README.md

---
license: bigscience-bloom-rail-1.0
language:
- es
- en
pipeline_tag: text-generation
library_name: transformers
tags:
- alpaca
- bloom
- LLM
datasets:
- tatsu-lab/alpaca
inference: false
widget:
- text: "Below is an instruction that describes a task, paired with an input that provides further context.\nWrite a response that appropriately completes the request.\n### Instruction:\nTell me about alpacas"
---

<div style="text-align:center;width:250px;height:250px;">
    <img src="https://huggingface.co/platzi/chivoom/resolve/main/chivoom_logo-removebg-preview.png">
</div>



# Chivoom: Spanish Alpaca (Chiva) 🐐 + BLOOM 💮

# IMPORTANT: This is just a PoC and still WIP!


## Adapter Description
This adapter was created with the [PEFT](https://github.com/huggingface/peft) library and allowed the base model **BigScience/BLOOM 7B1** to be fine-tuned on the **Stanford's Alpaca Dataset** (translated to Spanish) by using the method **LoRA**.

## Model Description
BigScience Large Open-science Open-access Multilingual Language Model

[BLOOM 7B1](https://huggingface.co/bigscience/bloom-7b1)

## Training data

We translated to Spanish the Alpaca dataset.

Alpaca is a dataset of **52,000** instructions and demonstrations generated by OpenAI's `text-davinci-003` engine. This instruction data can be used to conduct instruction-tuning for language models and make the language model follow instruction better.

The authors built on the data generation pipeline from [Self-Instruct framework](https://github.com/yizhongw/self-instruct) and made the following modifications:

- The `text-davinci-003` engine to generate the instruction data instead of `davinci`.
- A [new prompt](https://github.com/tatsu-lab/stanford_alpaca/blob/main/prompt.txt) was written that explicitly gave the requirement of instruction generation to `text-davinci-003`.
- Much more aggressive batch decoding was used, i.e., generating 20 instructions at once, which significantly reduced the cost of data generation.
- The data generation pipeline was simplified by discarding the difference between classification and non-classification instructions.
- Only a single instance was generated for each instruction, instead of 2 to 3 instances as in Self-Instruct.

This produced an instruction-following dataset with 52K examples obtained at a much lower cost (less than $500). 
In a preliminary study, the authors also found that the 52K generated data to be much more diverse than the data released by [Self-Instruct](https://github.com/yizhongw/self-instruct/blob/main/data/seed_tasks.jsonl).

### Training procedure

TBA

## How to use
```py
import torch
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig

peft_model_id = "platzi/chivoom"
config = PeftConfig.from_pretrained(peft_model_id)
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, return_dict=True, load_in_8bit=True, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("bigscience/bloom-7b1")

model = PeftModel.from_pretrained(model, peft_model_id)
model.eval()

# Based on the inference code by `tloen/alpaca-lora`
def generate_prompt(instruction, input=None):
    if input:
        return f"""A continuación se muestra una instrucción que describe una tarea, emparejada con una entrada que proporciona más contexto. Escribe una respuesta que complete adecuadamente la petición.
### Instrucción:
{instruction}
### Entrada:
{input}
### Respuesta:"""
    else:
        return f"""A continuación se muestra una instrucción que describe una tarea. Escribe una respuesta que complete adecuadamente la petición.
### Instrucción:
{instruction}
### Respuesta:"""

def generate(
        instruction,
        input=None,
        temperature=0.1,
        top_p=0.75,
        top_k=40,
        num_beams=4,
        **kwargs,
):
    prompt = generate_prompt(instruction, input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].cuda()
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=256,
        )
    s = generation_output.sequences[0]
    output = tokenizer.decode(s)
    return output.split("### Response:")[1]

instruction = "¿Qué es un chivo?"

print("Instrucción:", instruction)
print("Respuesta:", generate(instruction))
```