In this work, we focus on fine-tuning an OpenAI GPT-2 pre-trained model for generating patent claims. GPT-2 has demonstrated impressive efficacy of pre-trained language models on various tasks, particularly coherent text generation. Patent claim language itself has rarely been explored in the past and poses a unique ch…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,657 papers · 148 categories
Trend · papers per month
48 results for “GPT-2”
DARE uses GPT-2 to generate relation examples, improving RE performance.
problem Challenges in Relation Extraction, especially limited data and class imbalance.
method Data augmentation with GPT-2 fine-tuning to generate relation examples.
result Improves RE performance by up to 11 F1 score points.
Alpha-GPT 2.0 integrates human insights into AI-driven investment research.
problem Efficiency and precision in quantitative investment research.
method Iterative Human-AI interaction using large language models.
result Enhanced efficiency and precision in quantitative investment research.
We present an open-source tool for visualizing multi-head self-attention in Transformer-based language representation models. The tool extends earlier work by visualizing attention at three levels of granularity: the attention-head level, the model level, and the neuron level. We describe how each of these views can he…
New algorithms improve privacy and utility of large language models.
problem Privacy-preserving fine-tuning of large language models.
method Meta-framework for differentially private fine-tuning, inspired by recent success in fine-tuning.
result Private fine-tuned models achieve utility close to non-private models, with improved privacy and efficiency.
Project aims to improve coherence in language generation models.
problem Models often generate inconsistent text that diverges from the prompt.
method Trained a sentence pair coherence classifier and co-trained GPT-2 with this coherence objective.
result Fine-tuned model generates coherent paragraphs without diverging.
BatchTopK SAEs improve GPT-2 and Gemma activations with adjustable sparsity.
problem Interpreting language model activations using sparse autoencoders.
method Adapting the TopK constraint to the batch-level, allowing variable number of active latents per sample.
result BatchTopK SAEs consistently outperform TopK SAEs in reconstructing GPT-2 and Gemma activations.
Pretrained LLM Adapted with LoRA as a Decision Transformer for Offline RL in Quantitative Tradingq-fin.CP
Pre-trained LLM adapted with LoRA improves offline RL for quantitative trading.
problem Challenges in offline RL for quantitative trading due to complex temporal dependencies and overfitting.
method Integrates pre-trained GPT-2 weights and LoRA for efficient fine-tuning of a Decision Transformer.
result Outperforms existing offline RL methods in certain trading scenarios.
Automates translating natural language to Verilog for digital design.
problem Manual translation of natural language specifications to Verilog is time-consuming and error-prone.
method Fine-tuned GPT-2 to derive Verilog from English, using a dataset of design tasks.
result GPT-2 achieved 94.8% correct translation across simple and abstract design tasks.
Signed pairwise interactions conflate uniqueness, redundancy, and synergy
problem Signed pairwise interactions conflate uniqueness, redundancy, and synergy
method Stochastic Hi-Fi
result Stochastic Hi-Fi recovers structure missed by scalar baselines
Theory predicts neural scaling exponents from language statistics.
problem No existing theory could quantitatively predict neural scaling exponents.
method Isolated two key statistical properties of language.
result Derives a simple formula predicting neural scaling exponents.
Pretrained masked language models (MLMs) require finetuning for most NLP tasks. Instead, we evaluate MLMs out of the box via their pseudo-log-likelihood scores (PLLs), which are computed by masking tokens one by one. We show that PLLs outperform scores from autoregressive language models like GPT-2 in a variety of task…
MARS optimizes large model training by reducing variance, outperforming AdamW.
problem Training large models efficiently and scalably.
method Unified optimization framework MARS combining preconditioned gradient updates and variance reduction.
result MARS outperforms AdamW in training GPT-2 models.
New method boosts performance of diffusion models on discrete data like natural language.
problem Performance of diffusion models on discrete data like natural language is poor.
method Proposes score entropy, a novel loss that extends score matching to discrete spaces.
result Significantly boosts performance on language modeling tasks.
IVON optimizes large neural networks, matching or outperforming Adam.
problem The inefficacy of variational learning in large neural networks.
method Improved Variational Online Newton (IVON) optimizer.
result IVON consistently matches or outperforms Adam for large networks.
Study combines sentiment analysis with traditional models for better S&P 500 trading.
problem Improving trading performance in volatile markets.
method Sentiment analysis from financial news, GPT-2, FinBERT, combined with technical indicators and time-series models.
result Combining sentiment-driven insights with traditional models improves trading performance.
The Transformer is a fully attention-based alternative to recurrent networks that has achieved state-of-the-art results across a range of NLP tasks. In this paper, we analyze the structure of attention in a Transformer language model, the GPT-2 small pretrained model. We visualize attention for individual instances and…
In unsupervised learning, collecting more data is not always a costly process unlike the training. For example, it is not hard to enlarge the 40GB WebText used for training GPT-2 by modifying its sampling methodology considering how many webpages there are in the Internet. On the other hand, given that training on this…
Generative models predict page quality without training, useful for low-resource settings.
problem Detecting low-quality content in web articles.
method Human evaluation and analysis of 500 million web articles.
result Generative models can predict page quality without training, useful for low-resource settings.
Paper improves privacy for language models against reconstruction attacks.
problem Reconstruction attacks can regenerate training data from language models.
method Uses Rényi differential privacy with optimized privacy budgets.
result Better privacy guarantees for extraction of rare secrets.
This research discovers model architecture and training dataset characteristics through strategic input probing.
problem Discovering model architecture and training dataset characteristics in black box models.
method Structured input probes and model outputs are used to train a deep classifier for image and text classification.
result The approach successfully distinguishes between different image and text datasets and architectures.
New framework improves interpretability of trainable prompts.
problem Improving task-specific LLM performance with soft prompts remains a black-box method.
method Developed a theoretical framework for evaluating interpretability of trainable prompts, inspired new objective functions.
result Found a fundamental trade-off between interpretability and task performance in trainable prompts.
Log-Normal Multiplicative Dynamics improves low-precision training of neural networks.
problem Training large neural networks with low precision is unstable.
method Derive a Bayesian learning rule with log-normal posterior distributions and multiplicative updates.
result LMD achieves stable and accurate training for Vision Transformer and GPT-2.
This paper enhances language models with knowledge awareness.
problem Understanding how much knowledge pretrained language models grasp.
method Inserting explicit knowledge layers into pretraining without changing transformer architecture.
result Significantly more knowledge packed into transformer parameters.
A tailored HTR system improves CER to 0.015 for medieval Latin.
problem Digitizing handwritten medieval Latin records for a low-resource language.
method End-to-end pipeline using image segmentation and transformer-based models with extensive data augmentation.
result Best-performing setup achieved CER of 0.015, superior to commercial models.
Biomedical research papers use significantly different language and jargon when compared to typical English text, which reduces the utility of pre-trained NLP models in this domain. Meanwhile Medline, a database of biomedical abstracts, introduces nearly a million new documents per-year. Applications that could benefit…
Random feature models can outperform a weak teacher with early stopping.
problem Generalization from a weak to a strong model in random feature networks.
method Random feature models, early stopping, proving weak-to-strong generalization.
result Random feature models can outperform a weak teacher with early stopping.
Optimus pre-trains sentences in a latent space for various NLP tasks.
problem Training large-scale language models for diverse NLP tasks.
method Pre-trained Variational Autoencoder (VAE) on large text corpus, fine-tuned for various tasks.
result Optimus achieves state-of-the-art on VAE language modeling benchmarks.
Study quantifies how LLMs capture higher-order statistical structure using cumulant expansion.
problem Understanding how LLMs internalize statistical structure during next-token prediction.
method Cumulant-expansion framework treating softmax entropy as perturbation around center distribution.
result Cumulants reveal distinct signatures for mathematical vs. general text prompts, quantifying feature-learning dynamics.
TensorGuide improves LoRA efficiency and expressivity through joint tensor-train optimization.
problem Limited expressivity and generalization of standard LoRA.
method TensorGuide uses a unified tensor-train structure with controlled Gaussian noise to generate correlated low-rank matrices.
result TensorGuide achieves superior accuracy and scalability with fewer parameters compared to standard LoRA and TT-LoRA.
The study analyzes numerical stability in large language models using mixed-precision arithmetic.
problem Numerical stability of large language models using low-precision arithmetic.
method Developed a mixed-precision analysis of transformer inference, deriving bounds for condition numbers and forward error.
result Established that numerical stability is determined by the interplay between weight magnitude and the growth of the residual stream.
Paper finds sharpness differences in transformer blocks accelerating LLM training.
problem Understanding and accelerating large language model pre-training.
method Uncovering sharpness disparity across transformer blocks and proposing Blockwise Learning Rate.
result Blockwise Learning Rate strategy accelerates LLM pre-training with lower loss and speedup.
Ray tracing sampler improves neural network sampling efficiency and resilience.
problem Sampling neural network posterior distributions efficiently and robustly.
method Markov Chain Monte Carlo using ray tracing through likelihood space.
result Significantly higher resilience to gradient heating compared to HMC.
BLAST optimizes deep model inference by learning efficient matrix structures.
problem Efficient inference of large-scale deep neural networks.
method Introduces BLAST matrix for adaptive structured weights.
result Significant compression and performance improvements for various model sizes.
We show how to convert ICL in linearized transformers into model weights.
problem Making in-context learning interpretable and permanent in large language models.
method Demonstrates equivalence between ICL and bias terms in linearized transformers, and develops ICLCA for exact conversion.
result Exact conversion of in-context learning into model weights is possible for linearized transformers.
Attention temperature improves robustness of ICL in high-dimensional settings.
problem ICL robustness failure under distribution shift in high dimensions.
method Analyzed a Transformer with approximate softmax attention, derived a closed-form error expression, and showed optimal temperature minimizes error.
result Optimal attention temperature minimizes ICL generalization error under distribution shift.
Simplified masked diffusion models improve discrete data generation.
problem Complex model formulations and unclear relationships hinder discrete data generative modeling.
method Developed a simple and general framework for masked diffusion models.
result Models trained on OpenWebText surpass prior diffusion language models and outperform autoregressive models.
AdamS uses momentum as a denominator to optimize LLMs efficiently.
problem Optimizing large language models (LLMs) with efficient and effective methods.
method AdamS introduces a novel denominator based on the root of the weighted sum of squares of momentum and current gradient.
result AdamS achieves superior optimization performance with minimal memory and compute requirements.
Out-of-Distribution Generalization of In-Context Learning: A Low-Dimensional Subspace Perspectivestat.ML
Paper provides a mathematical model for transformer ICL out-of-distribution generalization.
problem Understanding when transformer in-context learning can generalize beyond pre-training data.
method Minimal mathematical model of linear regression tasks with low-rank covariance matrices, analyzing distribution shifts as varying angles between subspaces.
result Transformers can generalize to all angle shifts if pre-training tasks are drawn from a union of subspaces, but not from a single Gaussian.
New method improves inference for discrete diffusion models, achieving better quality and efficiency.
problem High dimensionality of discrete diffusion models causes inference challenges.
method Developed high-order numerical inference schemes for discrete diffusion models.
result Second-order accuracy of the -Trapezoidal method in KL divergence.
IGSD separates task-specific content channels in transformer components by comparing activation replacement with zero ablation.
problem Mechanistic interpretability of transformer components
method IGSD: paired-intervention framework for comparing activation replacement with zero ablation
result IGSD identifies an early-layer content channel in transformer components that standard importance methods underestimate.
Unified framework connects different neural network models.
problem Understanding the geometry of neural network loss landscapes.
method Unified framework capturing four symmetry classes.
result First discovery of low- and zero-barrier linear interpolation paths.
Social bias in machine learning has drawn significant attention, with work ranging from demonstrations of bias in a multitude of applications, curating definitions of fairness for different contexts, to developing algorithms to mitigate bias. In natural language processing, gender bias has been shown to exist in contex…
AutoScale improves LLM pre-training by adjusting data mixtures at different scales.
problem Data mixtures that work well at small scales may not perform as well at larger scales.
method AutoScale uses a two-stage approach: fitting a model to predict loss under different compositions and extrapolating optimal compositions to larger scales.
result AutoScale accelerates convergence and improves downstream performance.
Improving optimization for iterate-averaged language models
problem How to optimize the averaged model returned by Language Model pipelines
method Formulating optimizer design as an optimal-control problem
result Proven convergence rate and strict improvement in squared error
Neural dialogue models, despite their successes, still suffer from lack of relevance, diversity, and in many cases coherence in their generated responses. These issues can attributed to reasons including (1) short-range model architectures that capture limited temporal dependencies, (2) limitations of the maximum likel…
New framework analyzes LLM personalization trade-offs under congestion.
problem Tension between personalization and resource sharing in LLMs.
method Developed a statistical-economic framework to model user incentives.
result Congestion can flip rankings of SFT and ICL, and offers both methods never hurt profits.
Backpropagation-free trunk training improves model performance on various benchmarks.
problem Memory inefficiency and noisy gradient estimates in deep network training.
method Split Forward Gradient (Split-FG) method that splits network into trunk and head, estimating only trunk gradient.
result Split-FG achieves better performance than pure forward-gradient training and backpropagation on various benchmarks.