elite-fleet-qwen-7b

🚖 Elite Fleet AI Dispatcher

A fine-tuned Qwen2.5-7B-Instruct model trained to act as the Head of Dispatch for an Ultra-Elite Chauffeur Service. Given a mission brief, the model selects the optimal vehicle from an available fleet and justifies its decision based on client tier, distance, driver fatigue, energy levels, and maintenance status.

---

Model Details

Model Description

This model was fine-tuned using LoRA (Low-Rank Adaptation) via the Unsloth framework on a proprietary dataset of chauffeur dispatch scenarios. Each training example consists of a structured mission brief containing client tier, passenger count, distance, time, and a list of available fleet vehicles — each with energy level, maintenance status, and driver fatigue score. The model learns to select the single best vehicle and explain its reasoning in the style of a professional dispatcher.

• Developed by: Rodina222 & Raneem-Hussien11
• Model type: Causal Language Model (Fine-tuned)
• Base Model: Qwen/Qwen2.5-7B-Instruct
• Fine-tuning Method: QLoRA via Unsloth
• Language: English
• License: Apache 2.0
• Demo: Elite Fleet Dispatcher Space

---

How to Get Started with the Model

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

MODEL_ID = "Rodina222/elite-fleet-qwen-7b"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype    = torch.float16,
    device_map     = "auto",
    low_cpu_mem_usage = True,
)
model.eval()

SYSTEM_PROMPT = (
    "You are the Head of Dispatch for an Ultra-Elite Chauffeur Service. "
    "Review the mission constraints (Client Tier, Pax, Distance) and the available fleet. "
    "Select the optimal vehicle to maximize client satisfaction and safety. "
    "Explain your reasoning precisely."
)

mission_brief = """MISSION BRIEF:
- Type: Airport Transfer
- Client Tier: VVIP (Royalty/Diplomat)
- Passengers: 3
- Distance: 18.1 miles
- Time: 20:00

AVAILABLE FLEET OPTIONS:
Option 1: Mercedes V-Class (Gas, Van) | Energy: 97% | Maint: NO | Fatigue: 0.41
Option 2: Mercedes S-Class (Gas, Sedan) | Energy: 38% | Maint: NO | Fatigue: 0.05
Option 3: Mercedes S-Class (Gas, Sedan) | Energy: 44% | Maint: NO | Fatigue: 0.16
Option 4: Mercedes V-Class (Gas, Van) | Energy: 35% | Maint: NO | Fatigue: 0.55
Option 5: Lucid Air (Electric, Sedan) | Energy: 63% | Maint: NO | Fatigue: 0.36"""

messages = [
    {"role": "system", "content": SYSTEM_PROMPT},
    {"role": "user",   "content": mission_brief},
]

inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt = True,
    return_tensors        = "pt",
    return_dict           = True,
)
input_ids      = inputs["input_ids"].to(model.device)
attention_mask = inputs["attention_mask"].to(model.device)

with torch.no_grad():
    outputs = model.generate(
        input_ids      = input_ids,
        attention_mask = attention_mask,
        max_new_tokens = 128,
        do_sample      = False,
        use_cache      = True,
    )

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response.split("assistant")[-1].strip())

---

Mission Brief Format

The model expects inputs in this exact structure:

MISSION BRIEF:
- Type: <Airport Transfer | Business Meeting | Cross-City Commute | Night Out>
- Client Tier: <VVIP (Royalty/Diplomat) | VIP | Standard Elite>
- Passengers: <number>
- Distance: <X.X miles>
- Time: <HH:MM>

AVAILABLE FLEET OPTIONS:
Option 1: <Vehicle> (<Fuel>, <Type>) | Energy: <X%> | Maint: <YES/NO> | Fatigue: <0.00>
Option 2: ...

Fleet Vehicles Supported

• Mercedes S-Class (Gas, Sedan)
• Mercedes V-Class (Gas, Van)
• BMW i7 (Electric, Sedan)
• Audi A8 (Gas, Sedan)
• Lucid Air (Electric, Sedan)
• Cadillac Escalade (Gas, SUV)

---

Training Details

Training Data

A proprietary dataset of Ultra-Elite Chauffeur dispatch scenarios in JSONL format. Each example is a 3-turn conversation (system, user, assistant) where the assistant selects the optimal vehicle and provides structured reasoning.

• Dataset size: Custom proprietary dataset
• Train/Test split: 90% / 10% (seed=3407)
• Format: ChatML (Qwen instruction format)

Training Procedure

---

Evaluation

Metrics

Key Business Rules Learned

• Never selects a vehicle with Maint: YES
• Prioritises prestige vehicles for VVIP clients (Rolls-Royce, Mercedes S-Class, BMW i7)
• Avoids high driver fatigue (Fatigue > 0.7 considered risky)
• Avoids low energy for electric vehicles on long routes
• Selects vans/SUVs (V-Class, Escalade) when passenger count exceeds sedan capacity

---

Intended Uses

Direct Use

Fleet dispatching for chauffeur and premium transportation services. The model takes a structured mission brief and outputs an optimal vehicle selection with reasoning.

Downstream Use

Can be integrated into dispatch management systems, booking platforms, or operational dashboards for luxury transportation companies.

Out-of-Scope Use

• Standard taxi or ride-sharing dispatch (not trained on this domain)
• Non-English mission briefs
• Fleets with vehicle types not seen during training

---

Technical Specifications

Model Architecture

• Architecture: Qwen2.5 Transformer (Decoder-only)
• Parameters: 7 Billion
• Quantisation: 4-bit NF4 (QLoRA)
• Fine-tuning: LoRA adapters on attention + MLP layers

Compute Infrastructure

---

Framework Versions

• PEFT 0.18.1
• Transformers ≥ 4.40.0
• Unsloth (latest)
• TRL (latest)
• PyTorch ≥ 2.1.0