What Makes a High-Quality Training Dataset for Large Language Models: A Practitioners' Perspective
Bibliographic record
Abstract
Large Language Models (LLMs) have demonstrated remarkable performance in various application domains, largely due to their self-supervised pre-training on extensive high-quality text datasets. However, despite the importance of constructing such datasets, many leading LLMs lack documentation of their dataset construction and training procedures, leaving LLM practitioners with a limited understanding of what makes a high-quality training dataset for LLMs. To fill this gap, we initially identified 18 characteristics of high-quality LLM training datasets, as well as 10 potential data pre-processing methods and 6 data quality assessment methods, through detailed interviews with 13 experienced LLM professionals. We then surveyed 219 LLM practitioners from 23 countries across 5 continents. We asked our survey respondents to rate the importance of these characteristics, provide a rationale for their ratings, specify the key data pre-processing and data quality assessment methods they used, and highlight the challenges encountered during these processes. From our analysis, we identified 13 crucial characteristics of high-quality LLM datasets that receive a high rating, accompanied by key rationale provided by respondents. We also identified some widely-used data pre-processing and data quality assessment methods, along with 7 challenges encountered during these processes. Based on our findings, we discuss the implications for researchers and practitioners aiming to construct high-quality training datasets for optimizing LLMs.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.236 | 0.557 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.005 | 0.006 |
| Science and technology studies | 0.003 | 0.006 |
| Scholarly communication | 0.012 | 0.021 |
| Open science | 0.004 | 0.006 |
| Research integrity | 0.004 | 0.005 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".