{"url":"/method/ternary-weight-splitting","slug":"ternary-weight-splitting","name":"Ternary Weight Splitting","full_name":"Ternary Weight Splitting","full_name_withheld":false,"description_markdown":"**Ternary Weight Splitting** is a ternarization approach used in [BinaryBERT](https://www.paperswithcode.com/method/binarybert) that exploits the flatness of ternary loss landscape as the optimization proxy of the binary model. We first train the half-sized ternary BERT to convergence, and then split both the latent full-precision weight $\\mathbf{w}^{t}$ and quantized $\\hat{\\mathbf{w}}^{t}$ to their binary counterparts $\\mathbf{w}\\_{1}^{b}, \\mathbf{w}\\_{2}^{b}$ and $\\hat{\\mathbf{w}}\\_{1}^{b}, \\hat{\\mathbf{w}}\\_{2}^{b}$ via the TWS operator. To inherit the performance of the ternary model after splitting, the TWS operator requires the splitting equivalency (i.e., the same output given the same input):\r\n\r\n$$\r\n\\mathbf{w}^{t}=\\mathbf{w}\\_{1}^{b}+\\mathbf{w}\\_{2}^{b}, \\quad \\hat{\\mathbf{w}}^{t}=\\hat{\\mathbf{w}}\\_{1}^{b}+\\hat{\\mathbf{w}}\\_{2}^{b}\r\n$$\r\n\r\nWhile solution to the above equation is not unique, we constrain the latent full-precision weights after splitting $\\mathbf{w}\\_{1}^{b}, \\mathbf{w}\\_{2}^{b}$ to satisfy $\\mathbf{w}^{t}=\\mathbf{w}\\_{1}^{b}+\\mathbf{w}\\_{2}^{b}$. See the paper for more details.","description_state":"present","introduced_year":null,"introduced_by":{"title":"BinaryBERT: Pushing the Limit of BERT Quantization","paper":"/paper/binarybert-pushing-the-limit-of-bert","first_author":"Haoli Bai","n_authors":9,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/binarybert-pushing-the-limit-of-bert"},"source":{"url":"https://arxiv.org/abs/2012.15701v2","title":"BinaryBERT: Pushing the Limit of BERT Quantization","url_on_a_paper_host":true},"code_snippet_url":null,"code_snippet_url_on_a_code_host":false,"categories":[{"area":"General","area_id":"general","collection":"Ternarization","url":"/methods/category/ternarization","pwc_aliases":[]}],"n_papers_tagged":2,"archive_num_papers":2,"papers_newest_first":[{"paper":null,"title":"Co-Designing Binarized Transformer and Hardware Accelerator for Efficient End-to-End Edge Deployment","date":"2024-07-16","arxiv_id":"2407.12070","n_code_links":0,"syntology":null},{"paper":"/paper/binarybert-pushing-the-limit-of-bert","title":"BinaryBERT: Pushing the Limit of BERT Quantization","date":"2020-12-31","arxiv_id":"2012.15701","n_code_links":1,"syntology":{"ran":4,"of":6,"unverified":2,"pointer_only":6}}],"papers_shown":2,"tasks":[{"task":"/task/quantization","name":"Quantization","papers":2},{"task":"/task/binarization","name":"Binarization","papers":1},{"task":"/task/model-compression","name":"Model Compression","papers":1},{"task":"/task/scheduling","name":"Scheduling","papers":1}],"tasks_shown":4,"n_tasks":4,"usage_by_year":[{"year":"2020","papers":1},{"year":"2024","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/ternary-weight-splitting"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}