Multilingual Fault Localization for Deep Learning Compilers
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
Deep learning compilers play an increasingly important role in implementing learned algorithms efficiently.These compilers are commonly implemented using a set of different programming languages: languages suitable for manipulating high-level tensor graph representations differ from those used to implement efficient low-level operations on accelerator devices.Finding faults in these compilers remains a challenging problem, and previously proposed fault localization techniques have limitations when working with a multilingual codebase.To overcome the aforementioned limitations, this thesis proposes a multilingual fault localization technique based on a language-independent approach to mutant generation.We evaluated this technique using eleven real faults in a deep learning compiler codebase.The results of the empirical evaluation show that the proposed approach can precisely locate four of the eleven faults and correctly ranks the faulty elements as the most suspicious.i B MFL Mutation Operators 88 C DLA Fault Localization Results 90 vi List of Tables 3.1 Notation for spectra metrics collected during program execution. . . .3.2 Notation for mutant metrics computed from mutant test results. . . .5.1 Suspiciousness scores of highest-ranking mutants for defective mid programs in different languages. . . . . . . . . . . . . . .
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 | 0.001 |
| 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 | 0.000 |
| 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