llama32-1b-gsm-coconut-checkpoint7

CoCoNut (Chain of Continuous Thought) trained checkpoint 7 based on Llama-3.2-1B.

Training

• Base model: meta-llama/Llama-3.2-1B
• Training method: CoCoNut (continuous latent reasoning)
• Dataset: GSM8K
• Checkpoint: 7

Usage

import torch
checkpoint = torch.load("pytorch_model.bin", map_location='cpu')
# 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.*.