| --- |
| license: mit |
| tags: |
| - training-artifacts |
| - checkpoints |
| - logs |
| - causal-lm |
| --- |
| |
|  |
|
|
| # QED-75M Artifacts |
|
|
| This repository stores training artifacts for **QED-75M**. |
|
|
| ## What is inside |
|
|
| - Training checkpoints (`.pt`) |
| - Training/evaluation logs |
| - Auxiliary files used for reproducibility (configs, summaries, intermediate outputs) |
|
|
| ## Related model repository |
|
|
| - Main model card and inference-ready model: [levossadtchi/QED-75M](https://huggingface.co/levossadtchi/QED-75M). It contains a checkpoint at step 7400 SFT step. |
|
|
| ## Training summary |
|
|
| - Pretraining data volume: **12.6B tokens** |
| - Multi-stage pipeline: pretraining -> long-context annealing -> SFT |
|
|
| ## Notes |
|
|
| - These files are intended for reproducibility, inspection, and research workflows. |
| - For normal inference, use the main model repository instead of this artifacts repository. |
| - data/pretokenized contains pretokenized data for stage 1. |
|
|
