---
language:
- el
- en
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
base_model: glossAPI/apertus-8b-greek-cpt
datasets:
- glossAPI/apertus-8b-greek-cpt-modern-greek-train
- glossAPI/greek-apertus-post-training-data
tags:
- apertus
- greek
- glossapi
- instruct
- public-three-benchmarks
model-index:
- name: Greek Apertus 8B — Greek maths tuning
  results:
  - task:
      type: text-generation
    dataset:
      name: GreekMMLU
      type: greekmmlu
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 70.44
  - task:
      type: text-generation
    dataset:
      name: Greek IFEval
      type: ifeval-el
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 68.02
  - task:
      type: text-generation
    dataset:
      name: Greek GSM8K
      type: gsm8k-el
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 61.49
---

# Greek Apertus 8B — Greek maths tuning

5,482 Greek maths rows and 4,496 replay rows. Three epochs, learning rate 5e-6, sequence length 4,096.

## Benchmarks

Accuracy (%).

| GreekMMLU | Greek IFEval | Greek GSM8K |
|---:|---:|---:|
| 70.44 | 68.02 | 61.49 |

## Repositories and checkpoints

| Resource | Links |
|---|---|
| Instruction models | [Parent — Stage 1](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage1-sft-S1), [Next stage — Conversation](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage3-sft-chat-P), [Repository](https://huggingface.co/glossAPI/greek-apertus-8b-instruct) |
| Post-training data | [Training data — Greek maths](https://huggingface.co/datasets/glossAPI/greek-apertus-post-training-data/tree/main/second_stage_sft/stage2-sft-recipe3), [Repository](https://huggingface.co/datasets/glossAPI/greek-apertus-post-training-data) |
| Greek base | [base-avg30B-50B](https://huggingface.co/glossAPI/apertus-8b-greek-cpt/tree/base-avg30B-50B) |
| Pre-training data | [Repository](https://huggingface.co/datasets/glossAPI/apertus-8b-greek-cpt-modern-greek-train) |
| SFT data | [Repository](https://huggingface.co/datasets/glossAPI/greek-apertus-sft-data) |
| Tokenizer | [Repository](https://huggingface.co/fffoivos/apertus-tokenizer-extension) |

## Licence

Apache-2.0 for the model weights.

## Acknowledgements

This work was implemented thanks to a grant by Swiss AI for compute on CSCS.