need-ai-conversational-model

NEED AI Conversational Model

Multi-task conversational AI model for the NEED service marketplace platform.

Model Description

This model performs three simultaneous tasks:

1. Intent Classification - Identifies user intent (10 classes)
2. Category Classification - Classifies service category (20 classes)
3. Response Generation - Generates contextual responses

Architecture

• Type: Custom Transformer (Encoder-Decoder)
• Parameters: ~95.9M
• Dimensions: 512 (model), 6 layers, 8 heads
• Vocabulary: 50,257 tokens (GPT-2 tokenizer)
• Context Length: 512 tokens

Usage

Installation

pip install torch transformers huggingface_hub

Loading the Model

import torch
from transformers import AutoTokenizer
from huggingface_hub import hf_hub_download

# Download model and custom code
model_path = hf_hub_download(repo_id="yogami9/need-ai-conversational-model", filename="pytorch_model.bin")
modeling_path = hf_hub_download(repo_id="yogami9/need-ai-conversational-model", filename="modeling_need.py")

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("yogami9/need-ai-conversational-model")

# Import and load model
import sys
import os
sys.path.insert(0, os.path.dirname(modeling_path))
from modeling_need import NEEDConversationalModel

model = NEEDConversationalModel.from_pretrained(model_path)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
model.eval()

Inference Example

def generate_response(text: str, max_length: int = 100):
    # Prepare input
    input_text = f"Human: {text}"
    input_ids = tokenizer.encode(input_text, return_tensors='pt').to(device)
    speaker_ids = torch.zeros_like(input_ids)  # Human speaker

    # Generate
    with torch.no_grad():
        output_ids = model.generate(
            input_ids=input_ids,
            speaker_ids=speaker_ids,
            max_length=max_length,
            temperature=0.8,
            top_k=50
        )

    # Decode
    response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
    return response

# Test
response = generate_response("I need a house cleaner in Lagos")
print(response)

Training Data

Trained on conversations from the NEED service marketplace covering:

• Service requests and inquiries
• Pricing questions
• Location-based queries
• Booking requests
• Category navigation help

Languages: English (Nigerian context)

Training Details

• Framework: PyTorch
• Training Hardware: Google Colab (Tesla T4 GPU)
• Optimizer: AdamW (lr=5e-5, weight_decay=0.01)
• Scheduler: CosineAnnealingLR
• Batch Size: 2-4
• Epochs: 3+ (early stopping)

Limitations

• Trained primarily on Nigerian English context
• Limited to 512 token context window
• May require fine-tuning for other domains
• Custom architecture requires manual loading

Intended Use

Primary: Customer service automation for NEED platform

Suitable for:

• Service request routing
• Basic customer support
• Intent detection
• Category recommendation

Not suitable for:

• Medical/legal advice
• Financial transactions
• High-stakes decision making

Citation

@misc{need-ai-2025,
  title={NEED AI Conversational Model},
  author={NEED Service App Team},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/yogami9/need-ai-conversational-model}
}

Contact

• Email: [email protected]
• Repository: GitHub

License

Apache 2.0