# Prism: Code Education Data Processing

📑 [Paper](https://huggingface.co/papers/yyyy.yyyyy) | 🌐 [Project Page](https://prismml.github.io/) | 💾 [Released Resources](https://huggingface.co/collections/toolevalxm/prism-67b978e28fd926b56a4f55a3) | 📦 [Repo](https://github.com/xmhtoolathlon/Annoy-DataSync)

This is the resource page for our Prism project resources on Huggingface.

**Dataset**

| Dataset | Link |
|---------|------|
| Prism-CodeEdu-Synth | [🤗](https://huggingface.co/datasets/toolevalxm/Prism-CodeEdu-Synth) |

Please also check the raw data after our processing: [toolevalxm/Prism-CodeEdu-Raw](https://huggingface.co/datasets/toolevalxm/Prism-CodeEdu-Raw).

**Models**

| Base Model / Training | Prism | | Prism++ | |
|----------------------|-------|-------|---------|-------|
| | Stage 1 | Stage 2 | Stage 1 | Stage 2 |
| Qwen 2.5 7B Coder | [🤗](https://huggingface.co/toolevalxm/qwen2.5-7b-coder_prism_stage1) | [🤗](https://huggingface.co/toolevalxm/qwen2.5-7b-coder_prism) | [🤗](https://huggingface.co/toolevalxm/qwen2.5-7b-coder_prism_pp_stage1) | [🤗](https://huggingface.co/toolevalxm/qwen2.5-7b-coder_prism_pp) |

**Introduction**

We use a fully LLM-based approach for synthesizing all the desired responses using Qwen2.5-72B-Instruct, as it provides excellent performance for code understanding tasks at reasonable cost.

*Due to compliance requirements, we only release the CodeEdu subset (this page) of full dataset.

**Commercial Usage**

The commercial usage policy for this dataset is Qwen.

**License**

The license for this dataset is Qwen.