π¦’ SmollStories-5M
Part of the SmollStories family β released during the Mayo 2026 Tramo by Tralalabs π΄
Specs
|
|
| Total params |
4,851,968 (4.85M) |
| Architecture |
GPT-style decoder-only |
| Layers |
4 |
| Heads |
4 |
| Hidden dim |
256 |
| Context length |
512 |
| Vocab size |
6144 (custom BPE) |
| Final loss |
2.810 |
Training data
Mixed 1:1:1 from three children's story datasets:
- π ajibawa-2023/Children-Stories-Collection
- π― SimpleStories/SimpleStories
- π£ roneneldan/TinyStories
## Family
The full SmollStories lineup (Mayo 2026):
- π₯ SmollStories-1K
- π± SmollStories-10K
- π£ SmollStories-100K
- π₯ SmollStories-500K
- π¦ SmollStories-1M
- π¦’ SmollStories-5M (you are here)
- π¦
SmollStories-15M
## License
MIT
π΄ Mayo 2026 Tramo Release β Tralalabs