A fine-tuned LoRA adapter for Qwen/Qwen2.5-1.5B-Instruct, trained on 16,000 multi-turn conversations to act as 8 AI hotel staff roles at a luxury 5-star hotel.

🤖 The 8 AI Roles

#	Role	Type	Description
1	Reservation Agent	Guest-Facing	Room bookings, modifications, cancellations, pricing
2	Concierge	Guest-Facing	Local attractions, dining, spa, transportation, events
3	Guest Help Desk	Guest-Facing	Wi-Fi, amenities, complaints, maintenance, policies
4	Room Service Agent	Guest-Facing	Food/beverage orders, dietary needs, menu guidance
5	Virtual Front Desk	Guest-Facing	Check-in, check-out, room upgrades, key cards
6	Cashier Assistant	Staff-Facing	Billing, invoices, tax calculations, loyalty discounts
7	Housekeeping Coordinator	Staff-Facing	Room priorities, cleaning schedules, inventory
8	Security Assistant	Staff-Facing	Visitor logs, incident reports, emergency protocols

🚀 Quick Start

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model + LoRA adapter
base_model = "Qwen/Qwen2.5-1.5B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)

model = AutoModelForCausalLM.from_pretrained(
    base_model,
    device_map="auto",
    torch_dtype=torch.float16,
    trust_remote_code=True
)
model = PeftModel.from_pretrained(model, "himu1780/meridian-palace-v1")
model = model.merge_and_unload()

# Chat with the Reservation Agent
messages = [
    {"role": "system", "content": "You are the Reservation Agent at The Meridian Palace, an award-winning 5-star luxury hotel..."},
    {"role": "user", "content": "I'd like to book a suite for 3 nights next month."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7, top_p=0.9, do_sample=True)

response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
print(response)

📊 Training Details

Parameter	Value
Base Model	Qwen/Qwen2.5-1.5B-Instruct
Method	LoRA (Low-Rank Adaptation)
Training Data	16,000 multi-turn ChatML conversations
Train / Validation	15,200 / 800
Steps	300
Final Loss	0.1273
LoRA Rank	8
LoRA Alpha	32
Learning Rate	2e-4
Max Seq Length	512
Hardware	HuggingFace Free CPU (2 vCPU, 16GB RAM)
Training Time	~22 hours

📁 Dataset

Trained on himu1780/meridian-palace-training — 16,000 conversations in ChatML format covering:

50% Standard operations (routine interactions)
20% Complex scenarios (multi-step, nuanced situations)
10% Edge cases (unusual requests, system errors)
10% Difficult situations (complaints, angry guests)
5% VIP / Special (celebrity, wedding, high-value guests)
5% Cross-department handoffs

🎯 Hotel Context

The model is trained with knowledge of "The Meridian Palace" — a fictional 5-star luxury hotel with:

320 rooms across 25 floors (7 room types from $280–$3,200/night)
3 restaurants, 2 bars, full spa, fitness center, conference facilities
Meridian Rewards loyalty program (Silver → Gold → Platinum → Diamond)
Comprehensive policies (cancellation, children, pets, smoking)

🔗 Links

Demo Space: himu1780/meridian-palace-demo
Training Dataset: himu1780/meridian-palace-training

📄 License

MIT