Instructions to use readerbench/RoSummary-medium with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use readerbench/RoSummary-medium with Transformers:

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

pipe = pipeline("text-generation", model="readerbench/RoSummary-medium")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("readerbench/RoSummary-medium")
model = AutoModelForCausalLM.from_pretrained("readerbench/RoSummary-medium", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use readerbench/RoSummary-medium with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/readerbench/RoSummary-medium

SGLang

How to use readerbench/RoSummary-medium 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 "readerbench/RoSummary-medium" \
    --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": "readerbench/RoSummary-medium",
		"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 "readerbench/RoSummary-medium" \
        --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": "readerbench/RoSummary-medium",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use readerbench/RoSummary-medium with Docker Model Runner:
```
docker model run hf.co/readerbench/RoSummary-medium
```

RoSummary-medium

File size: 1,844 Bytes

Model card for RoSummary-medium

---
language:
- ro
---

# RoSummary

This is a version of the RoGPT2 model trained on the [AlephNews](https://huggingface.co/datasets/readerbench/AlephNews) dataset for the summarization task. There are 3 trained versions, they are available on the HuggingFace Hub:
* [base](https://huggingface.co/readerbench/RoSummary-base)
* [medium](https://huggingface.co/readerbench/RoSummary-medium)
* [large](https://huggingface.co/readerbench/RoSummary-large)


## Evaluation on [AlephNews](https://huggingface.co/datasets/readerbench/AlephNews)


| Model  | Decode Method  |           | BERTScore |          |          |   ROUGE  |          |
|:------:|:--------------:|:---------:|:---------:|:--------:|:--------:|:--------:|:--------:|
| 	 |                | Precision |   Recall  | F1-Score |  ROUGE-1 |  ROUGE-2 |  ROUGE-L |
| 	 |	Greedy	  |   0.7335  |   0.7399  |  0.7358  |  0.3360  |  0.1862  |  0.3333  |
| Base   |  Beam Search	  |   0.7354  |   0.7468  |  0.7404  |  0.3480  |  0.1991  |  0.3416  |
|	 | Top-p Sampling |   0.7296  |   0.7299  |  0.7292  |  0.3058  |  0.1452  |  0.2951  |
|	 |	Greedy	  |   0.7378  |   0.7401  |  0.7380  |  0.3422  |  0.1922  |  0.3394  |
| Medium |  Beam Search	  |   0.7390  | **0.7493**|**0.7434**|**0.3546**|**0.2061**|**0.3467**|
|	 | Top-p Sampling |   0.7315  |   0.7285  |  0.7294  |  0.3042  |  0.1400  |  0.2921  |
|	 |	Greedy	  |   0.7376  |   0.7424  |  0.7391  |  0.3414  |  0.1895  |  0.3355  |
| Large	 |  Beam Search	  | **0.7394**|   0.7470  |  0.7424  |  0.3492  |  0.1995  |  0.3384  |
|	 | Top-p Sampling |   0.7311  |   0.7301  |  0.7299  |  0.3051  |  0.1418  |  0.2931  |


## Acknowledgments

---

Research supported with [Cloud TPUs](https://cloud.google.com/tpu/) from Google's [TensorFlow Research Cloud (TFRC)](https://www.tensorflow.org/tfrc)