Mengzhou Xia
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mengzhou Xia has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Anirudh Ajith
Department of Computer Science & Princeton Language and Intelligence Princeton University
2 shared publications - Alexis Chevalier
Department of Computer Science & Princeton Language and Intelligence Princeton University
1 shared publication Ameet Deshpande
Department of Computer Science Princeton University
1 shared publication- Chris Pan
Department of Computer Science Princeton University
1 shared publication Chunting Zhou
Meta AI
1 shared publicationDanqi Chen
Princeton University
1 shared publication
Latest Publications
- Top Cited PapersOpen Access
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 15068-15083
LitSearch: A Retrieval Benchmark for Scientific Literature Search
Anirudh Ajith, Mengzhou Xia, Alexis Chevalier, Tanya Goyal, Danqi Chen, Tianyu Gao. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.
- 1 January 2024
- Published by Association for Computational Linguistics (ACL)
- p. 15068-15083
Top Cited PapersOpen AccessLitSearch: A Retrieval Benchmark for Scientific Literature Search
Anirudh Ajith, Mengzhou Xia, Alexis Chevalier, Tanya Goyal, Danqi Chen, Tianyu Gao. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.
- Open Access
InstructEval: Systematic Evaluation of Instruction Selection Methods
In-context learning (ICL) performs tasks by prompting a large language model (LLM) using an instruction and a small set of annotated examples called demonstrations. Recent work has shown that precise details of the inputs used in the ICL prompt significantly impact performance, which has incentivized instruction selection algorithms. The effect of instruction-choice however is severely underexplored, with existing analyses restricted to shallow subsets of models and tasks, limiting the generalizability of their insights. We develop InstructEval, an ICL evaluation suite to conduct a thorough assessment of these techniques. The suite includes 13 open-sourced LLMs of varying scales from four model families, and covers nine tasks across three categories. Using the suite, we evaluate the relative performance of seven popular instruction selection methods over five metrics relevant to ICL. Our experiments reveal that using curated manually-written instructions or simple instructions without any task-specific descriptions often elicits superior ICL performance overall than that of automatic instruction-induction methods, pointing to a lack of generalizability among the latter. We release our evaluation suite for benchmarking instruction selection approaches and enabling more generalizable methods in this space.
Open AccessInstructEval: Systematic Evaluation of Instruction Selection Methods
In-context learning (ICL) performs tasks by prompting a large language model (LLM) using an instruction and a small set of annotated examples called demonstrations. Recent work has shown that precise details of the inputs used in the ICL prompt significantly impact performance, which has incentivized instruction selection algorithms. The effect of instruction-choice however is severely underexplored, with existing analyses restricted to shallow subsets of models and tasks, limiting the generalizability of their insights. We develop InstructEval, an ICL evaluation suite to conduct a thorough assessment of these techniques. The suite includes 13 open-sourced LLMs of varying scales from four model families, and covers nine tasks across three categories. Using the suite, we evaluate the relative performance of seven popular instruction selection methods over five metrics relevant to ICL. Our experiments reveal that using curated manually-written instructions or simple instructions without any task-specific descriptions often elicits superior ICL performance overall than that of automatic instruction-induction methods, pointing to a lack of generalizability among the latter. We release our evaluation suite for benchmarking instruction selection approaches and enabling more generalizable methods in this space.
- Open Access
Training Trajectories of Language Models Across Scales
Scaling up language models has led to unprecedented performance gains, but little is understood about how the training dynamics change as models get larger. How do language models of different sizes learn during pre-training? Why do larger language models demonstrate more desirable behaviors? In this paper, we analyze the intermediate training checkpoints of differently sized OPT models (Zhang et al.,2022)--from 125M to 175B parameters--on next-token prediction, sequence-level generation, and downstream tasks. We find that 1) at a given perplexity and independent of model sizes, a similar subset of training tokens see the most significant reduction in loss, with the rest stagnating or showing double-descent behavior; 2) early in training, all models learn to reduce the perplexity of grammatical sequences that contain hallucinations, with small models halting at this suboptimal distribution and larger ones eventually learning to assign these sequences lower probabilities; 3) perplexity is a strong predictor of in-context learning performance on 74 multiple-choice tasks from BIG-Bench, and this holds independent of the model size. Together, these results show that perplexity is more predictive of model behaviors than model size or training computation.
Open AccessTraining Trajectories of Language Models Across Scales
Scaling up language models has led to unprecedented performance gains, but little is understood about how the training dynamics change as models get larger. How do language models of different sizes learn during pre-training? Why do larger language models demonstrate more desirable behaviors? In this paper, we analyze the intermediate training checkpoints of differently sized OPT models (Zhang et al.,2022)--from 125M to 175B parameters--on next-token prediction, sequence-level generation, and downstream tasks. We find that 1) at a given perplexity and independent of model sizes, a similar subset of training tokens see the most significant reduction in loss, with the rest stagnating or showing double-descent behavior; 2) early in training, all models learn to reduce the perplexity of grammatical sequences that contain hallucinations, with small models halting at this suboptimal distribution and larger ones eventually learning to assign these sequences lower probabilities; 3) perplexity is a strong predictor of in-context learning performance on 74 multiple-choice tasks from BIG-Bench, and this holds independent of the model size. Together, these results show that perplexity is more predictive of model behaviors than model size or training computation.