Token Distillation: Attention-aware Input Embeddings For New Tokens

• URL: http://arxiv.org/abs/2505.20133v2
• Date: Fri, 31 Oct 2025 13:44:06 GMT
• Title: Token Distillation: Attention-aware Input Embeddings For New Tokens
• Abstract summary: Existing embedding methods require expensive further training or pretraining of additional modules.<n>We propose Token Distillation and show that by distilling representations obtained using the original tokenization, we can quickly learn high-quality input embeddings for new tokens.
• Score: 30.409194986834
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Current language models rely on static vocabularies determined at pretraining time, which can lead to decreased performance and increased computational cost for domains underrepresented in the original vocabulary. New tokens can be added to solve this problem, when coupled with a good initialization for their new embeddings. However, existing embedding initialization methods require expensive further training or pretraining of additional modules. In this paper, we propose Token Distillation and show that by distilling representations obtained using the original tokenization, we can quickly learn high-quality input embeddings for new tokens. Experimental results with a wide range of open-weight models show that Token Distillation outperforms even strong baselines.

Related papers

• Grounded Token Initialization for New Vocabulary in LMs for Generative Recommendation [15.12832019023085]
  Language models (LMs) are increasingly extended with new learnable vocabulary tokens for domain-specific tasks.<n>Standard practice initializes these new tokens as the mean of existing vocabulary embeddings, then relies on supervised fine-tuning to learn their representations.<n>We propose the emphGrounded Token Initialization Hypothesis: linguistically grounding novel tokens in the pretrained embedding space before fine-tuning.
  arXiv  Detail & Related papers  (2026-04-02T17:59:19Z)
• LiteToken: Removing Intermediate Merge Residues From BPE Tokenizers [76.59130257385826]
  Intermediate merge residues in BPE vocabularies are frequent during merge learning so that retained in the final vocabulary, but are mostly further merged and rarely emitted when tokenizing the corpus during tokenizer usage.<n>We present a systematic empirical characterization of this phenomenon across commonly used tokenizers and introduce LiteToken, a simple method for removing residue tokens.<n>Experiments show that LiteToken reduces token fragmentation, reduces parameters, and improves robustness to noisy or misspelled inputs, while preserving overall performance.
  arXiv  Detail & Related papers  (2026-02-04T16:19:05Z)
• Reflection Pretraining Enables Token-Level Self-Correction in Biological Sequence Models [82.79223371188756]
  Chain-of-Thought (CoT) prompting has advanced task-solving capabilities in natural language processing with large language models.<n>Applying CoT to non-natural language domains, such as protein and RNA language models, is not yet possible.<n>We introduce pretraining, for the first time in a biological sequence model, which enables the model to engage in intermediate reasoning.
  arXiv  Detail & Related papers  (2025-12-24T05:25:17Z)
• Teaching Old Tokenizers New Words: Efficient Tokenizer Adaptation for Pre-trained Models [12.218365713546214]
  Tokenizer adaptation plays an important role in transferring pre-trained language models to new domains or languages.<n>The common approach to extension trains a new tokenizer on domain-specific text and appends the tokens that do not overlap with the existing vocabulary.<n>We propose continued BPE training, which adapts a pre-trained tokenizer by continuing the BPE merge learning process on new data.
  arXiv  Detail & Related papers  (2025-12-03T17:20:16Z)
• Textual Inversion for Efficient Adaptation of Open-Vocabulary Object Detectors Without Forgetting [1.1871535995163365]
  Textual Inversion (TI) allows extending the VLM vocabulary by learning new or improving existing tokens to accurately detect novel or fine-grained objects from as little as three examples.<n>The storage and gradient calculations are limited to the token embedding dimension, requiring significantly less compute than full-model fine-tuning.<n>We evaluate whether the method matches or outperforms the baseline methods that suffer from forgetting in a variety of quantitative and qualitative experiments.
  arXiv  Detail & Related papers  (2025-08-07T12:28:08Z)
• Vision Transformers Don't Need Trained Registers [17.412430704896455]
  A sparse set of neurons is responsible for concentrating high-norm activations on outlier tokens.<n>We create a training-free approach to mitigate these artifacts.<n>Our results suggest that test-time registers effectively take on the role of register tokens at test-time.
  arXiv  Detail & Related papers  (2025-06-09T17:59:57Z)
• Training-Free Tokenizer Transplantation via Orthogonal Matching Pursuit [45.18582668677648]
  We present a training-free method to transplant tokenizers in large language models.<n>We approximate each out-of-vocabulary token as a sparse linear combination of shared tokens.<n>We show that OMP achieves best zero-shot preservation of the base model's performance.
  arXiv  Detail & Related papers  (2025-06-07T00:51:27Z)
• Achieving Tokenizer Flexibility in Language Models through Heuristic Adaptation and Supertoken Learning [0.0]
  Pretrained language models (LLMs) are often constrained by their fixed tokenization schemes.<n> Tokenadapt is a model-agnostic tokenizer transplantation method.<n>Our framework introduces two innovations: first, Tokenadapt, a model-agnostic tokenizer transplantation method, and second, novel pre-tokenization for multi-word Supertokens.
  arXiv  Detail & Related papers  (2025-05-14T19:00:27Z)
• HYPEROFA: Expanding LLM Vocabulary to New Languages via Hypernetwork-Based Embedding Initialization [50.27950279695363]
  Many pre-trained language models (PLMs) exhibit suboptimal performance on mid- and low-resource languages.<n>A common strategy to address this is to introduce new tokens specific to the target languages, initialize their embeddings, and apply continual pre-training on target-language data.<n>We propose HYPEROFA, a hypernetwork-based approach for more adaptive token embedding.
  arXiv  Detail & Related papers  (2025-04-21T19:40:32Z)
• Retrieval Backward Attention without Additional Training: Enhance Embeddings of Large Language Models via Repetition [4.249842620609683]
  This paper focuses on improving the performance of pre-trained language models in zero-shot settings through a simple and easily implementable method.<n>We propose a novel backward attention mechanism to enhance contextual information encoding.
  arXiv  Detail & Related papers  (2025-02-28T05:19:18Z)
• Token Assorted: Mixing Latent and Text Tokens for Improved Language Model Reasoning [53.57895922042783]
  Large Language Models (LLMs) excel at reasoning and planning when trained on chainof-thought (CoT) data.<n>We propose a hybrid representation of the reasoning process, where we partially abstract away the initial reasoning steps using latent discrete tokens.
  arXiv  Detail & Related papers  (2025-02-05T15:33:00Z)
• SEP: Self-Enhanced Prompt Tuning for Visual-Language Model [93.94454894142413]
  We introduce a novel approach named Self-Enhanced Prompt Tuning (SEP) SEP explicitly incorporates discriminative prior knowledge to enhance both textual-level and visual-level embeddings. Comprehensive evaluations across various benchmarks and tasks confirm SEP's efficacy in prompt tuning.
  arXiv  Detail & Related papers  (2024-05-24T13:35:56Z)
• OFA: A Framework of Initializing Unseen Subword Embeddings for Efficient Large-scale Multilingual Continued Pretraining [49.213120730582354]
  Instead of pretraining multilingual language models from scratch, a more efficient method is to adapt existing pretrained language models (PLMs) to new languages via vocabulary extension and continued pretraining. We propose a novel framework: $textbfO$ne $textbfF$or $textbfA$ll, which wisely initializes the embeddings of unseen subwords and thus can adapt a PLM to multiple languages efficiently and effectively.
  arXiv  Detail & Related papers  (2023-11-15T10:40:45Z)
• Improving Input-label Mapping with Demonstration Replay for In-context Learning [67.57288926736923]
  In-context learning (ICL) is an emerging capability of large autoregressive language models. We propose a novel ICL method called Sliding Causal Attention (RdSca) We show that our method significantly improves the input-label mapping in ICL demonstrations.
  arXiv  Detail & Related papers  (2023-10-30T14:29:41Z)
• Language Model Pre-Training with Sparse Latent Typing [66.75786739499604]
  We propose a new pre-training objective, Sparse Latent Typing, which enables the model to sparsely extract sentence-level keywords with diverse latent types. Experimental results show that our model is able to learn interpretable latent type categories in a self-supervised manner without using any external knowledge.
  arXiv  Detail & Related papers  (2022-10-23T00:37:08Z)
• LASP: Text-to-Text Optimization for Language-Aware Soft Prompting of Vision & Language Models [67.19124099815645]
  We propose a novel Language-Aware Soft Prompting (LASP) learning method to alleviate base class overfitting. LASP is inherently amenable to including, during training, virtual classes, i.e. class names for which no visual samples are available. LASP matches and surpasses, for the first time, the accuracy on novel classes obtained by hand-crafted prompts and CLIP for 8 out of 11 test datasets.
  arXiv  Detail & Related papers  (2022-10-03T17:56:35Z)
• Imputing Out-of-Vocabulary Embeddings with LOVE Makes Language Models Robust with Little Cost [5.672132510411465]
  State-of-the-art NLP systems represent inputs with word embeddings, but these are brittle when faced with Out-of-Vocabulary words. We follow the principle of mimick-like models to generate vectors for unseen words, by learning the behavior of pre-trained embeddings using only the surface form of words. We present a simple contrastive learning framework, LOVE, which extends the word representation of an existing pre-trained language model (such as BERT) and makes it robust to OOV with few additional parameters.
  arXiv  Detail & Related papers  (2022-03-15T13:11:07Z)
• Token-wise Curriculum Learning for Neural Machine Translation [94.93133801641707]
  Existing curriculum learning approaches to Neural Machine Translation (NMT) require sufficient sampling amounts of "easy" samples from training data at the early training stage. We propose a novel token-wise curriculum learning approach that creates sufficient amounts of easy samples. Our approach can consistently outperform baselines on 5 language pairs, especially for low-resource languages.
  arXiv  Detail & Related papers  (2021-03-20T03:57:59Z)
• Active Learning for Sequence Tagging with Deep Pre-trained Models and Bayesian Uncertainty Estimates [52.164757178369804]
  Recent advances in transfer learning for natural language processing in conjunction with active learning open the possibility to significantly reduce the necessary annotation budget. We conduct an empirical study of various Bayesian uncertainty estimation methods and Monte Carlo dropout options for deep pre-trained models in the active learning framework. We also demonstrate that to acquire instances during active learning, a full-size Transformer can be substituted with a distilled version, which yields better computational performance.
  arXiv  Detail & Related papers  (2021-01-20T13:59:25Z)
• Word Shape Matters: Robust Machine Translation with Visual Embedding [78.96234298075389]
  We introduce a new encoding of the input symbols for character-level NLP models. It encodes the shape of each character through the images depicting the letters when printed. We name this new strategy visual embedding and it is expected to improve the robustness of NLP models.
  arXiv  Detail & Related papers  (2020-10-20T04:08:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.