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.
TPU-KNN: K Nearest Neighbor Search at Peak FLOP/s
Authors: Felix Chern, Blake Hechtman, Andy Davis, Ruiqi Guo, David Majnemer, Sanjiv Kumar
NeurIPS 2022 | Venue PDF | LLM Run Details | Input Tokens: 16,995 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 6,163 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 | We conduct experiments verifying our TPU implementation of the algorithm accurately aligned with the performance model and achieves state-of-the-art speed-recall trade-offs on standard nearest neighbor search benchmarks. |
| Researcher Affiliation | Industry | Felix Chern Google Research EMAIL Blake Hechtman Google Core ML EMAIL Andy Davis Google Core ML EMAIL Ruiqi Guo Google Research EMAIL David Majnemer Google Core ML EMAIL Sanjiv Kumar Google Research EMAIL |
| Pseudocode | Yes | Algorithm 1: Partial Reduce for MIPS |
| Open Source Code | Yes | Our work is available in the open-source package of Jax and Tensorflow on TPU. |
| Open Datasets | Yes | We select the Glove4 (Pennington et al., 2014) and Sift5 (Jegou et al., 2010) datasets from the ANN benchmarks. Their corresponding distances are the cosine distance and the Euclidean distance. See the code snippets in Appendix A.1 and A.2. (Footnote 4: Released in Apache license 2.0. Footnote 5: Released in CC0 public domain.) |
| Dataset Splits | No | Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [N/A] |
| Hardware Specification | Yes | In particular, we are interested in the novel platforms that deliver high FLOP/s for distance computation, namely Google TPU V3, V4, Nvidia GPU V100, and A100 in our analysis and evaluation. Table 1: Hardware specifications for the generalized roofline model |
| Software Dependencies | No | The paper mentions using Jax and TensorFlow, but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | No | Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [N/A] |