Notice: This website is continuously updated as new conference proceedings and journal volumes are published. Consequently, the figures presented here may differ from those reported in our publications.
A brief overview of our methodology is provided on the Methods page, along with references to publications documenting our approach and links to the code and data used in each study.
The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates.
Ex Uno Pluria: Insights on Ensembling in Low Precision Number Systems
Authors: Giung Nam, Juho Lee
NeurIPS 2024 | Venue PDF | LLM Run Details | Input Tokens: 25,405 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 3,426 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our empirical analysis demonstrates the effectiveness of our proposed low precision ensembling method compared to existing ensemble approaches. |
| Researcher Affiliation | Academia | Giung Nam Kim Jaechul Graduate School of AI KAIST, Daejeon, South Korea EMAIL Juho Lee Kim Jaechul Graduate School of AI KAIST, Daejeon, South Korea EMAIL |
| Pseudocode | No | The paper does not contain explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/cs-giung/lpe-bsr. |
| Open Datasets | Yes | We employed two datasets for our experiments: Image Net (Russakovsky et al., 2015) for Vi T and CLIP-Vi T models, and MMLU (Hendrycks et al., 2021) for LLa Ma. |
| Dataset Splits | Yes | Table 1 summarizes the evaluation results on a subset of the Image Net validation split, along with the parameter count for each model. |
| Hardware Specification | Yes | We conducted experiments using TPUv2/v3/v4 cores, with flexibility in selecting the cores based on the memory requirements of each experiment. |
| Software Dependencies | No | We built our experimental code using JAX (Bradbury et al., 2018) and Transformers (Wolf et al., 2020), both licensed under Apache-2.0. |
| Experiment Setup | Yes | The optimization process for CLIP-Vi T-L/14 models concludes after 100,000 iterations with a minibatch size of 64, employing a cosine decaying learning rate schedule. |