Instructions to use OpenLLM-Ro/RoLlama3-8b-Instruct-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OpenLLM-Ro/RoLlama3-8b-Instruct-4bit with Transformers:

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

pipe = pipeline("text-generation", model="OpenLLM-Ro/RoLlama3-8b-Instruct-4bit")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenLLM-Ro/RoLlama3-8b-Instruct-4bit")
model = AutoModelForCausalLM.from_pretrained("OpenLLM-Ro/RoLlama3-8b-Instruct-4bit")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use OpenLLM-Ro/RoLlama3-8b-Instruct-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/OpenLLM-Ro/RoLlama3-8b-Instruct-4bit

SGLang

How to use OpenLLM-Ro/RoLlama3-8b-Instruct-4bit 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 "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "OpenLLM-Ro/RoLlama3-8b-Instruct-4bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use OpenLLM-Ro/RoLlama3-8b-Instruct-4bit with Docker Model Runner:
```
docker model run hf.co/OpenLLM-Ro/RoLlama3-8b-Instruct-4bit
```

Model Card for Model ID

This model is a Bits&Bytes 4 bits quantization of the https://huggingface.co/OpenLLM-Ro/RoLlama3-8b-Instruct model.

The main advantages of this model are :

it runs on a GPU with 6GB of free ram. (So usually a user-grade gpu with 8 Gb VRAM, versus the standard model which needs 48+GB).
it is 2-3 times faster in inference time/token

The main drawback is that is less accurate than the full(original) model, although is up to you to decide if the compromise is a good fit for your use-case.

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

OpenLLM-Ro represents the first open-source effort to build a LLM specialized for Romanian. OpenLLM-Ro developed and publicly releases a collection of Romanian LLMs, both in the form of foundational model and instruct and chat variants.

Developed by: OpenLLM-Ro Language(s): Romanian License: cc-by-nc-4.0 Finetuned from model: Meta-Llama-3-8B-Instruct Trained using: RoAlpaca, RoAlpacaGPT4, RoDolly, RoSelfInstruct, RoNoRobots, RoOrca, RoCamel, RoOpenAssistant, RoUltraChat

Model Sources [optional]

Repository: https://github.com/OpenLLM-Ro/LLaMA-Factory Paper: https://arxiv.org/abs/2406.18266

ntended Use Intended Use Cases RoLlama3 is intented for research use in Romanian. Base models can be adapted for a variety of natural language tasks while instruction and chat tuned models are intended for assistant-like chat.

Out-of-Scope Use Use in any manner that violates the license, any applicable laws or regluations, use in languages other than Romanian.

How to Get Started with the Model Use the code below to get started with the model.

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("OpenLLM-Ro/RoLlama3-8b-Instruct") model = AutoModelForCausalLM.from_pretrained("OpenLLM-Ro/RoLlama3-8b-Instruct")

instruction = "Ce jocuri de societate pot juca cu prietenii mei?" chat = [ {"role": "system", "content": "Ești un asistent folositor, respectuos și onest. Încearcă să ajuți cât mai mult prin informațiile oferite, excluzând răspunsuri toxice, rasiste, sexiste, periculoase și ilegale."}, {"role": "user", "content": instruction}, ] prompt = tokenizer.apply_chat_template(chat, tokenize=False, system_message="")

inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt") outputs = model.generate(input_ids=inputs, max_new_tokens=128) print(tokenizer.decode(outputs[0]))

Academic Benchmarks

Model	Average	ARC	MMLU	Winogrande	Hellaswag	GSM8k	TruthfulQA
RoLlama-3-8B-Instruct-4Bit	NA	40.38	NA	NA	NA	NA	50.93
Llama-3-8B-Instruct	50.62	43.69	52.04	59.33	53.19	43.87	51.59
RoLlama3-8b-Instruct-2024-06-28	50.56	44.70	52.19	67.23	57.69	30.23	51.34
RoLlama3-8b-Instruct-2024-10-09	52.21	47.94	53.50	66.06	59.72	40.16	45.90
RoLlama3-8b-Instruct-DPO-2024-10-09	49.96	46.29	53.29	65.57	58.15	34.77	41.70

Downstream tasks

	LaRoSeDa				WMT
	Few-shot		Finetuned		Few-shot		Finetuned
Model	Binary (Macro F1)	Multiclass (Macro F1)	Binary (Macro F1)	Multiclass (Macro F1)	EN-RO (Bleu)	RO-EN (Bleu)	EN-RO (Bleu)	RO-EN (Bleu)
Llama-3-8B-Instruct	95.88	56.21	98.53	86.19	18.88	30.98	28.02	40.28
RoLlama3-8b-Instruct-2024-06-28	97.52	67.41	94.15	87.13	24.01	27.36	26.53	40.36
RoLlama3-8b-Instruct-2024-10-09	95.58	61.20	96.46	87.26	22.92	24.28	27.31	40.52
RoLlama3-8b-Instruct-DPO-2024-10-09	97.48	54.00	-	-	22.09	23.00	-	-

	XQuAD				STS
	Few-shot		Finetuned		Few-shot		Finetuned
Model	(EM)	(F1)	(EM)	(F1)	(Spearman)	(Pearson)	(Spearman)	(Pearson)
RoLlama-3-8B-Instruct-4Bit - F5 Scores	NA	NA	NA	NA	68.58	NA	70.57	NA
Llama-3-8B-Instruct	39.47	58.67	67.65	82.77	73.04	72.36	83.49	84.06
RoLlama3-8b-Instruct-2024-06-28	39.43	59.50	44.45	59.76	77.20	77.87	85.80	86.05
RoLlama3-8b-Instruct-2024-10-09	18.89	31.79	50.84	65.18	77.60	76.86	86.70	87.09
RoLlama3-8b-Instruct-DPO-2024-10-09	26.05	42.77	-	-	79.64	79.52	-	-

Hardware

Nvidia RTX 4090 16GB, Laptop Version

Software

[More Information Needed]

RoLlama3 Model Family

Model	Link
RoLlama3-8b-Instruct-2024-06-28	link
RoLlama3-8b-Instruct-2024-10-09	link
RoLlama3-8b-Instruct-DPO-2024-10-09	link

Citation

@misc{masala2024vorbecstiromanecsterecipetrain,
      title={"Vorbe\c{s}ti Rom\^ane\c{s}te?" A Recipe to Train Powerful Romanian LLMs with English Instructions}, 
      author={Mihai Masala and Denis C. Ilie-Ablachim and Alexandru Dima and Dragos Corlatescu and Miruna Zavelca and Ovio Olaru and Simina Terian-Dan and Andrei Terian-Dan and Marius Leordeanu and Horia Velicu and Marius Popescu and Mihai Dascalu and Traian Rebedea},
      year={2024},
      eprint={2406.18266},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2406.18266}, 
}

Downloads last month: 113

Safetensors

Model size

8B params

Tensor type

F16

F32

Paper for OpenLLM-Ro/RoLlama3-8b-Instruct-4bit

"Vorbeşti Româneşte?" A Recipe to Train Powerful Romanian LLMs with English Instructions

Paper • 2406.18266 • Published Jun 26, 2024 • 1