About :

This 🦙 Llama model was trained on a translated Alpaca dataset in Bahasa Indonesia. It uses Parameter Efficient Fine Tuning and LoRA to enable training on consumer-grade GPU hardware.

How to Use :

Load the 🦙 Alpaca-LoRA model

import torch
import bitsandbytes as bnb
from transformers import LlamaTokenizer, LlamaForCausalLM, GenerationConfig
from peft import PeftModel, PeftConfig, prepare_model_for_int8_training, LoraConfig, get_peft_model

peft_model_id = "firqaaa/indo-Alpaca-LoRA-7b"

tokenizer = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf")
model = LlamaForCausalLM.from_pretrained("decapoda-research/llama-7b-hf",
                                         load_in_8bit=True,
                                         device_map="auto")
# Load the LoRA model
model = PeftModel.from_pretrained(model, peft_model_id)

Prompt Template

Prepare the prompt template

instruction = "Tuliskan deret bilangan fibbonaci. Tulis jawaban/respons dalam Bahasa Indonesia."

PROMPT = f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{instruction}
### Response:"""

Evaluation

feel free to change the parameters inside GenerationConfig to get better result.

inputs = tokenizer(
    PROMPT,
    return_tensors="pt"
)
input_ids = inputs["input_ids"].cuda()

generation_config = GenerationConfig(
    temperature=0.1,
    top_p=0.95,
    top_k=40,
    num_beams=4,
    repetition_penalty=1.15,
)
print("Generating...")
print("Instruction : {}".format(instruction))

generation_output = model.generate(
    input_ids=input_ids,
    generation_config=generation_config,
    return_dict_in_generate=True,
    output_scores=True,
    max_new_tokens=512,
)
print("Response : ")
for s in generation_output.sequences:
    print(tokenizer.decode(s).split("### Response:")[1])

Note :

Due to the high loss and lack of compute unit, we will update this model frequently to ensure the quality of generated text