POKASH.PL
/
ably.do
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

mod gpt

This commit is contained in:
l.gabrysiak 2025-02-26 10:29:26 +01:00
parent 8ee5f5cbd9
commit f846ddeabe
2 changed files with 15 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
gpt.py
View File

@ -44,7 +44,6 @@ def prepare_dataset_from_file(file_path):
return formatted_articles return formatted_articles
def main(): def main():
# Inicjalizacja tokenizera # Inicjalizacja tokenizera
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME) tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
@ -61,17 +60,17 @@ def main():
examples["text"], examples["text"],
truncation=True, truncation=True,
padding="max_length", padding="max_length",
max_length=256, # Zwiększono dla dłuższych artykułów max_length=2048, # Zwiększono dla dłuższych artykułów
return_tensors="pt" return_tensors="pt"
) )
tokenized["labels"] = tokenized["input_ids"].clone() tokenized["labels"] = tokenized["input_ids"].clone()
return tokenized return tokenized
tokenized_dataset = dataset.map(tokenize_function, batched=True) tokenized_dataset = dataset.map(tokenize_function, batched=True, remove_columns=dataset.column_names)
# Model i data collator # Model i data collator
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME) model = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
model.resize_token_embeddings(len(tokenizer), mean_resizing=False) model.resize_token_embeddings(len(tokenizer))
data_collator = DataCollatorForLanguageModeling( data_collator = DataCollatorForLanguageModeling(
tokenizer=tokenizer, tokenizer=tokenizer,
@ -81,12 +80,17 @@ def main():
# Konfiguracja treningu # Konfiguracja treningu
training_args = TrainingArguments( training_args = TrainingArguments(
output_dir="./results", output_dir="./results",
num_train_epochs=8, # Zwiększono liczbę epok num_train_epochs=15, # Zwiększono liczbę epok
per_device_train_batch_size=2, per_device_train_batch_size=4, # Zwiększono rozmiar batcha
learning_rate=5e-5, learning_rate=2e-5, # Zmniejszono learning rate
weight_decay=0.01, # Dodano weight decay
logging_steps=10, logging_steps=10,
save_steps=500, # Dodano zapisywanie modelu co 500 kroków
eval_steps=500, # Dodano ewaluację co 500 kroków
evaluation_strategy="steps",
load_best_model_at_end=True, # Ładowanie najlepszego modelu na końcu
report_to="none", report_to="none",
save_strategy="no" save_total_limit=2, # Ograniczenie liczby zapisywanych checkpointów
) )
# Trainer # Trainer
@ -94,6 +98,7 @@ def main():
model=model, model=model,
args=training_args, args=training_args,
train_dataset=tokenized_dataset, train_dataset=tokenized_dataset,
eval_dataset=tokenized_dataset, # Używamy tego samego zbioru do ewaluacji
data_collator=data_collator data_collator=data_collator
) )

2
test.py
View File

@ -17,6 +17,6 @@ def generate_response(prompt, max_length=1000):
response = tokenizer.decode(outputs[0], skip_special_tokens=True) response = tokenizer.decode(outputs[0], skip_special_tokens=True)
return response return response
prompt = "Zacytuj art. 154 kodeksu pracy" prompt = "Jak brzmi art. 154 kodeksu pracy"
response = generate_response(prompt) response = generate_response(prompt)
print(response) print(response)