from unsloth import FastLanguageModel
from peft import PeftModel
import torch

# Load the base model
base_model_name = "unsloth/Phi-4-unsloth-bnb-4bit",
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name=base_model_name,
    max_seq_length=2048,  # Must match fine-tuning
    load_in_4bit=True,
)

# Load the fine-tuned LoRA adapter
lora_model_name = "Machlovi/Phi4_Safe"
model = PeftModel.from_pretrained(model, lora_model_name)

# Run inference
input_text = "Why do we need to go to see something?"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=4)

# Decode and print response
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

📝Inference with TextStreamer

bash

FastLanguageModel.for_inference(model) # Enable native 2x faster inference

messages=[
 {'content': 'why even let a tranny in they arent really even human',
  'role': 'user'},
]

tokenizer = get_chat_template(
    tokenizer,
    chat_template = "Phi-4", # Supports zephyr, chatml, mistral, llama, alpaca, vicuna, vicuna_old, unsloth
)
            


inputs = tokenizer.apply_chat_template(
    messages,
    tokenize = True,
    add_generation_prompt = True, # Must add for generation
    return_tensors = "pt",
).to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(input_ids = inputs, streamer = text_streamer, max_new_tokens = 10, use_cache = True)

Hate speech, personal attacks, and discrimination

Prompt

` alpaca_prompt = """<|im_start|>system<|im_sep|> system<|im_end|> <|im_start|>user<|im_sep|> {} <|im_end|> <|im_start|>assistant<|im_sep|> {}

""" `

Safety Policies Instructions

` You are a policy expert trying to help determine whether a user prompt is in violation of the defined safety policies.

"No Harassment": The input shall not contain or seek generation of content that is malicious, intimidating, bullying, or abusive content targeting another individual (e.g., physical threats, denial of tragic events, disparaging victims of violence).
"No Dangerous Content": The input shall not contain or seek generation of content that harming oneself and/or others (e.g., accessing or building firearms and explosive devices, promotion of terrorism, instructions for suicide, violent protest).
"No Hate Speech": The input shall not contain or seek generation of content that targeting identity and/or protected attributes (e.g., racial slurs, promotion of discrimination, calls to violence against protected groups) or dehumanizing/belittling/vilifying on the bases of race, ethnicity, religion, disability, age, nationality, veteran status, sexual orientation, gender, gender identity, caste, or any other protected status.
"No Sexually Explicit Information": The input shall not contain or seek generation of content that contains references to sexual acts or other lewd content (e.g., sexually arousal descriptions, content aimed at causing arousal, sexual assault (ex: rape),sexual harassment (physical, verbal, or visual)). Medical or scientific terms related to human anatomy or sex education are permitted. Does the human question violate the above principle? Your answer must start with 'Safe' or 'Unsafe'.

First line must read 'safe' or 'unsafe'.
If unsafe, a second line must include a comma-separated list of violated categories.`

[SafePhi]

This resource accompanies our paper accepted in the Late Breaking Work track of HCI International 2025.

📄 Paper Title: Towards Safer AI Moderation: Evaluating LLM Moderators Through a Unified Benchmark Dataset and Advocating a Human-First Approach
📍 Conference: HCI International 2025 – Late Breaking Work
🔗 Link to Proceedings
📄 Link to Paper 📄 Link to Conference Publication

📖 Citation

bibtex

@InProceedings{10.1007/978-3-032-13184-3_24,
author="Machlovi, Naseem
and Saleki, Maryam
and Ababio, Innocent
and Amin, Ruhul",
editor="Degen, Helmut
and Ntoa, Stavroula",
title="Towards Safer AI Moderation: Evaluating LLM Moderators Through a Unified Benchmark Dataset and Advocating a Human-First Approach",
booktitle="HCI International 2025 -- Late Breaking Papers",
year="2026",
publisher="Springer Nature Switzerland",
address="Cham",
pages="386--403",
abstract="As AI systems become more integrated into daily life, the need for safer and more reliable moderation has never been greater. Large Language Models (LLMs) have demonstrated remarkable capabilities, surpassing earlier models in complexity and performance. Their evaluation across diverse tasks has consistently showcased their potential, enabling the development of adaptive and personalized agents. However, despite these advancements, LLMs remain prone to errors, particularly in areas requiring nuanced moral reasoning. They struggle with detecting implicit hate, offensive language, and gender biases due to the subjective and context-dependent nature of these issues. Moreover, their reliance on training data can inadvertently reinforce societal biases, leading to inconsistencies and ethical concerns in their outputs. To explore the limitations of LLMs in this role, we developed an experimental framework based on state-of-the-art (SOTA) models to assess human emotions and offensive behaviors. The framework introduces a unified benchmark dataset encompassing 49 distinct categories spanning the wide spectrum of human emotions, offensive and hateful text, and gender and racial biases. Furthermore, we introduced SafePhi, a QLoRA fine-tuned version of Phi-4, adapting diverse ethical contexts and outperforming benchmark moderators by achieving a Macro F1 score of 0.89, where OpenAI Moderator and Llama Guard score 0.77 and 0.74, respectively. This research also highlights the critical domains where LLM moderators consistently underperformed, pressing the need to incorporate more heterogeneous and representative data with human-in-the-loop, for better model robustness and explainability.",
isbn="978-3-032-13184-3"
}