Leo Boytsov
@srchvrs.bsky.social
Machine learning scientist and engineer speaking πtorch & C++ (ph-D CMU) working on (un)natural language processing, speaking πtorch & C++. Opinions sampled from MY OWN 100T param LM.
🧵Perhaps everything you need to know about compression of generative models.
1. It's hard to remove more than 50% of the parameters.
2. Compression is achieved via a combination of sparsification, distillation, and (optionally) quantization.
↩️
1. It's hard to remove more than 50% of the parameters.
2. Compression is achieved via a combination of sparsification, distillation, and (optionally) quantization.
↩️
November 26, 2024 at 12:22 PM
🧵Perhaps everything you need to know about compression of generative models.
1. It's hard to remove more than 50% of the parameters.
2. Compression is achieved via a combination of sparsification, distillation, and (optionally) quantization.
↩️
1. It's hard to remove more than 50% of the parameters.
2. Compression is achieved via a combination of sparsification, distillation, and (optionally) quantization.
↩️