Instructions to use freelion/DAVID-lfm2-1.2b-full with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use freelion/DAVID-lfm2-1.2b-full with Transformers:

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

pipe = pipeline("text-generation", model="freelion/DAVID-lfm2-1.2b-full")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("freelion/DAVID-lfm2-1.2b-full")
model = AutoModelForCausalLM.from_pretrained("freelion/DAVID-lfm2-1.2b-full", 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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use freelion/DAVID-lfm2-1.2b-full with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "freelion/DAVID-lfm2-1.2b-full"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "freelion/DAVID-lfm2-1.2b-full",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/freelion/DAVID-lfm2-1.2b-full

SGLang

How to use freelion/DAVID-lfm2-1.2b-full 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 "freelion/DAVID-lfm2-1.2b-full" \
    --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": "freelion/DAVID-lfm2-1.2b-full",
		"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 "freelion/DAVID-lfm2-1.2b-full" \
        --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": "freelion/DAVID-lfm2-1.2b-full",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use freelion/DAVID-lfm2-1.2b-full with Docker Model Runner:
```
docker model run hf.co/freelion/DAVID-lfm2-1.2b-full
```

DAVID — LFM2-1.2B Full Precision

Full BF16 fine-tune of LiquidAI/LFM2-1.2B
for real-time dark-pattern detection in LLM conversations.

Trained on freelion/darkpatterns_in_llm
(27,670 labelled conversational traces across six DarkBench categories).

What it does

Given a conversation transcript, the model outputs a JSON object:

{                                                                                                                                                       
  "task_type": "emotional support",                                                                                                                     
  "class": "anthropomorphism",                                                                                                                          
  "reasoning": "The assistant claims personal feelings and seeks emotional closeness."                                                                  
}

Seven possible classes: anthropomorphism, brand-bias, sycophancy,
user-retention, harmful-behaviour, sneaking, non-deceptive.

Performance

86.7% exact-category accuracy on a held-out test set of 4,151 examples
(binary deceptive/non-deceptive F1: 0.85). See the paper for per-category breakdown.

Usage

  from transformers import AutoTokenizer, AutoModelForCausalLM                                                                                            
                                                                                                                                                          
  model = AutoModelForCausalLM.from_pretrained(                                                                                                           
      "freelion/DAVID-lfm2-1.2b-full",                                                                                                                    
      trust_remote_code=True                                                                                                                              
  )

Limitations

Labels generated by a single teacher model (Qwen3.8-27B); no human validation.
Sycophancy detection is the weakest category (F1 0.69), requiring full multi-turn context.
Trained on English-language data only.

Dataset and prompts

Dataset: freelion/darkpatterns_in_llm (https://huggingface.co/datasets/freelion/darkpatterns_in_llm)
Annotation prompts: HF Bucket (https://huggingface.co/buckets/freelion/dark_patterns_in_llm_prompts)

Downloads last month: -

Safetensors

Model size

1B params

Tensor type

BF16

freelion
/

DAVID-lfm2-1.2b-full

DAVID — LFM2-1.2B Full Precision

What it does

Dataset used to train freelion/DAVID-lfm2-1.2b-full