---
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
- glossAPI/greek-apertus-sft-data
tags:
- apertus
- greek
- glossapi
- instruct
- sft
- chat
- v0.5-preview
- card-eval-20261001
- public-three-benchmarks
model-index:
- name: Greek Apertus 8B Instruct
  results:
  - task:
      type: text-generation
    dataset:
      name: GreekMMLU (official, no chat template)
      type: greekmmlu-official-no-chat-template
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 70.52
  - task:
      type: text-generation
    dataset:
      name: IFEval-el (prompt-strict)
      type: ifeval-el-prompt-strict
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 65.8
  - task:
      type: text-generation
    dataset:
      name: GSM8K-el
      type: gsm8k-el
    metrics:
    - type: accuracy
      name: accuracy (%)
      value: 59.67
---

# Greek Apertus 8B Instruct v0.5

Greek Apertus is an 8B instruction model based on [Apertus 8B](https://huggingface.co/swiss-ai/Apertus-8B-2509).

Greek continued pre-training followed by general instruction, Greek maths and conversation fine-tuning. Greek-extended vocabulary: 148,992 tokens.

## Benchmarks

Accuracy (%).

| Model | GreekMMLU | Greek IFEval | Greek GSM8K |
|---|---:|---:|---:|
| [Krikri 1.0](https://huggingface.co/ilsp/Llama-Krikri-8B-Instruct) | 67.94 | 63.77 | 71.72 |
| [Krikri 1.5](https://huggingface.co/ilsp/Llama-Krikri-8B-Instruct-v1.5) | 67.69 | 54.90 | **73.01** |
| [Apertus Instruct](https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509) | 66.50 | 54.90 | 56.18 |
| [Greek Apertus — general instruction tuning](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage1-sft-S1) | **69.95** | **67.28** | 54.21 |
| **Greek Apertus Instruct** | **70.52** | **65.80** | 59.67 |

## Training

[Greek continued pre-training](https://huggingface.co/glossAPI/apertus-8b-greek-cpt/tree/base-avg30B-50B).

| Stage | Checkpoint | Data |
|---|---|---|
| 1 | [Stage 1](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage1-sft-S1) | [Supervised mixtures](https://huggingface.co/datasets/glossAPI/greek-apertus-sft-data/tree/main/stage1-sft-S1) |
| 2 | [Stage 2](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage2-sft-recipe3) | [Greek maths](https://huggingface.co/datasets/glossAPI/greek-apertus-post-training-data/tree/main/second_stage_sft/stage2-sft-recipe3) |
| 3 | [Stage 3](https://huggingface.co/glossAPI/greek-apertus-8b-instruct/tree/stage3-sft-chat-P) | [Greek conversations](https://huggingface.co/datasets/glossAPI/greek-apertus-post-training-data/tree/main/second_stage_sft/stage3-sft-chat-P) |

## Repositories and checkpoints

| Resource | Links |
|---|---|
| Greek base | [Repository](https://huggingface.co/glossAPI/apertus-8b-greek-cpt) |
| 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) |
| Post-training data | [Repository](https://huggingface.co/datasets/glossAPI/greek-apertus-post-training-data) |
| Original Apertus | [Repository](https://huggingface.co/swiss-ai/Apertus-8B-2509) |
| 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.

Collection: [Greek Apertus 8B](https://huggingface.co/collections/glossAPI/greek-apertus-8b-v05-preview-6abecae3e0362dff3324b647).