File size: 3,960 Bytes
2cfee1a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
eb9164d
2cfee1a
 
eb9164d
2cfee1a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
annotations_creators:
- expert-generated
language_creators:
- expert-generated
language:
- en
license:
- mit
multilinguality:
- monolingual
size_categories:
- n<1K
source_datasets:
- original
task_categories:
- text-generation
- question-answering
pretty_name: Nano-Start Learning Dataset
tags:
- educational
- llm-training
- chat
- completions
- oxidizr
configs:
- config_name: completions
  data_files:
  - split: train
    path: completions.jsonl
- config_name: qa
  data_files:
  - split: train
    path: qa.jsonl
- config_name: chat
  data_files:
  - split: train
    path: chat.jsonl
---

# Nano-Start Learning Dataset

A small educational dataset for learning how to train language models from scratch.

## Dataset Description

This dataset contains simple, factual examples designed to demonstrate LLM training concepts:

- **Completions**: Factual statements the model learns to continue
- **Q&A**: Question-answer pairs using chat special tokens
- **Chat**: Multi-turn conversations with system prompts

The dataset is intentionally small (~276 examples) so models can be trained quickly on CPU. The goal is education, not production-quality models.

## Dataset Statistics

| Split | Examples | Description |
|-------|----------|-------------|
| completions | 129 | Factual statements about geography, math, science, etc. |
| qa | 96 | Q&A pairs with `<\|user\|>` and `<\|assistant\|>` tokens |
| chat | 51 | Multi-turn conversations with `<\|system\|>` prompts |

## Data Format

All files are JSONL (JSON Lines) with a single `text` field:

### Completions
```json
{"text": "The capital of France is Paris. Paris is known for the Eiffel Tower."}
{"text": "1 + 1 = 2. This is the most basic addition problem in mathematics."}
{"text": "Water boils at 100 degrees Celsius at sea level."}
```

### Q&A
```json
{"text": "<|user|>What is 1+1?<|assistant|>1+1 equals 2."}
{"text": "<|user|>What is the capital of France?<|assistant|>The capital of France is Paris."}
```

### Chat
```json
{"text": "<|system|>You are a helpful assistant.<|user|>Hello!<|assistant|>Hello! How can I help you today?"}
{"text": "<|system|>You are a math tutor.<|user|>What is 5x5?<|assistant|>5x5 equals 25."}
```

## Special Tokens

The dataset uses OpenAI-compatible special tokens from the `cl100k_base` vocabulary:

| Token | ID | Purpose |
|-------|------|---------|
| `<\|endoftext\|>` | 100257 | End of document (added during tokenization) |
| `<\|system\|>` | 100277 | System instructions |
| `<\|user\|>` | 100278 | User input |
| `<\|assistant\|>` | 100279 | Model response |

## Usage

### Download

**Option A: Using hf**
```bash
pip install huggingface_hub
hf download fs90/nano-start-data --local-dir raw --repo-type dataset
```

**Option B: Direct download**

Download files from the [Files tab](https://huggingface.co/datasets/fs90/nano-start-data/tree/main).

### View with Python
```python
from datasets import load_dataset

ds = load_dataset("fs90/nano-start-data", "completions")
for example in ds["train"][:3]:
    print(example["text"])
```

### For Training

This raw data shows what the text looks like **before tokenization**. For training, use the pre-tokenized version: [fs90/nano-start-data-bin](https://huggingface.co/datasets/fs90/nano-start-data-bin)

To learn how to tokenize your own data, see the [splintr](https://github.com/farhan-syah/splintr) project.

## Related Resources

- **Pre-tokenized data**: [fs90/nano-start-data-bin](https://huggingface.co/datasets/fs90/nano-start-data-bin)
- **Training framework**: [oxidizr](https://github.com/farhan-syah/oxidizr)
- **Tokenization**: [splintr](https://github.com/farhan-syah/splintr) - Learn how to tokenize your own data

## License

MIT License

## Citation

```bibtex
@dataset{nano_start_2024,
  title={Nano-Start: Educational Dataset for LLM Training},
  author={fs90},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/fs90/nano-start-data}
}
```