Paikallisten suurien kielimallinen hyödyntäminen liiketoiminnan sovelluksissa
Bibliographic record
Abstract
Large Language Models (LLMs) have gained popularity in various use cases due to their capabilities. Currently third party services are commonly used, but these solutions contain drawbacks, such as data privacy concerns. Due to this, the interest for local solutions has increased causing international companies to release their own models, which are comparable to closed solutions. This thesis explores how local large language models can be utilized for business applications. The goal of this thesis is to form a comprehensive view of the state of LLMs, including their capabilities and limitations by researching them from various sources. Additionally, experiments are conducted to analyze the inference requirements, assess the impact of quantization on them, evaluate the language capabilities of the models and determine their capability to follow instructions and generate coherent output. The experiments include applying Retrieval Augmented Generation (RAG) using internal company data and fine-tuning a model to improve language capabilities with limited computational resources. As a part of research, a customized method was created and is used to evaluate the effectiveness of retrieval augmented generation. This is done by automatically creating a question-answer dataset with over a thousand entries. The dataset can be used by an LLM to evaluate the factuality and relevance of the context or the model output. The result of the thesis is a comprehensive study of current LLMs, tools and methods, which can be applied as a foundation to build new products in the future. The results indicate that LLMs are suitable for many use cases, although they do have limitations.
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 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) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.006 | 0.004 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.014 | 0.124 |
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; both teacher heads agree on what is shown here.
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".