glossAPI/apertus-8b-greek-cpt
Text Generation • 8B • Updated • 8
Datasets:
You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This repository holds the complete continued pre-training data of Greek Apertus 8B: the packed training stream, the Modern Greek documents, the raw HPLT slice, the replay reservoir and a build input. Parts contain third-party text under their own licences (HPLT, GlossAPI sources, replay sources); the replay sources grant no redistribution rights. By requesting access you agree to use the data for research and to respect the licences of each part.
Log in or Sign Up to review the conditions and access this dataset content.
Greek Apertus 8B: continued pre-training data
Packed training stream, Modern-Greek documents, raw HPLT slice, replay reservoir and build input.
Contents
| Part | Path |
|---|---|
| Training stream | megatron/, schedules/, sequences/ |
| Documents | data/*.parquet |
| HPLT source | sources/hplt_greek_ge8_no_mt_clean60_wave4/ |
| Replay | replay_reservoir_v1/ |
| Build input | build_inputs/c3_dedup_audit/ |
| Run records | run_records/ |
76,685,490,476 active training tokens.
Repositories and checkpoints
| Resource | Links |
|---|---|
| Greek base | Repository, base-avg30B-50B |
| Pre-training data | Revision |
| Instruction models | Repository |
| SFT data | Repository |
| Post-training data | Repository |
| Tokenizer | Repository |
Licences
Mixed, per part; each part's original card states its own terms. Replay sources are not redistributable beyond the gated access.
Acknowledgements
This work was implemented thanks to a grant by Swiss AI for compute on CSCS.
Collection: Greek Apertus 8B.
- Downloads last month
- 445