = data.shard(num_shards=4, index=0)

model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")


# Tokenize the data
tokenized_training_data = tokenizer(train_data["text"], return_tensors="pt", padding=True, truncation=True, 
                                    max_length=64)

tokenized_test_data = tokenizer(test_data["text"], return_tensors="pt", padding=True, truncation=True, 
                                max_length=64)
Pengantar LLM di Python

Keluaran tokenisasi

print(tokenized_training_data)

{'input_ids': tensor([[  101,  1045, 12524,  1045,  2572,  8025,  1011,  3756,  
2013,  2026, 2678,  3573,  2138,  1997,  2035,  1996,  6704,  2008,  5129,  2009, 
2043,  2009, 2001,  2034,  2207,  1999,  3476,  1012,  1045,  2036, ...
Pengantar LLM di Python

Tokenisasi per baris

def tokenize_function(text_data):
    return tokenizer(text_data["text"], return_tensors="pt", padding=True, truncation=True, max_length=64)

# Tokenize in batches
tokenized_in_batches = train_data.map(tokenize_function, batched=True)


# Tokenize row by row
tokenized_by_row = train_data.map(tokenize_function, batched=False)

Dataset({
    features: ['text', 'label', 'input_ids', 'token_type_ids', 'attention_mask'],
    num_rows: 1563
})
Pengantar LLM di Python

Tokenisasi subword

  • Umum pada tokenizer modern
  • Kata dipecah menjadi sub-bagian bermakna

 

Unbelievably

Pengantar LLM di Python

Tokenisasi subword

  • Umum pada tokenizer modern
  • Kata dipecah menjadi sub-bagian bermakna

 

Unbelievably di-tokenisasi menjadi un, believ, ably.

Pengantar LLM di Python

Ayo berlatih!

Pengantar LLM di Python

Preparing Video For Download...