pruned_olmo3_5120_24_32

WARNING: This model is PRUNED ONLY, NOT retrained or distilled!

Performance will be degraded compared to the original model. This is a structural pruning checkpoint intended as a starting point for knowledge distillation or fine-tuning.

Description

Structurally pruned version of allenai/OLMo-3-7B-Instruct.

Pruning Configuration

Parameter Original Pruned
Intermediate size (MLP) 11008 5120
Attention heads 32 24
Layers 32 32
Hidden size 4096 4096 (unchanged)

Important Notes

  1. This model has NOT been retrained after pruning
  2. Performance will be significantly degraded compared to the original
  3. Intended use: Starting checkpoint for distillation/fine-tuning
  4. For the distillation training data, see hbfreed/dolci-distill-packed
Downloads last month
13
Safetensors
Model size
4B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support