Instructions to use glossAPI/apertus-8b-greek-cpt with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use glossAPI/apertus-8b-greek-cpt with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="glossAPI/apertus-8b-greek-cpt")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("glossAPI/apertus-8b-greek-cpt")
model = AutoModelForCausalLM.from_pretrained("glossAPI/apertus-8b-greek-cpt", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use glossAPI/apertus-8b-greek-cpt with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "glossAPI/apertus-8b-greek-cpt"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "glossAPI/apertus-8b-greek-cpt",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/glossAPI/apertus-8b-greek-cpt

SGLang

How to use glossAPI/apertus-8b-greek-cpt with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "glossAPI/apertus-8b-greek-cpt" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "glossAPI/apertus-8b-greek-cpt",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "glossAPI/apertus-8b-greek-cpt" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "glossAPI/apertus-8b-greek-cpt",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use glossAPI/apertus-8b-greek-cpt with Docker Model Runner:
```
docker model run hf.co/glossAPI/apertus-8b-greek-cpt
```

apertus-8b-greek-cpt / README.md

fffoivos

release: label Greek Apertus v0.5

09eb891 verified 9 days ago

preview code

Raw History Blame Contribute Delete

5.54 kB

metadata

language: el
license: apache-2.0
library_name: transformers
pipeline_tag: text-generation
base_model: swiss-ai/Apertus-8B-2509
datasets:
  - glossAPI/apertus-8b-greek-cpt-modern-greek-train
tags:
  - apertus
  - greek
  - glossapi
  - continued-pretraining
model-index:
  - name: apertus-8b-greek-cpt main
    results:
      - task:
          type: text-generation
        dataset:
          name: GreekMMLU (decontaminated, n=16,159)
          type: dascim/GreekMMLU
        metrics:
          - type: accuracy
            value: 54.85
            verified: false
      - task:
          type: text-generation
        dataset:
          name: ASEP MCQA (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 55.08
            verified: false
      - task:
          type: text-generation
        dataset:
          name: DemosQA (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 46.58
            verified: false
      - task:
          type: text-generation
        dataset:
          name: GPCR (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 62.89
            verified: false
      - task:
          type: text-generation
        dataset:
          name: Medical MCQA (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 38.42
            verified: false
      - task:
          type: text-generation
        dataset:
          name: OYXOY metaphor (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 33.89
            verified: false
      - task:
          type: text-generation
        dataset:
          name: OYXOY NLI (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 38.73
            verified: false
      - task:
          type: text-generation
        dataset:
          name: OYXOY WiC (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 33.64
            verified: false
      - task:
          type: text-generation
        dataset:
          name: OYXOY WSD (strict contamination-filtered)
          type: fffoivos/native-greek-suite
        metrics:
          - type: accuracy
            value: 38.06
            verified: false
      - task:
          type: text-generation
        dataset:
          name: Apertus Table-14 retention macro
          type: apertus-pretraining-suite
        metrics:
          - type: accuracy
            value: 62.95
            verified: false

Apertus 8B Greek CPT — full checkpoint trajectory

Continued pretraining on 76.685B active tokens. Base models, not instruction tuned.

Training

79% Modern Greek, 20% foreign replay, 1% Old Greek. Extended vocabulary: 148,992 tokens. 18 training checkpoints and two checkpoint averages. main: terminal checkpoint.

Recommended checkpoint

base-avg30B-50B: GreekMMLU 56.78% (best single checkpoint: 56.81%). Apertus pretraining-suite macro: 64.58% (terminal: 62.95%).

Repositories and checkpoints

Resource	Links
Original Apertus	Repository, Revision
Instruction models	Repository, Stage 1, Greek maths, Conversation
Greek base	base-avg30B-50B, 17-step18284-tokens77B, base-avg30B-50B, CHECKPOINTS.md, checkpoint-index.json
Pre-training data	Repository, Revision
SFT data	Stage 1, Repository
Post-training data	Greek maths, Conversation, Repository
Tokenizer	Revision, Repository
Benchmark audit	Repository

Acknowledgements

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

Collection: Greek Apertus 8B.