Instructions to use Xkev/Llama-3.2-3B-Instruct-multihop-BES with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Xkev/Llama-3.2-3B-Instruct-multihop-BES with Transformers:

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

pipe = pipeline("text-generation", model="Xkev/Llama-3.2-3B-Instruct-multihop-BES")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Xkev/Llama-3.2-3B-Instruct-multihop-BES")
model = AutoModelForCausalLM.from_pretrained("Xkev/Llama-3.2-3B-Instruct-multihop-BES")
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 Xkev/Llama-3.2-3B-Instruct-multihop-BES with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Xkev/Llama-3.2-3B-Instruct-multihop-BES

SGLang

How to use Xkev/Llama-3.2-3B-Instruct-multihop-BES 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 "Xkev/Llama-3.2-3B-Instruct-multihop-BES" \
    --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": "Xkev/Llama-3.2-3B-Instruct-multihop-BES",
		"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 "Xkev/Llama-3.2-3B-Instruct-multihop-BES" \
        --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": "Xkev/Llama-3.2-3B-Instruct-multihop-BES",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Xkev/Llama-3.2-3B-Instruct-multihop-BES with Docker Model Runner:
```
docker model run hf.co/Xkev/Llama-3.2-3B-Instruct-multihop-BES
```

Llama-3.2-3B-Instruct — Multi-Hop BES

Paper Link: https://arxiv.org/abs/2605.28814

Multi-hop retrieval-augmented QA agent fine-tuned from meta-llama/Llama-3.2-3B-Instruct with Bidirectional Evolutionary Search (BES) on the 3–4-hop subset of MuSiQue.

For the 8B variant see Xkev/Llama-3.1-8B-Instruct-multihop-BES.

Training

Base model: meta-llama/Llama-3.2-3B-Instruct
Dataset: 3–4-hop subset of MuSiQue (~5.5k examples)
Retrieval: E5 + FAISS over wiki-18

Intended use

Research on multi-hop reasoning and post-training. Not intended for general dialog or production.

License

Llama 3.2 Community License for fine-tuning artifacts. Base model meta-llama/Llama-3.2-3B-Instruct is governed by Meta's Llama 3.2 License, which still applies to derived weights.