arXiv：cs.AI· Avinash Benki, Prasanjith Pasunuti, Sutanshu Saksena Raj, Parijat Chatterjee, Nayanjit Sarkar, Daksh Maan·· 23 小时前
Nemotron-3-Nano-30B-A3B 原地扩展 Indic 词表并持续预训练
In-Place Indic Tokenizer Expansion and Continual Pretraining of Nemotron-3-Nano-30B-A3B
AI 导读
研究团队在不改动 Nemotron-3-Nano-30B-A3B 主干架构的前提下适配八种印度语言，用续训 BPE 和 AdaptMix 将词表从 131,072 原地扩至 163,072，每词 token 数减少 38.6%。
正文
Abstract:We adapted Nemotron-3-Nano-30B-A3B for eight Indian languages without changing its backbone architecture. We expanded the vocabulary in place from 131,072 to 163,072 entries using continued byte-pair encoding and AdaptMix rebalancing, reducing tokens per word by 38.6%. We initialized new input and output rows by averaging their original constituent rows, then trained them on 90% Indic and 10% English text before updating the backbone. Continual pretraining then progressed through broad training, high-quality annealing, and long-context training over approximately 1.27 trillion tokens. The annealed checkpoint gained 12.9 points on the Indic MILU benchmark but lost 3.4 on English MMLU. We also report practical negative results. Direct token addition underperformed continued byte-pair encoding. AdamW weight decay changed rows intended to stay fixed. Perplexity on an Indic-heavy holdout did not select the learning rate that best preserved English performance.
| Comments: | Accepted at the GlobalSouthAI Workshop, NeurIPS 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15373 [cs.CL] |
| (or arXiv:2610.15373v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15373 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sutanshu Saksena Raj [view email]
[v1]
Sun, 11 Oct 2026 05:53:00 UTC (1,325 KB)
来源：arXiv：cs.AI · arxiv.org