CoCoNut (Chain of Continuous Thought) trained checkpoint 11 based on Llama-3.2-1B.
Training
Base model: meta-llama/Llama-3.2-1B
Training method: CoCoNut (continuous latent reasoning)
Dataset: GSM8K
Checkpoint: 11
Usage
python
1import torch
2checkpoint = torch.load("pytorch_model.bin", map_location='cpu')3# This is a raw PyTorch checkpoint (state dict with 'base_causallm.' prefix)
Model Details
This checkpoint contains the model weights from intermediate training steps.
The state dict keys have the prefix base_causallm.model.*.
