meomeo163/medical_chatbot

Model Details

Model Description

• Model type: Text generation
• Language(s) (NLP): Vietnamese
• Finetuned from model: arcee-ai/Arcee-VyLinh

How to Get Started with the Model

from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import TextStreamer
from peft import PeftModel
import torch
# Load the model and tokenizer
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
base_model = AutoModelForCausalLM.from_pretrained("arcee-ai/Arcee-VyLinh")
tokenizer = AutoTokenizer.from_pretrained("arcee-ai/Arcee-VyLinh")
adapter_model = PeftModel.from_pretrained(base_model, "meomeo163/medical_chatbot").to(device)
prompt = "mình đang có hiện tượng bị đau bụng dưới, thì thoảng thấy buồn nôn, các chứng bệnh mình có thể gặp phải là gì"
messages = [
    {"role": "system", "content": "Bạn là trợ lý y tế chuyên nghiệp"},
    {"role_type": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

streamer = TextStreamer(
    tokenizer,
    skip_prompt=True,     
    skip_special_tokens=True 
)

generated_ids = adapter_model.generate(
    model_inputs.input_ids,
    attention_mask=model_inputs.attention_mask, # Add attention mask
    pad_token_id=tokenizer.eos_token_id, # Set pad token id
    max_new_tokens=512,
    eos_token_id=tokenizer.eos_token_id,
    temperature=0.25,
    streamer=streamer,
    no_repeat_ngram_size=3 # Add or increase no_repeat_ngram_size
)

generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids)[0]
print(response)

Training Details

Training Data

A question-asnwer data about medical [hungnm/vietnamese-medical-qa]

Training Hyperparameters

Training regime:

training_args = TrainingArguments(
    per_device_train_batch_size = 4,
    gradient_accumulation_steps = 4,
    warmup_steps = 100,

    num_train_epochs = 3,
    learning_rate = 2e-4,
    bf16 = True,

    logging_steps = 25,
    output_dir = "finetuned_medical_qa_full",

    optim = "adamw_8bit",
    eval_strategy = "steps",
    eval_steps = 100,
    save_strategy = "steps",
    save_steps = 100,
    
    load_best_model_at_end = True,
    metric_for_best_model = "eval_loss",
    greater_is_better = False,
    save_total_limit = 2,
)

• PEFT 0.17.1