Pruned Language Model: olusegunola/phi-3-mini-pruned-wanda

This is a structurally pruned version of the microsoft/Phi-3-mini-4k-instruct model using the Wanda (Weight-Activation Magnitude) pruning method.

  • Base Model: microsoft/Phi-3-mini-4k-instruct
  • Parameter Reduction: 44.26% of total trainable parameters were removed.
  • Average Reconstruction Loss (MSE): 0.048602

Evaluation Results

Performance on common language model benchmarks (0-shot), limited to 100 samples:

Task Metric Value
Evaluation not run
Downloads last month
1
Safetensors
Model size
2B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support