|
# Dataset description |
|
|
|
This release incorporates the complete data sequence used in our CrystalCoder training, covering data sequences from the three pre-training stages. The data is a combination of two previous works: [SlimPajama dataset](https://huggingface.co/datasets/cerebras/SlimPajama-627B) and [StarCoder](https://huggingface.co/datasets/bigcode/starcoderdata). We divide them across three stages with different weights. |
|
|
|
## Stage 1 |
|
During this stage, we utilize half of the [SlimPajama data](https://huggingface.co/datasets/cerebras/SlimPajama-627B). |
|
|
|
## Stage 2 |
|
In the second stage, the remaining half of the [SlimPajama data](https://huggingface.co/datasets/cerebras/SlimPajama-627B) is employed, along with two epochs of [StarCoder data](https://huggingface.co/datasets/bigcode/starcoderdata). For StarCoder data, we implement [FIM augmentation](https://arxiv.org/abs/2207.14255) with an FIM rate of 0.9. |
|
|
|
## Stage 3 |
|
The third stage involves the reuse of Python, HTML, CSS, and JavaScript data from the [StarCoder data](https://huggingface.co/datasets/bigcode/starcoderdata) for training over three epochs. In this stage, FIM with a rate of 0.3 is applied. Additionally, a small portion of the SlimPajama dataset except the Github part is also reused. |
|
|
|
# Primary usage |
|
|
|
This dataset is used for our training of our CrystalCoder, and also for further reproduction. |
|
|
|
# License: |