Opening paragraph. Say the result first, then explain how you got there — people reading a training writeup want the number before the method.
Setup
Body text with a link, some inline_code, and whatever else the post needs.
- Base model, parameter count, tokenizer.
- Token budget and hardware.
- Anything that would let someone reproduce this.
Results
| Model | Params | ARC-e | HellaSwag |
|---|---|---|---|
| Baseline | 2.9M | 31.4 | 26.8 |
| This run | 2.9M | 34.1 | 27.9 |
Use a pull quote for the one claim you want people to remember.
Config
d_model: 256
n_layers: 8
n_heads: 4
tokens: 6B
What's next
Closing paragraph: what you'd change, what's already running, where the weights are.
