paladim-sentiment

PALADIM Model - Sentiment Analysis Demo

This is a PALADIM (Pre Adaptive Learning Architecture of Dual-Process Hebbian-MoE Schema) model fine-tuned on IMDB sentiment analysis.

Model Details

• Base Model: distilbert-base-uncased
• Task: Sentiment Analysis (Binary Classification)
• LoRA Rank: 16
• Training Data: IMDB dataset (subset)
• Parameters: Only ~0.3% trainable (LoRA adapters)

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
from peft import PeftModel

# Load base model
model_name = "distilbert-base-uncased"
model = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Load PALADIM LoRA adapters
model = PeftModel.from_pretrained(model, "nickagge/paladim-sentiment")

# Inference
text = "This movie was absolutely fantastic! I loved every minute of it."
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
outputs = model(**inputs)
prediction = torch.argmax(outputs.logits, dim=-1)
print("Sentiment:", "Positive" if prediction == 1 else "Negative")

PALADIM Architecture

This model demonstrates PALADIM's Plastic Memory component using LoRA adapters for rapid task adaptation.

Full PALADIM includes:

• 🧠 Plastic Memory (LoRA) - Fast adaptation
• 🛡️ Consolidation Engine (EWC + KD) - Prevent forgetting
• 🔀 Mixture of Experts - Sparse activation
• 🎯 Meta-Controller - Adaptive learning

Training

• Epochs: 2
• Final Loss: 0.0000
• Final Accuracy: 100.00%

Citation

@software{paladim2024,
  title={PALADIM: Pre Adaptive Learning Architecture of Dual-Process Hebbian-MoE Schema},
  author={nickagge},
  year={2025},
  url={https://huggingface.co/nickagge/paladim}
}