Comments (1)
fine-tuning the parameters rows per band and number of bands in MinHash depends heavily on the data and the specific use case, such as fuzzy deduplication of text pairs in your scenario. These parameters control the trade-off between accuracy and performance (speed and memory usage):
- Accuracy: Increasing r and b generally increases the accuracy of the MinHash similarity estimates, as more hash functions are utilized, capturing more aspects of the data.
- Speed: However, higher values of r and b can slow down the computation, as more hash functions need to be evaluated.
- Memory Usage: Likewise, more memory is required to store the additional hash values.
The ideal settings for r and b could vary depending on:
- The size and nature of your dataset.
- The level of similarity/dissimilarity among text pairs in your dataset.
- The computational resources available to you.
- Your tolerance for error versus your need for speed and lower memory usage.
An empirical approach, where you run experiments with different values of r and b on a representative subset of your data, can be very informative. By analyzing how the performance metrics (speed, memory usage, and accuracy) change with different settings, you can better understand the trade-offs and find an optimal configuration for your specific use case.
from open-nllb.
Related Issues (20)
- Obtain high quality Serbian parallel corpus (currently 0 support in our public bi-text)
- Setup a pipeline for mined data (use Allen AI's OSS dataset replication)
- Understand how to do 4-stage curriculum learning from the paper
- Estimate the necessary compute and number of GPUs for Open-NLLB effort
- Get a compute grant
- [Modeling] Release a 615M HBS (Croatian, Bosnian, Serbian) Open-NLLB checkpoint
- [Modeling] Release a 1.3B Slavic languages Open-NLLB checkpoint
- [Modeling] Release a 3.3B Open-NLLB checkpoint (~202 languages)
- [Modeling] Release a 615M English -> HBS Open-NLLB checkpoint
- [Data] Acquire additional high-quality (non-public) parallel corpora for HBS
- [Future - outside current project scope] 7B lang-family-specific Open-NLLB checkpoint
- [Future - outside current project scope] non-English LLMs (Serbian LLM, etc.)
- Weird line length spikes in Serbian, Croatian, Bosnian (data analysis task) HOT 4
- Creation of a small model file for a few languages
- Standard Moroccan Tamazight is mislabeled
- Provide context for input and output
- Native language visualizations HOT 2
- LID model peak probabilities
- Choosing the LID model
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from open-nllb.