Answering User Questions About Machine Learning Models Through Standardized Model Cards
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Reusing pre-trained machine learning models is becoming very popular due to model hubs such as Hugging Face (HF). However, similar to when reusing software, many issues may arise when reusing an ML model. In many cases, users resort to asking questions on discussion forums such as the HF community forum. In this paper, we study how we can reduce the community's workload in answering these questions and increase the likelihood that questions receive a quick answer. We analyze 11,278 discussions from the HF model community that contain user questions about ML models. We focus on the effort spent handling questions, the high-level topics of discussions, and the potential for standardizing responses in model cards based on a model card template. Our findings indicate that there is not much effort involved in responding to user questions, however, 40.1% of the questions remain open without any response. A topic analysis shows that discussions are more centered around technical details on model development and troubleshooting, indicating that more input from model providers is required. We show that 42.5% of the questions could have been answered if the model provider followed a standard model card template for the model card. Based on our analysis, we recommend that model providers add more development-related details on the model's architecture, algorithm, data preprocessing and training code in existing documentation (sub)sections and add new (sub)sections to the template to address common questions about model usage and hardware requirements.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it