| --- |
| license: mit |
| --- |
| # Reflection Model Outputs |
|
|
| This repository contains model output results from various LLMs across multiple tasks and configurations. |
|
|
| ## 📂 Dataset Structure |
|
|
| We have 3 runs of data, and all files are organized under the main directory: |
| - `EssentialAI/reflection_model_outputs_run1/` |
| - `EssentialAI/reflection_model_outputs_run2/` |
| - `EssentialAI/reflection_model_outputs_run3/` |
|
|
| Within this, you will find results grouped by **model architecture and checkpoint size**, including: |
|
|
| - `OLMo-2 7B` |
| - `OLMo-2 13B` |
| - `OLMo-2 32B` |
| - `Qwen2.5 0.5B` |
| - `Qwen2.5 3B` |
| - `Qwen2.5 7B` |
| - `Qwen2.5 14B` |
| - `Qwen2.5 32B` |
| - `Qwen2.5 72B` |
|
|
| Each model folder contains outputs from multiple task setups. |
|
|
| Example task folders include: |
|
|
| | Original Task Folder | Adversarial Task Folder | |
| |---------------------------|-----------------------------| |
| | `bbh_cot_fewshot` | `bbh_adv` | |
| | `cruxeval_i` | `cruxeval_i_adv` | |
| | `cruxeval_o` | `cruxeval_o_adv` | |
| | `gsm8k-platinum_cot` | `gsm8k-platinum_adv` | |
| | `gsm8k_cot` | `gsm8k_adv` | |
| | `triviaqa` | `triviaqa_adv` | |
|
|
| --- |
|
|
| ## 💾 How to Download the Dataset |
|
|
| To download the entire dataset locally using the Hugging Face Hub, run the following Python snippet with the run results you're interested in: |
|
|
| ```python |
| from huggingface_hub import snapshot_download |
| snapshot_download( |
| repo_id="EssentialAI/reflection_model_outputs_run1", |
| repo_type="dataset", |
| local_dir="reflection_model_outputs_run1" |
| ) |
| ``` |
