The empirical results indicate that the reported gains depend heavily on the experiment setup and the choice of random seed, although the trend of stable gains is confirmed with 128-Dyck pretraining of small models with the Llama tokenizer for most of the examined languages.
Abstract
Pretraining LLMs on artificial languages ("pre-pretraining") is a technique that could reportedly increase token efficiency by 33%, i.e., save up to 33% of training tokens needed to reach a certain performance. We validate this prior result for English on a larger set of natural languages across four language families, using two different tokenizers and varying model sizes. We also relate the observed gains (or losses) in token efficiency to quantified linguistic properties of the languages, such as sentence length, morphological richness, and features of dependency syntactic trees (tree depth, number of children, number of crossing dependencies). Our empirical results indicate that the reported gains depend heavily on the experiment setup and the choice of random seed, although we can confirm the trend of stable gains with 128-Dyck pretraining of small models with the Llama tokenizer for most of the examined languages. On a general note, we argue that multiple training runs should be carried out at least for a subset of experiments to avoid the community adopting unstable approaches.
In this paper, we study the connection between an LLM's output distribution and the data used to train it. Specifically, we study the degree to which an LLM's next-token distribution agrees with the empirical next-token distribution (ENTD) given the context in the training data. The ENTD is an appealing target because it is the unrestricted global minimizer of the next-token cross entropy loss used for pretraining, as well as an easily interpretable function of the pretraining corpus. We find that for a significant fraction of inputs, the LLM's distribution agrees with the ENTD almost perfectly, and the agreement generally increases with model scale and training compute. Nevertheless, there is a long tail of input sequences where the LLM and ENTD differ significantly, and we examine several possible sources of this discrepancy across the transformer architecture, training procedure, and finite-sample noise in the ENTD estimate itself. More broadly, we hope our findings will encourage more work on ``data-centric mechanistic interpretability,''a complement to standard mechanistic interpretability that opens the black box of how model behaviors arise from the data, rather than how they are encoded in the learned weights.
Language model tokenizers are typically selected with minimal evaluation, despite the fact that their design choices directly impact model capabilities. This can be partly attributed to a limited understanding of which tokenizer properties affect which aspects of downstream performance. We introduce TokEval, a framework of tokenizer evaluation metrics that goes beyond standard measures like fertility and compression rate to capture linguistically and structurally meaningful properties, e.g., UTF-8 character boundary integrity and digit place-value boundary alignment for mathematics. To validate whether these metrics are predictive of downstream model performance, we conduct controlled language model pretraining experiments, varying solely the tokenizers'training data mixture, pretokenization strategy, and training algorithm. We evaluate the resulting models on bits-per-byte (a tokenizer-agnostic version of perplexity) and several benchmarks, spanning linguistic understanding, mathematical reasoning, and code generation. Our experiments suggest that different intrinsic properties have different impacts on model abilities: information-theoretic metrics predict language modeling abilities (Spearman rho up to 0.80), while structure-sensitive metrics, such as those measuring digit and line-break handling, correlate with task accuracy. We hope TokEval enables more principled tokenizer evaluation, replacing pretraining sweeps with intrinsic measurement wherever the two agree.
Pretrained language models (PLMs) have established state-of-the-art performance across diverse natural language understanding (NLU) tasks. This study reveals that seman-tic-rich explanations of lexical units can effectively guide PLM learning processes. We propose a novel language understanding enhancement method with token interpretation (LUETI) that addresses two critical limitations in conventional PLMs: Incomplete token semantics caused by isolated contextual learning and insufficient semantic encoding in embedding matrices. LUETI operates through dual mechanisms, augmenting token represen-tations by integrating hidden states with corresponding token interpretations and refining embedding spaces using interpretation-derived semantic vectors for token prediction. LUETI, which is implemented as a plug-in module for standard architectures, demonstrates significant improvements on BERT and GLM, achieving average performance gains of 3.36% and 4.87% respectively on the SuperGLUE benchmark with equivalent parameters and training data. Note that LUETI-equipped models attain comparable performance to baseline PLMs using only 60% of pretraining data. Findings establish token interpretation as a computationally efficient but semantically powerful enhancement strategy for language model pretraining.
Tianyi Chen, Yashen Wang, Huan Chang et al.· IEEE/CAA Journal of Automati...· 0 citations
This work finds that likelihood-trained TPMs can result in failed generations due to overly large corrections to the LM’s logits, and trains TPMs with LM-aligned objectives that better align with the LM token-probability space.
Hanzhang Liu, William Zhao, Zilei Shao et al.· 0 citations
A tokenizer fixed at the start of pre-training allocates vocabulary in proportion to the pre-training corpus, reflecting the deployment priorities at that time. When those priorities shift, languages added later are split into many more tokens per word, which can raise latency, compute, and energy consumption for users of those languages. Cloud models can afford a broad vocabulary because the embedding and LM-head matrices are a small fraction of their parameters. On a compact model those matrices are a material share of per-token decode bandwidth, so on-device models ship small vocabularies and accept fragmentation outside a fixed language set. We present tokenizer expansion, an in-place recipe for upgrading a pre-trained model's tokenizer when the model producer controls its design. We continue the existing tokenizer's BPE merges on a multilingual corpus, so most source tokens carry over unchanged as single tokens and every new token has an exact decomposition into source tokens. We copy the carried-over embedding rows unchanged and initialize new rows as the mean of their source sub-token embeddings. A two-stage adaptation, embedding-only training then full-model continued pre-training, recovers source-checkpoint quality. We apply the recipe to a continued pre-trained checkpoint of LFM2-8B-A1B, an 8B-parameter Mixture-of-Experts model, to help produce LFM2.5-8B-A1B with a 128K tokenizer. The expanded tokenizer encodes Hindi and Vietnamese in roughly $2.4\times$ and $2.6\times$ fewer tokens than the source (up to $4.0\times$ on Thai). Combining these reductions with the measured per-token cost of the larger vocabulary, we estimate a $2.2$-$3.7\times$ per-character decode speedup for these languages across our reference devices. We release the model weights and the expanded tokenizer, and report the negative findings that shaped the recipe.
Jimmy T.H. Smith, Tarek Dakhran, Alberto Cabrera et al.· 0 citations
Control studies on model depth, learning-rate magnitude and scheduling, the interaction between learning-rate decay and data quality, and the effect of including instruction data during pre-training are reported, and an unexplained surge in benchmark performance midway through the constant-learning-rate phase is documented.
Lusheng Zhang, Shien He, Tianxing Yan et al.· 0 citations