Scalable Data Ablation Approximations for Language Models through Modular Training and Merging
- URL: http://arxiv.org/abs/2410.15661v1
- Date: Mon, 21 Oct 2024 06:03:49 GMT
- Title: Scalable Data Ablation Approximations for Language Models through Modular Training and Merging
- Authors: Clara Na, Ian Magnusson, Ananya Harsh Jha, Tom Sherborne, Emma Strubell, Jesse Dodge, Pradeep Dasigi,
- Abstract summary: We propose an efficient method for approximating data ablations which trains individual models on subsets of a training corpus.
We find that, given an arbitrary evaluation set, the perplexity score of a single model trained on a candidate set of data is strongly correlated with perplexity scores of parameter averages of models trained on distinct partitions of that data.
- Score: 27.445079398772904
- License:
- Abstract: Training data compositions for Large Language Models (LLMs) can significantly affect their downstream performance. However, a thorough data ablation study exploring large sets of candidate data mixtures is typically prohibitively expensive since the full effect is seen only after training the models; this can lead practitioners to settle for sub-optimal data mixtures. We propose an efficient method for approximating data ablations which trains individual models on subsets of a training corpus and reuses them across evaluations of combinations of subsets. In continued pre-training experiments, we find that, given an arbitrary evaluation set, the perplexity score of a single model trained on a candidate set of data is strongly correlated with perplexity scores of parameter averages of models trained on distinct partitions of that data. From this finding, we posit that researchers and practitioners can conduct inexpensive simulations of data ablations by maintaining a pool of models that were each trained on partitions of a large training corpus, and assessing candidate data mixtures by evaluating parameter averages of combinations of these models. This approach allows for substantial improvements in amortized training efficiency -- scaling only linearly with respect to new data -- by enabling reuse of previous training computation, opening new avenues for improving model performance through rigorous, incremental data assessment and mixing.
Related papers
- Optimize Cardinality Estimation Model Pretraining by Simplifying the Training Datasets [0.0]
We introduce a simplified training dataset, which has been reduced to a fraction of the size of existing pretraining datasets.
Sufficient experimental results demonstrate that the pre-trained cardinality estimator based on this simplified dataset can still achieve comparable performance to existing models in zero-shot setups.
arXiv Detail & Related papers (2025-02-20T08:06:16Z) - DUET: Optimizing Training Data Mixtures via Feedback from Unseen Evaluation Tasks [40.91931801667421]
This paper presents a novel global-to-local algorithm called DUET that can exploit the feedback loop by interleaving a data selection method with Bayesian optimization.
As a result, DUET can efficiently refine the training data mixture from a pool of data domains to maximize the model's performance on the unseen evaluation task.
arXiv Detail & Related papers (2025-02-01T01:52:32Z) - Optimizing Pretraining Data Mixtures with LLM-Estimated Utility [52.08428597962423]
Large Language Models improve with increasing amounts of high-quality training data.
We find token-counts outperform manual and learned mixes, indicating that simple approaches for dataset size and diversity are surprisingly effective.
We propose two complementary approaches: UtiliMax, which extends token-based $200s by incorporating utility estimates from reduced-scale ablations, achieving up to a 10.6x speedup over manual baselines; and Model Estimated Data Utility (MEDU), which leverages LLMs to estimate data utility from small samples, matching ablation-based performance while reducing computational requirements by $simx.
arXiv Detail & Related papers (2025-01-20T21:10:22Z) - Data Shapley in One Training Run [88.59484417202454]
Data Shapley provides a principled framework for attributing data's contribution within machine learning contexts.
Existing approaches require re-training models on different data subsets, which is computationally intensive.
This paper introduces In-Run Data Shapley, which addresses these limitations by offering scalable data attribution for a target model of interest.
arXiv Detail & Related papers (2024-06-16T17:09:24Z) - Distilled Datamodel with Reverse Gradient Matching [74.75248610868685]
We introduce an efficient framework for assessing data impact, comprising offline training and online evaluation stages.
Our proposed method achieves comparable model behavior evaluation while significantly speeding up the process compared to the direct retraining method.
arXiv Detail & Related papers (2024-04-22T09:16:14Z) - Data Mixing Laws: Optimizing Data Mixtures by Predicting Language Modeling Performance [55.872926690722714]
We study the predictability of model performance regarding the mixture proportions in function forms.
We propose nested use of the scaling laws of training steps, model sizes, and our data mixing law.
Our method effectively optimize the training mixture of a 1B model trained for 100B tokens in RedPajama.
arXiv Detail & Related papers (2024-03-25T17:14:00Z) - Efficient Online Data Mixing For Language Model Pre-Training [101.45242332613944]
Existing data selection methods suffer from slow and computationally expensive processes.
Data mixing, on the other hand, reduces the complexity of data selection by grouping data points together.
We develop an efficient algorithm for Online Data Mixing (ODM) that combines elements from both data selection and data mixing.
arXiv Detail & Related papers (2023-12-05T00:42:35Z) - Dataless Knowledge Fusion by Merging Weights of Language Models [51.8162883997512]
Fine-tuning pre-trained language models has become the prevalent paradigm for building downstream NLP models.
This creates a barrier to fusing knowledge across individual models to yield a better single model.
We propose a dataless knowledge fusion method that merges models in their parameter space.
arXiv Detail & Related papers (2022-12-19T20:46:43Z)
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.