---
library_name: peft
base_model: microsoft/phi-2
tags:
- peft
- lora
- conversational
- dark-humor
- phi-2
- finetuned
license: mit
language: en
---

# Dark Humor Bot - LoRA Adapter

This is a LoRA adapter fine-tuned on a dark humor dataset. It's designed to generate witty, cynical responses with dark humor.

## Model Details

- **Base Model:** microsoft/phi-2
- **Fine-tuning Method:** LoRA (Low-Rank Adaptation)
- **Training Data:** Custom dark humor conversations
- **Training Date:** 2026-03-06
- **Language:** English

## Description

A fine-tuned phi-2 model for generating dark humor responses

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-2",
    device_map="auto",
    torch_dtype=torch.float16
)
tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "fausap/dark-phi")

# Generate response
prompt = "### System:
You are a witty, cynical chatbot...\n\n### User:\nTell me a joke\n\n### Assistant:\n"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
Training Details
Quantization: 4-bit QLoRA

LoRA Rank: 8

LoRA Alpha: 16

Batch Size: 1 with gradient accumulation

Learning Rate: 2e-4

Example Response
User: Tell me a dark joke about modern life
Assistant: [Generated response will be here]

Limitations
Optimized for 6GB VRAM

May generate inappropriate content (by design - it's dark humor!)

Best used with the provided system prompt