---
license: bigscience-bloom-rail-1.0
tags:
- generated_from_trainer
model-index:
- name: Bloom-Alpaca-560m
  results: []
widget:
- text: |-
    ### Instruction:
    Write a poem about pancakes.

    ### Response:
datasets:
- tatsu-lab/alpaca
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# Bloom-Alpaca-560m

This model is a fine-tuned version of [bigscience/bloom-560m](https://huggingface.co/bigscience/bloom-560m) on the [tatsu-lab/alpaca](https://huggingface.co/datasets/tatsu-lab/alpaca) dataset. This is a test to check if the concept of Databrick's [databricks/dolly-v1-6b](https://huggingface.co/databricks/dolly-v1-6b) can be applied to smaller models.

## Model description

This is an LLM trained to follow instructions, similar to ChatGPT, except much much smaller.

## Sample Use

```python
from transformers import pipeline

generator = pipeline('text-generation', model='linkanjarad/Bloom-Alpaca-560m')

prompt = '''### Instruction:
Who is the first person on the moon?

### Response:'''

output = generator(prompt, do_sample=True, min_length=10, max_length=50, temperature=0.5)
print(output[0]['generated_text'])
```

## Intended uses & limitations

The model works best when given input in the format of the dataset it was trained on. Here is an example of the format of the [tatsu-lab/alpaca](https://huggingface.co/datasets/tatsu-lab/alpaca) dataset:

```
### Instruction:
What are the three primary colors?

### Response:
The three primary colors are red, blue, and yellow.
```

Here are a few sample model outputs:

```
### Instruction:
Write a poem about pancakes in the style of Dr. Suess.

### Response:
Gently bake the delicious cup of pancake goodness away
In the sweet light of a sunken window
Stay away from those who take it for granted
The magic of those who will bake and preserve
The sweetness of baked pancakes.
Once more I'm gonna bake my dreams for ever
No matter what my journey might bring
No matter how hard I come
```

```
### Instruction:
Generate a program in Python to generate prime numbers.

### Response:
"""
Generate a program in Python to generate prime numbers.
"""

import random
import math
import math.exp

def generate_primes():
    m = random.randint(1, 10)
    result = random.randint(1, 10)
    
    for i in range(m):
        result = generate_primes()
        if result == result:
            break
    return result
    
```

```
### Instruction:
How was the steam engine invented?

### Response:
The steam engine was invented by William George Taylor, a French engineer and inventor who used the technology of steam turbines to create engines. Taylor first designed a turbine that used a rotating disc with a spiral gear that accelerated the air through an annular passage. By expanding the annular passage, Taylor's turbine gained power and became a powerful engine. Taylor then used the spiral gear to create a turbine that drove a rail with a
```

```
### Instruction:
Generate a list of marketing strategies to promote a new mobile app.

### Response:
1. Create a campaign to promote the app's launch.
2. Provide a promotional video and posters.
3. Create a mobile app store campaign.
4. Publish a newsletter to attract potential customers.
5. Make a list of key features of the app.
6. Establish a referral program.
7. Attend marketing conferences and events.
8. Participate in customer surveys.
9. Offer loyalty rewards and discounts.
10. Utilize social media to promote the app.
11. Create a user-friendly advertising campaign.
12. Utilize social media to promote the app.
```

## Training Procedure

The model was trained with Huggingface's 🤗 Trainer for about 1 epoch.

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 128
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.15
- num_epochs: 1

### Framework versions

- Transformers 4.27.4
- Pytorch 1.13.0
- Datasets 2.1.0
- Tokenizers 0.13.2