The Effectiveness of Compact Fine-Tuned LLMs in Log Parsing
Bibliographic record
Abstract
Log parsing is defined as the process of extracting structured information from unstructured log data. It is an important step prior to many log analytics tasks. The emergence of Large Language Models (LLMs), like Generative Pre-trained Transformers (GPTs), has driven the development of novel log parsing methods. Existing studies have examined the effectiveness of large-scale general-purpose LLMs in log parsing. In this paper, we argue that the long-term adoption of such LLMs pose challenges of data privacy, cost, and tool integration. To address these challenges, we explore the viability of supervised fine-tuning of an open-source compact LLM for log parsing as a prospective alternative. To this end, we fine-tune the Mistral-7B-Instruct LLM on a diverse set of log files and evaluate its performance, in terms of both accuracy and robustness, against OpenAI's GPT-4-Turbo using different configuration settings. We apply two evaluation approaches, namely metric-based and LLM-based. Our overall findings show that fine-tuning a compact LLM such as Mistral-7B provides similar and sometimes better results than using a large-scale LLM, in our case GPT-4-Turbo. These findings are important because they enable companies to use a smaller LLM that they can readily adapt to parsing their log data, and integrate into their log analytics tools, without the need to rely on third-party LLM providers.
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 | 0.007 | 0.044 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.003 | 0.007 |
| Open science | 0.004 | 0.003 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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".