Instructions to use Kushtrim/llama-3.2-3b-instruct-sq-en with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Kushtrim/llama-3.2-3b-instruct-sq-en with Transformers:

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

pipe = pipeline("text-generation", model="Kushtrim/llama-3.2-3b-instruct-sq-en")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

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

tokenizer = AutoTokenizer.from_pretrained("Kushtrim/llama-3.2-3b-instruct-sq-en")
model = AutoModelForCausalLM.from_pretrained("Kushtrim/llama-3.2-3b-instruct-sq-en", device_map="auto")
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=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Kushtrim/llama-3.2-3b-instruct-sq-en with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Kushtrim/llama-3.2-3b-instruct-sq-en"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Kushtrim/llama-3.2-3b-instruct-sq-en",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Kushtrim/llama-3.2-3b-instruct-sq-en

SGLang

How to use Kushtrim/llama-3.2-3b-instruct-sq-en 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 "Kushtrim/llama-3.2-3b-instruct-sq-en" \
    --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": "Kushtrim/llama-3.2-3b-instruct-sq-en",
		"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 "Kushtrim/llama-3.2-3b-instruct-sq-en" \
        --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": "Kushtrim/llama-3.2-3b-instruct-sq-en",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Kushtrim/llama-3.2-3b-instruct-sq-en with Docker Model Runner:
```
docker model run hf.co/Kushtrim/llama-3.2-3b-instruct-sq-en
```

You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Llama 3.2 3B Instruct Albanian–English

Fine-tuned, released, and maintained by Kushtrim Visoka.

Kushtrim/llama-3.2-3b-instruct-sq-en is a task- and language-adapted checkpoint from the Meta Llama 3.2 family. The repository name describes the intended variant; the technical properties below are taken from its saved configuration.

Model details

Property	Value
Model type	Decoder-only causal language model
Architecture	`LlamaForCausalLM`
Intended language(s)	sq, en
Saved precision	BF16
Context or generation limit	131,072 tokens (configuration maximum)

Intended use

Research and evaluation in the languages and task indicated by the repository name.
Comparison with the upstream model family.
Further task- or domain-specific adaptation where the applicable license permits.

Evaluation and limitations

A standardized benchmark suite is not documented in this model card yet.
Evaluate the checkpoint on representative held-out data before deployment.
Fine-tuning can preserve or amplify limitations and biases from the upstream model and training data.
Generated text or transcripts may be inaccurate. Human review is required for consequential use.
The configured maximum context does not guarantee reliable quality at every sequence length.

Development and attribution

Original architecture and base-model family: Meta Llama Team
Albanian–English adaptation, fine-tuning, checkpoint packaging, release, and repository maintenance: Kushtrim Visoka
Model card: Kushtrim Visoka

Acknowledgements

This work builds on Meta Llama 3.2 and the work of its original authors. The upstream architecture and base weights remain the work of their respective creators. This repository documents the derivative fine-tuning, packaging, and release work carried out by Kushtrim Visoka and does not claim authorship of the original model. Users should also follow the upstream model's license and citation requirements.

Citation

If you use this checkpoint, please cite this repository and the original upstream model.

@misc{visoka_llama32_3b_instruct_sq_en_2026,
  author       = {Visoka, Kushtrim},
  title        = {Llama 3.2 3B Instruct Albanian–English},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/Kushtrim/llama-3.2-3b-instruct-sq-en},
  note         = {Fine-tuned derivative of the Meta Llama 3.2 family}
}

Downloads last month: -

Safetensors

Model size

3B params

Tensor type

F32

BF16

Model tree for Kushtrim/llama-3.2-3b-instruct-sq-en

Base model

meta-llama/Llama-3.2-3B-Instruct

Finetuned

(2,049)

this model