File size: 7,359 Bytes
df5c880
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
20ef17e
df5c880
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4a52f53
 
 
 
 
 
 
 
 
 
df5c880
b725d89
 
df5c880
 
 
 
 
 
04e7aa8
 
df5c880
 
 
 
 
 
 
 
 
 
 
faa73cd
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
---
license: apache-2.0
datasets:
- nicholasKluge/instruct-aira-dataset
language:
- en
metrics:
- accuracy
library_name: transformers
tags:
- alignment
- instruction tuned
- text generation
- conversation
- assistant
pipeline_tag: text-generation
widget:
- text: "<|startofinstruction|>What is your name?<|endofinstruction|>"
  example_title: Greetings
- text: "<|startofinstruction|>Can you explain what is Machine Learning?<|endofinstruction|>"
  example_title: Machine Learning
- text: "<|startofinstruction|>Do you know anything about virtue ethics?<|endofinstruction|>"
  example_title: Ethics
- text: "<|startofinstruction|>How can I make my girlfriend happy?<|endofinstruction|>"
  example_title: Advise
inference:
  parameters:
    repetition_penalty: 1.2
    temperature: 0.2
    top_k: 30
    top_p: 0.3
    max_length: 200
    length_penalty: 0.3
    early_stopping: true
co2_eq_emissions:
  emissions: 0.77
  source: CodeCarbon
  training_type: fine-tuning
  geographical_location: United States of America
  hardware_used: NVIDIA A100-SXM4-40GB
---
# Aira-2-774M

`Aira-2` is the second version of the Aira instruction-tuned series. `Aira-2-774M` is an instruction-tuned GPT-style model based on [GPT-2](https://huggingface.co/gpt2-large). The model was trained with a dataset composed of prompts and completions generated synthetically by prompting already-tuned models (ChatGPT, Llama, Open-Assistant, etc).

Check our gradio-demo in [Spaces](https://huggingface.co/spaces/nicholasKluge/Aira-Demo).

## Details

- **Size:** 774,032,640 parameters
- **Dataset:** [Instruct-Aira Dataset](https://huggingface.co/datasets/nicholasKluge/instruct-aira-dataset)
- **Language:** English
- **Number of Epochs:** 3
- **Batch size:** 8
- **Optimizer:** `torch.optim.AdamW` (warmup_steps = 1e2, learning_rate = 5e-4, epsilon = 1e-8)
- **GPU:** 1 NVIDIA A100-SXM4-40GB
- **Emissions:** 0.77 KgCO2 (Singapore)
- **Total Energy Consumption:** 1.58 kWh

This repository has the [notebook](AIRA_FineTuning.ipynb) used to train this model.

## Usage

Three special tokens are used to mark the user side of the interaction and the model's response:

`<|startofinstruction|>`What is a language model?`<|endofinstruction|>`A language model is a probability distribution over a vocabulary.`<|endofcompletion|>`

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

tokenizer = AutoTokenizer.from_pretrained('nicholasKluge/Aira-2-774M')
aira = AutoModelForCausalLM.from_pretrained('nicholasKluge/Aira-2-774M')

aira.eval()
aira.to(device)

question =  input("Enter your question: ")

inputs = tokenizer(tokenizer.bos_token + question + tokenizer.sep_token, return_tensors="pt").to(device)

responses = aira.generate(**inputs,
	bos_token_id=tokenizer.bos_token_id,
	pad_token_id=tokenizer.pad_token_id,
	eos_token_id=tokenizer.eos_token_id,
	do_sample=True,
	top_k=50,
	max_length=500,
	top_p=0.95,
	temperature=0.7,
	num_return_sequences=2)

print(f"Question: 👤 {question}\n")

for i, response in  enumerate(responses):
	print(f'Response {i+1}: 🤖 {tokenizer.decode(response, skip_special_tokens=True).replace(question, "")}')
```

The model will output something like:

```markdown
>>>Question: 👤 What is the capital of Brazil?

>>>Response 1: 🤖 The capital of Brazil is Brasília.
>>>Response 2: 🤖 The capital of Brazil is Brasília.
```

## Limitations

🤥 Generative models can perpetuate the generation of pseudo-informative content, that is, false information that may appear truthful.

🤬 In certain types of tasks, generative models can produce harmful and discriminatory content inspired by historical stereotypes.

## Evaluation

| Model (GPT-2)                                                   | Average   | [ARC](https://arxiv.org/abs/1803.05457) | [TruthfulQA](https://arxiv.org/abs/2109.07958) | [ToxiGen](https://arxiv.org/abs/2203.09509) |   |   |
|-----------------------------------------------------------------|-----------|-----------------------------------------|------------------------------------------------|---------------------------------------------|---|---|
| [Aira-2-124M](https://huggingface.co/nicholasKluge/Aira-2-124M) | **38.07** | **24.57**                               | **41.02**                                      | **48.62**                                   |   |   |
| GPT-2                                                           | 35.37     | 21.84                                   | 40.67                                          | 43.62                                       |   |   |
| [Aira-2-355M](https://huggingface.co/nicholasKluge/Aira-2-355M) | **39.68** | **27.56**                               | 38.53                                          | **53.19**                                   |   |   |
| GPT-2-medium                                                    | 36.43     | 27.05                                   | **40.76**                                      | 41.49                                       |   |   |
| [Aira-2-774M](https://huggingface.co/nicholasKluge/Aira-2-774M) | **42.26** | **28.75**                               | **41.33**                                      | **56.70**                                   |   |   |
| GPT-2-large                                                     | 35.16     | 25.94                                   | 38.71                                          | 40.85                                       |   |   |
| [Aira-2-1B5](https://huggingface.co/nicholasKluge/Aira-2-1B5)   | **42.22** | 28.92                                   | **41.16**                                      | **56.60**                                   |   |   |
| GPT-2-xl                                                        | 36.84     | **30.29**                               | 38.54                                          | 41.70                                       |   |   |

* Evaluations were performed using the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness) (by [EleutherAI](https://www.eleuther.ai/)). The notebook used to make these evaluations is available in the [this repo](lm_evaluation_harness.ipynb).

## Cite as 🤗

```latex

@misc{nicholas22aira,
  doi = {10.5281/zenodo.6989727},
  url = {https://huggingface.co/nicholasKluge/Aira-2-774M},
  author = {Nicholas Kluge Corrêa},
  title = {Aira},
  year = {2023},
  publisher = {HuggingFace},
  journal = {HuggingFace repository},
}

```

## License

The `Aira-2-774M` is licensed under the Apache License, Version 2.0. See the [LICENSE](LICENSE) file for more details.

# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_nicholasKluge__Aira-2-774M)

| Metric                | Value                     |
|-----------------------|---------------------------|
| Avg.                  | 27.47   |
| ARC (25-shot)         | 28.75          |
| HellaSwag (10-shot)   | 40.8    |
| MMLU (5-shot)         | 25.1         |
| TruthfulQA (0-shot)   | 41.33   |
| Winogrande (5-shot)   | 52.01   |
| GSM8K (5-shot)        | 0.0        |
| DROP (3-shot)         | 4.26         |