mcp-memory-auto-trigger-model

MCP Memory Auto-Trigger Model

🎯 Model Description

This model was trained to automatically decide when to save information to memory, search existing memory, or take no action based on user conversations. It's designed for intelligent memory management in AI assistants.

📊 EXCEPTIONAL PERFORMANCE

• Accuracy: 0.9956 (99.56%) 🔥
• F1 Macro: 0.9964
• F1 Weighted: 0.9956

📚 Training Data

• Dataset: PiGrieco/mcp-memory-auto-trigger-ultimate
• Total Examples: 47,516
• Real Data: 68% (BANKING77, CLINC150)
• Synthetic Data: 32% (high-quality generated)
• Language: English

🎯 Classes

• SAVE_MEMORY (0): Save important information to memory
• SEARCH_MEMORY (1): Search for existing information in memory
• NO_ACTION (2): Normal conversation requiring no memory action

💻 Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("PiGrieco/mcp-memory-auto-trigger-model")
model = AutoModelForSequenceClassification.from_pretrained("PiGrieco/mcp-memory-auto-trigger-model")

# Example usage
text = "I need to remember this configuration setting for later"
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)

with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predicted_class = torch.argmax(predictions, dim=-1).item()

class_names = ["SAVE_MEMORY", "SEARCH_MEMORY", "NO_ACTION"]
print(f"Predicted action: {class_names[predicted_class]}")
print(f"Confidence: {predictions[0][predicted_class]:.4f}")

🏋️ Training Details

• Base Model: distilbert-base-uncased
• Training Framework: Hugging Face Transformers
• Hardware: Google Colab A100 GPU
• Training Time: ~3-4 hours
• Epochs: 3
• Batch Size: 32
• Learning Rate: 2e-5
• Mixed Precision: Yes (fp16)

🚀 Production Ready

This model achieves world-class performance and is ready for immediate production deployment in MCP Memory Server systems.

📈 Model Performance

With 99.56% accuracy, this model represents state-of-the-art performance for memory trigger classification tasks.