Puzzletron Pruning

Distillation results for models compressed with Puzzletron MIP-based heterogeneous pruning, followed by Megatron-Bridge knowledge distillation.

Results

Model	File
Llama-3.1-8B-Instruct and Qwen3-8B	Llama-3.1-8B-Instruct.md

Related

• Puzzletron pruning example
• Megatron-Bridge distillation instructions
• Megatron dataset tokenization
• Minitron pruning instructions