Update README.md
Browse files
README.md
CHANGED
@@ -9,6 +9,46 @@ Obviously, it's still not perfect (I won't lie, the original dataset was very fl
|
|
9 |
|
10 |
If you want to support me, you can [here](https://ko-fi.com/adamcodd).
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
## Acknowledgment and citation
|
13 |
```bibtex
|
14 |
@inproceedings{bien-etal-2020-recipenlg,
|
|
|
9 |
|
10 |
If you want to support me, you can [here](https://ko-fi.com/adamcodd).
|
11 |
|
12 |
+
## Token Distribution Analysis
|
13 |
+
|
14 |
+
We analyzed the token count distribution of the dataset using the LLAMA-3 tokenizer. This information is crucial for understanding the length of prompts and planning model training.
|
15 |
+
|
16 |
+
### Key Statistics
|
17 |
+
- **Minimum tokens**: 164
|
18 |
+
- **Maximum tokens**: 3,285
|
19 |
+
- **Median (50th percentile)**: 274 tokens
|
20 |
+
|
21 |
+
### Decile Distribution
|
22 |
+
|
23 |
+
```ascii
|
24 |
+
| Percentile | Token Count |
|
25 |
+
|------------|-------------|
|
26 |
+
| 10% | 192 |
|
27 |
+
| 20% | 209 |
|
28 |
+
| 30% | 228 |
|
29 |
+
| 40% | 249 |
|
30 |
+
| 50% | 274 |
|
31 |
+
| 60% | 302 |
|
32 |
+
| 70% | 337 |
|
33 |
+
| 80% | 386 |
|
34 |
+
| 90% | 467 |
|
35 |
+
| 100% | 3,285 |
|
36 |
+
```
|
37 |
+
|
38 |
+
### Interpretation
|
39 |
+
|
40 |
+
1. **Range**: The dataset contains prompts ranging from 164 to 3,285 tokens, indicating significant variation in prompt lengths.
|
41 |
+
|
42 |
+
2. **Central Tendency**: The median token count is 274, meaning half of the prompts have 274 tokens or fewer.
|
43 |
+
|
44 |
+
3. **Distribution**:
|
45 |
+
- 90% of prompts have 467 tokens or fewer.
|
46 |
+
- There's a notable jump from the 90th percentile (467 tokens) to the maximum (3,285 tokens), suggesting some outliers with very high token counts.
|
47 |
+
|
48 |
+
4. **Implications for Training**:
|
49 |
+
- A sequence length of 400-500 tokens would cover the majority of prompts.
|
50 |
+
- Special handling may be needed for outliers with high token counts (e.g., truncation or splitting).
|
51 |
+
|
52 |
## Acknowledgment and citation
|
53 |
```bibtex
|
54 |
@inproceedings{bien-etal-2020-recipenlg,
|