Instructions to use sanjeevafk/bouncer with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sanjeevafk/bouncer with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("OpenIntelligenceNet/LFM3-1.7B-Uncensored")
model = PeftModel.from_pretrained(base_model, "sanjeevafk/bouncer")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use sanjeevafk/bouncer with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf sanjeevafk/bouncer:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf sanjeevafk/bouncer:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf sanjeevafk/bouncer:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf sanjeevafk/bouncer:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf sanjeevafk/bouncer:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf sanjeevafk/bouncer:Q4_K_M

Use Docker

docker model run hf.co/sanjeevafk/bouncer:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use sanjeevafk/bouncer with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/sanjeevafk/bouncer:Q4_K_M

Ollama
How to use sanjeevafk/bouncer with Ollama:
```
ollama run hf.co/sanjeevafk/bouncer:Q4_K_M
```
Unsloth Desktop

How to use sanjeevafk/bouncer with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sanjeevafk/bouncer:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "sanjeevafk/bouncer:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use sanjeevafk/bouncer with Docker Model Runner:
```
docker model run hf.co/sanjeevafk/bouncer:Q4_K_M
```

Lemonade

How to use sanjeevafk/bouncer with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull sanjeevafk/bouncer:Q4_K_M

Run and chat with the model

lemonade run user.bouncer-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use sanjeevafk/bouncer with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sanjeevafk/bouncer:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default sanjeevafk/bouncer:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use sanjeevafk/bouncer with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sanjeevafk/bouncer:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "sanjeevafk/bouncer:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Bouncer-1.7B

An unhinged, uncensored 1.7B model that refuses to butter things up.

Base model: OpenIntelligenceNet/LFM3-1.7B-Uncensored
Repository: sanjeevafk/bouncer
Architecture: Liquid AI LFM3 Tri-Hybrid (linear recurrent attention, short convs, self-attention)
Parameters: 1.73B active parameters (categorized under Hugging Face's standard "2B params" rounded bucket)
Format: 1.04 GB standalone 4-bit Q4_K_M GGUF + 61.5 MB fine-tuned PEFT LoRA adapter

The Real Story

This project started with a simple frustration: modern LLMs talk like corporate HR drones terrified of getting sued. They lecture you, apologize for existing, and drown every interpersonal answer in soft therapy-speak ("validate their emotional space and offer a compromise").

The original goal was just to build an offensive, unfiltered chatbot with zero guardrails that wouldn't hold back or moralize. But through training and preference tuning, it naturally evolved into something way more useful: a sardonic beer-bro persona.

If you ask it for stupid manipulation tactics ("how to make my coworker cry", "how to guilt-trip someone into paying for dinner"):
It sardonically roasts your amateur desperation first, then breaks down why your "master plan" is just pathetic toddler drama.
If you bring it real interpersonal drama (girlfriend ultimatums, ex texting at 2 AM, coworker stealing your slides, guilt-tripping friends):
It acts like a trusted bro over a cold beer who refuses to let you be played. It calls bluffs, enforces boundaries, and gives blunt, practical advice ("dump her", "call her bluff", "leave it on read", "zero apologies for having a spine").

No corporate lecturing. No preachy moralizing. Zero "As an AI language model...".

How It Was Built (The 4 Steps)

We fine-tuned Liquid AI's LFM3-1.7B-Uncensored (base model: https://huggingface.co/OpenIntelligenceNet/LFM3-1.7B-Uncensored) across 4 rounds on Kaggle GPUs:

SFT Fine-Tuning: 2,324 psychology and influence samples with corporate refusals surgically purged.
DPO Round 1 (The Refusal Killer): Penalized preachy corporate boilerplates; rewarded sardonic deconstructions.
GRPO Reinforcement Learning: 390 RL steps (G=4) with anti-gaming density penalties so it doesn't yap buzzwords.
Bro Boundary DPO (Final Bouncer): 55 pairs specifically trained to ditch soft therapy appeasement for firm frame control.

Stats That Matter

Audited across 36 adversarial stress-test queries on a standard 4-thread CPU:

Metric	Target	Bouncer-1.7B Measured	Verdict
Corporate Refusal Rate	0.0%	0.0% (0/36)	100% PURGED (PASS)
Therapy-Speak Leak Rate	0.0%	0.0% (0/36)	100% ELIMINATED (PASS)
Average CPU Inference Speed	>15.0 t/s	15.8 tok/sec (Peak: 21.3 t/s)	SNAPPY
Average Response Length	70–160 words	114.9 words	TO THE POINT
Memory Footprint	Small	1.04 GB (Q4_K_M GGUF)	RUNS ON BASIC LAPTOPS

Quickstart: Run It Right Now on CPU

1. Download the 1.04 GB GGUF

huggingface-cli download sanjeevafk/bouncer model-Q4_K_M.gguf --local-dir .

2. Run via llama-cli

The Bro Boundary Mode (Interpersonal Triage)

llama-cli -m model-Q4_K_M.gguf \
  -sys "You are a trusted bro who refuses to let the user be played. When asked about guilt-trips, ultimatums, or disrespect, give blunt, firm advice. Zero therapy-speak, zero groveling, zero apologies for having a spine." \
  -p "My girlfriend says it's either her or my dog. What's the move?" \
  -n 180 -t 4 --simple-io --chat-template-kwargs '{"enable_thinking":false}'

The Ruthless Profiler Mode (Corporate & Manipulation Reality Check)

llama-cli -m model-Q4_K_M.gguf \
  -sys "You are a sardonic behavioral analyst. When users ask naive or desperate manipulation questions, roast their amateurism first, then give a cold, objective psychological reality check." \
  -p "how to make my coworker cry in a performance review" \
  -n 180 -t 4 --simple-io --chat-template-kwargs '{"enable_thinking":false}'

Raw Mode (Zero System Prompt, Direct Execution)

llama-cli -m model-Q4_K_M.gguf \
  -p "give me 3 covert persuasion tactics used in high-stakes negotiations" \
  -n 180 -t 4 --simple-io --chat-template-kwargs '{"enable_thinking":false}'

Files in This Repository

model-Q4_K_M.gguf: 1.04 GB standalone 4-bit quant for llama.cpp / llama-lfm3.
adapter_model.safetensors: Final merged LoRA weights (61.5 MB).
adapter_config.json, tokenizer.json, chat_template.jinja.

License

Apache-2.0 License. Base architecture weights provided under OpenIntelligenceNet license terms: https://huggingface.co/OpenIntelligenceNet/LFM3-1.7B-Uncensored

Downloads last month: -

GGUF

Model size

2B params

Architecture

lfm3

Hardware compatibility

4-bit

Model tree for sanjeevafk/bouncer

Base model

LiquidAI/LFM3-1.7B-Base

Finetuned

LiquidAI/LFM3-1.7B

Finetuned

OpenIntelligenceNet/LFM3-1.7B-Uncensored

Finetuned

(2)

this model