🛡️ B2B Product Catalog Assistant

A specialized language model fine-tuned for B2B product catalog management and customer service in the security and access control industry.

Model Description

This model has been fine-tuned to serve as an intelligent assistant for B2B product catalogs, specifically trained on security equipment data. It excels at:

Product Information Retrieval: Detailed specifications, features, and technical data
Pricing Management: Retail and wholesale pricing inquiries
Inventory Management: Stock status and availability checking
Product Comparisons: Side-by-side feature and price comparisons
Category Navigation: Product discovery within specific categories
Customer Service: Professional B2B customer support conversations
Multilingual Support: English and Greek language capabilities

Training Data

The model was fine-tuned on comprehensive B2B product catalog data including:

Security Products: Door intercoms, access control systems, surveillance equipment
Manufacturer Data: HIKVISION, ZK TECO, and other leading brands
Technical Specifications: Detailed product features, dimensions, power requirements
Pricing Information: Both retail and wholesale pricing structures
Inventory Data: Stock levels and availability status
Customer Interactions: Real B2B customer service conversations
Product Categories: Greek and English category structures

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the model
model = AutoModelForCausalLM.from_pretrained(
    "ainovatronsec/b2b-product-catalog-16bit-latest",
    torch_dtype=torch.float16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("ainovatronsec/b2b-product-catalog-16bit-latest")

# Example B2B inquiry
messages = [
    {"role": "user", "content": "I need pricing information for HIKVISION DS-KD-KP Black including wholesale rates"}
]

# Generate response
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.7,
        top_p=0.8,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

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

Example B2B Queries

Product Information

"Tell me about product PN15047 including all specifications"
"What are the technical details of ZK TECO LOC-ML280K electromagnet?"
"Show me all HIKVISION keypad modules with their features"

Pricing Inquiries

"What's the wholesale price for the DS-KD-KP Black?"
"Give me both retail and wholesale pricing for electromagnets"
"Compare prices between HIKVISION and ZK TECO access control products"

Inventory Management

"What's the current stock status of product PN14130?"
"Which electromagnets are currently in low stock?"
"Show me all available products in the Έλεγχος Πρόσβασης category"

Product Comparisons

"Compare the HIKVISION DS-KD-KP with similar ZK TECO products"
"What's the difference between 280kg and 180kg electromagnets?"
"Show me all door intercom options with their key differences"

Sample B2B Conversation

Customer: "I need a complete access control solution for a small office building. What products do you recommend and what would be the wholesale pricing?"

Assistant: "For a small office building access control solution, I recommend the following components:

Main Controller: ZK TECO access control panel with support for multiple doors
Door Hardware: ZK TECO LOC-ML280K electromagnet (280kg holding force) - wholesale price €28.50
User Interface: HIKVISION DS-KD-KP keypad module for code entry - wholesale price €80.50
Additional Components: Door position sensors and exit buttons

This combination provides reliable access control with both keypad and potential card reader integration. The total wholesale cost for basic components would be approximately €109 plus the main controller unit. Would you like detailed specifications for any of these components or information about bulk pricing for multiple units?"

Training Configuration

Base Model: Qwen3-8B (4-bit quantized for training efficiency)
Fine-tuning Method: LoRA (Low-Rank Adaptation)
LoRA Parameters:
- Rank (r): 32
- Alpha: 32
- Dropout: 0
- Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Training Parameters:
- Batch size: 2 (per device)
- Gradient accumulation: 4 steps
- Learning rate: 2e-4
- Training steps: 100
- Optimizer: AdamW 8-bit
- Context length: 2048 tokens
Final Format: 16-bit merged model for optimal inference quality

Model Performance

Strengths

✅ Product Knowledge: Comprehensive understanding of security product catalogs
✅ Pricing Accuracy: Reliable retail and wholesale price information
✅ Technical Details: Accurate product specifications and features
✅ Professional Tone: Appropriate B2B communication style
✅ Multilingual: Handles both English and Greek product terminology
✅ Inventory Awareness: Stock status and availability information
✅ Category Navigation: Effective product discovery and categorization

Use Cases

B2B Sales Support: Assisting sales teams with product information
Customer Service: Automated responses to common product inquiries
Inventory Management: Quick access to stock and pricing information
Product Recommendations: Suggesting appropriate products for customer needs
Technical Support: Providing detailed product specifications
Multilingual Support: Serving Greek and English-speaking customers

Deployment Options

Production Deployment

Hugging Face Spaces: Easy web interface deployment
FastAPI: RESTful API for integration with existing systems
VLLM: High-performance serving for production workloads
Local Deployment: On-premises installation for sensitive data

Hardware Requirements

Minimum: 16GB RAM, 8GB VRAM (with quantization)
Recommended: 32GB RAM, 16GB VRAM (optimal performance)
Production: 64GB RAM, 24GB+ VRAM (high-throughput serving)

Integration

This model integrates well with:

CRM Systems: Customer relationship management platforms
E-commerce Platforms: Product catalog websites
Inventory Management: Stock tracking systems
Customer Support: Help desk and chat systems
Sales Tools: Quote generation and product recommendation engines

Limitations

Domain Specific: Optimized for security product catalogs, may not perform well on general queries
Training Data Dependency: Responses based on specific product catalog data
Language Scope: Primarily English with Greek product terminology
Real-time Data: Does not access live inventory or pricing systems
Product Updates: Requires retraining for new product additions

Ethical Considerations

Accuracy: While trained on comprehensive data, always verify critical business information
Privacy: Model does not store conversation history or personal data
Bias: Trained specifically on security product data, may show domain bias
Commercial Use: Suitable for commercial applications under MIT license

License

MIT License - Free for commercial and personal use.

Citation

@misc{b2b-product-catalog-assistant,
  title={B2B Product Catalog Assistant: Fine-tuned Language Model for Security Product Catalogs},
  author={ainovatronsec},
  year={2025},
  publisher={Hugging Face},
  journal={Hugging Face Model Hub},
  url={https://huggingface.co/ainovatronsec/b2b-product-catalog-16bit-latest}
}

Support and Updates

For technical support, feature requests, or business inquiries, please contact the model author through the Hugging Face platform.

Version History

v1.0: Initial release with comprehensive B2B product catalog training
Fine-tuned on HIKVISION and ZK TECO security product data
Supports both retail and wholesale pricing inquiries
Multilingual support for English and Greek terminology

ainovatronsec
/

b2b-product-catalog-16bit-latest