metadata

license: apache-2.0

phibode-3-mini-4k-ultraalpaca

phibode-3-mini-4k-ultraalpaca is an SFT fine-tuned version of microsoft/Phi-3-mini-4k-instruct using a custom training dataset. This model was made with Phinetune

Process

Learning Rate: 1.41e-05
Maximum Sequence Length: 2048
Dataset: recogna-nlp/ultra-alpaca-ptbr
Split: train

💻 Usage

!pip install -qU transformers
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

model = "recogna-nlp/phibode-3-mini-4k-ultraalpaca"
tokenizer = AutoTokenizer.from_pretrained(model)

# Example prompt
messages = [
    {"role": "system", "content": "Você é assistente de IA chamado PhiBode. O PhiBode é um modelo de língua conversacional projetado para ser prestativo, honesto e inofensivo."},
    {"role": "user", "content": "<Insira seu prompt aqui>"},
]

# Generate a response
model = AutoModelForCausalLM.from_pretrained(model, trust_remote_code=True)
pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
generation_args = {
    "max_new_tokens": 500,
    "return_full_text": False,
    "temperature": 0.0,
    "do_sample": False,
}
outputs = pipeline(messages, **generation_args)
print(outputs[0]["generated_text"])