| --- |
| license: apache-2.0 |
| task_categories: |
| - text-generation |
| - question-answering |
| language: |
| - en |
| tags: |
| - finemath |
| - fineweb |
| - edu |
| - pretrain |
| - pretraining |
| - web |
| - reason |
| - reasoning |
| - dense |
| - fine |
| - fine data |
| - ultra |
| - pro |
| - finemath 3plus |
| - finemath 4plus |
| - thought |
| - thoughts |
| - think |
| - thinking |
| - post train |
| - posttrain |
| size_categories: |
| - 100K<n<1M |
| --- |
| # DCLM Ultra |
|
|
| **WebMathUltra** is a heavily filtered and deduplicated, improved version of **finemath-3plus**. |
|
|
| It achieves **superior performance** on math benchmarks including arithmark 3.0, and achieving better math loss. |
|
|
| **9M model trained on only finemath-3plus vs WebMathUltra, same architecture, 500 millon tokens.** |
|
|
| | Metric | finemath 3+ | WebMathUltra | |
| |---|---:|---:| |
| | Math loss | **2.93** | **2.93** | |
| | Arithmark 3.0 | 35.30% | **38.70%** | |
|
|
| ## Dataset Details |
|
|
| - **Curated by:** AInion |
| - **Funded by:** ArrowAI |
| - **Shared by:** MoreThought |
| - **Language:** english |
| - **License:** apache 2.0 |
|
|
| ### Dataset Sources |
|
|
| - **Main Source:** https://huggingface.co/datasets/HuggingFaceTB/finemath |
| - **Ultra Source:** https://huggingface.co/datasets/Geralt-Targaryen/finemath |
|
|
| ## Uses |
|
|
| - PreTraining SLMs on clean, heavily filtered math. |
| - Continued PreTraining SLMs on clean, heavily filtered math. |
| - PostTraining LLMs on clean, heavily filtered math. |
