Views
No views yet
1{% if messages %}{% for message in messages %}{% if message['role'] == 'user' %}<|start_header_id|>user<|end_header_id|>
2
3{{ message['content'] }}<|eot_id|>{% elif message['role'] == 'assistant' %}<|start_header_id|>assistant<|end_header_id|>
4
5{{ message['content'] }}<|eot_id|>{% endif %}{% endfor %}{% if add_generation_prompt %}<|start_header_id|>assistant<|end_header_id|>
6
7{% endif %}{% endif %}<|begin_of_text|>: Beginning of text token (ID: 128000)<|end_of_text|>: End of text token (ID: 128001)<|start_header_id|>: Start of role header (ID: 128006)<|end_header_id|>: End of role header (ID: 128007)<|eot_id|>: End of turn token (ID: 128009)1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model = AutoModelForCausalLM.from_pretrained("path/to/model")
4tokenizer = AutoTokenizer.from_pretrained("path/to/model")1messages = [
2 {"role": "user", "content": "Hello, how are you?"},
3 {"role": "assistant", "content": "I'm doing well, thank you! How can I help you today?"},
4 {"role": "user", "content": "Can you explain what you are?"}
5]
6
7# Apply chat template
8formatted_text = tokenizer.apply_chat_template(
9 messages,
10 tokenize=False,
11 add_generation_prompt=True
12)
13
14# Tokenize
15inputs = tokenizer(formatted_text, return_tensors="pt")
16
17# Generate
18outputs = model.generate(**inputs, max_new_tokens=256)
19response = tokenizer.decode(outputs[0], skip_special_tokens=False)1from transformers import Trainer, TrainingArguments
2
3training_args = TrainingArguments(
4 output_dir="./results",
5 num_train_epochs=3,
6 per_device_train_batch_size=4,
7 gradient_accumulation_steps=4,
8 learning_rate=2e-5,
9 warmup_steps=100,
10 logging_steps=10,
11 save_steps=500,
12)
13
14trainer = Trainer(
15 model=model,
16 args=training_args,
17 train_dataset=train_dataset,
18 eval_dataset=eval_dataset,
19)
20
21trainer.train()1@article{llama3,
2 title={Llama 3: Open Foundation and Fine-Tuned Language Models},
3 author={Meta AI},
4 year={2024}
5}