PEFT
PyTorch
gpt_bigcode
code
File size: 4,755 Bytes
9d48ec9
 
 
 
 
 
 
 
 
 
 
30a2056
9d48ec9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
30a2056
9d48ec9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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

---
license: bigcode-openrail-m
datasets:
- bigcode/guanaco-commits
metrics:
- code_eval
library_name: peft
tags:
- code
---
# Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models
<p align="center" width="100%">
<a ><img src="https://github.com/bigcode-project/astraios/blob/main/visuals/banner.png?raw=true" alt="Astraios" style="width: 20%; min-width: 300px; display: block; margin: auto;"></a>
</p>

# Table of Contents

1. [Model Summary](#model-summary)
2. [Use](#use)
3. [Training](#training)
4. [Citation](#citation)

# Model Summary

> Astraios-FFT is an instruction tuned model with 15.5B parameters created by finetuning StarCoderBase on CommitPackFT & OASST as described in the Astraios paper.

- **Repository:** [bigcode-project/astraios](https://github.com/bigcode-project/astraios)
- **Paper:** [Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models]()
- **Languages:** 80+ Programming languages
- **✨Astraios:**
<table>
<tr>
<th>Data</t> 
<td><a href=https://huggingface.co/datasets/bigcode/guanaco-commits>CommitPackFT+OASST</a></td>
<td>Filtered version of CommitPack and OASST for high-quality commit messages that resemble instructions</td>
</tr>
<tr>
<th>Model</t> 
<td><a href=https://huggingface.co/collections/bigcode/astraios-1b-6576ff1b8e449026ae327c1c>Astraios-1B</a></td>
<td>Collection of StarCoderBase-1B models instruction tuned on CommitPackFT + OASST with different tuning methods</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/collections/bigcode/astraios-3b-6577127317ee44ff547252d3>Astraios-3B</a></td>
<td>Collection of StarCoderBase-3B (3B parameters) models instruction tuned on CommitPackFT + OASST with different tuning methods</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/collections/starpeft/starcoderbase-7b-650c1f028b45cfec8e72c265>Astraios-7B</a></td>
<td>Collection of StarCoderBase-7B (7B parameters) models instruction tuned on CommitPackFT + OASST with different tuning methods</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/collections/bigcode/astraios-16b-65788b7476b6de79781054cc>Astraios-16B</a></td>
<td>Collection of StarCoderBase-16B (16B parameters) models instruction tuned on CommitPackFT + OASST with different tuning methods</td>
</tr>
<tr>
<th>Evaluation</t> 
<td><a href=https://huggingface.co/datasets/code_x_glue_cc_clone_detection_big_clone_bench>BigCloneBench</a></td>
<td>Dataset for clone detection; We use 2,000 samples for evaluation</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/datasets/code_x_glue_cc_defect_detection>Devign</a></td>
<td>Dataset for defect detection; We use 2,000 samples for evaluation</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/datasets/bigcode/humanevalpack>HumanEvalPack</a></td>
<td>Extension of OpenAI's HumanEval to cover 3 scenarios across 6 languages</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/datasets/RaymondLi/perturbed_humaneval>ReCode</a></td>
<td>Dataset for the robustness of code generation, covering 4 variants</td>
</tr>
<tr>
<th></t> 
<td><a href=https://huggingface.co/datasets/moyix/asleep_keyboard>Asleep At The Keyboard</a></td>
<td>Datasets for security of code generation; We use DoW for evaluation</td>
</tr>
</table>


# Use

## Intended use

The model follows instructions provided in the input. You should always preface your input with "Question: " and finish it with "Answer:", for example: "Question: Please write a function in Python that performs bubble sort.

Answer:"

**Feel free to share your generations in the Community tab!**

## Generation
```python
# pip install -q transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

checkpoint = "bigcode/astraios-fft"
model = AutoModelForCausalLM.from_pretrained(checkpoint)
device = "cuda" # for GPU usage or "cpu" for CPU usage

tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)

inputs = tokenizer.encode("Question: Please write a function in Python that performs bubble sort.

Answer:", return_tensors="pt").to(device)
outputs = model.generate(inputs)
print(tokenizer.decode(outputs[0]))
```

# Training

## Model

- **Architecture:** GPT-2 model with multi-query attention and Fill-in-the-Middle objective
- **Steps:** 250k pretraining & 200 instruction tuning
- **Precision:** fp32

## Hardware

- **Pretraining:**
  - **GPUs:** 512 Tesla A100
  - **Training time:** 24 days
- **Instruction tuning:**
  - **GPUs:** 8 Tesla A100

## Software

- **Orchestration:** [Megatron-LM/Transformers](https://github.com/bigcode-project/octopack#training)
- **Neural networks:** [PyTorch](https://github.com/pytorch/pytorch)

# Citation

```bibtex
```