Instructions to use Lalakai/Hunter-1-1.7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Lalakai/Hunter-1-1.7B with Transformers:

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

pipe = pipeline("text-generation", model="Lalakai/Hunter-1-1.7B", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("Lalakai/Hunter-1-1.7B", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Lalakai/Hunter-1-1.7B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Lalakai/Hunter-1-1.7B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Lalakai/Hunter-1-1.7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Lalakai/Hunter-1-1.7B

SGLang

How to use Lalakai/Hunter-1-1.7B 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 "Lalakai/Hunter-1-1.7B" \
    --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": "Lalakai/Hunter-1-1.7B",
		"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 "Lalakai/Hunter-1-1.7B" \
        --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": "Lalakai/Hunter-1-1.7B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Lalakai/Hunter-1-1.7B with Docker Model Runner:
```
docker model run hf.co/Lalakai/Hunter-1-1.7B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Hunter 1.7B

Hunter 1.7B is a compact 1.7B-parameter language model from Lalakai Labs, tuned for defensive security work: incident triage, incident-response question answering, and causal reasoning about attack paths. It is the smallest model in the Hunter 1 series and is designed to run on ordinary laptops.

Hunter 1.7B is built on Spark-X2.5-1.7B from XHToken.

Highlights

Tuned for defensive triage, incident response, and causal attack-path analysis
1.7B parameters in BF16 safetensors; the adapter is merged into the weights, so no separate LoRA loading is needed
Requires trust_remote_code=True (custom Spark2_5ForCausalLM architecture)
Apache-2.0 license

Model details


Developer	Lalakai Labs
Model type	Causal language model (conversational, text generation)
Parameters	1.7B
Precision	BF16
Training sequence length	1,024 tokens
Language	English
License	Apache-2.0

Usage

Install transformers and load the model with trust_remote_code=True:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Lalakai/Hunter-1-1.7B"

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)

prompt = (
    "<|im_start|>system\nYou are a defensive security analyst.<|im_end|>\n"
    "<|im_start|>user\nSummarize the most likely attack path implied by these alerts: ...<|im_end|>\n"
    "<|im_start|>assistant\n"
)

inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=256, do_sample=False)
print(tokenizer.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Prompt format: use ChatML (<|im_start|>role ... <|im_end|>), which matches the fine-tuning format. The chat template bundled with the tokenizer is inherited from the base model and may not match, so build prompts manually as shown above.

Training

Data: 2,500 curated cybersecurity examples, filtered to causal reasoning, offensive security, and incident QA task types
Format: ChatML, truncated to 1,024 tokens
Method: LoRA (r=16, alpha=16, dropout=0.05) on gate_proj and up_proj only; 7.8M trainable parameters (0.45%), merged into the BF16 base weights
Optimization: 150 steps, global batch size 16 sequences, learning rate 3e-4 with cosine decay and 15 warmup steps, BF16, gradient checkpointing
Final training loss: ~1.0 (run average 1.21)

Evaluation

No standardized benchmark results have been published for this model yet.

Limitations

Attention layers were not adapted. The LoRA targets from the original specification (q_proj, v_proj, o_proj) do not exist in this architecture, which uses fused q_k_v_proj and out_proj modules. Only the MLP projections were adapted.
Chat template mismatch. Training used ChatML tags; the bundled chat template is the base model's. See the prompt format note above.
Context length. The model was trained on 1,024-token sequences. Behavior on longer contexts has not been tested.
Not security advice. Outputs may be incorrect or incomplete and should be reviewed by a qualified analyst before any action is taken.

Intended use

Triage support and summarization for security analysts
Drafting and structuring incident-response notes
Reasoning about causal chains in security events
Local experimentation on consumer hardware

Out of scope: operational offensive use, unauthorized testing of systems, and autonomous response actions without human review.

Responsible use

Use this model under human supervision and within the bounds of applicable law and your organization's policies. Do not use it to target systems you are not authorized to test.

License

Released under the Apache License, Version 2.0 (https://www.apache.org/licenses/LICENSE-2.0).

Acknowledgements

Hunter 1.7B is fine-tuned from XHToken/Spark-X2.5-1.7B (https://huggingface.co/XHToken/Spark-X2.5-1.7B) (Apache-2.0). We thank the authors of the base model.