Update README.md

README.md

@@ -17,21 +17,26 @@ This dataset contains **100,000 rows** sampled from the `allenai/c4` English spl
 
 ## Methods
 
-### FastText-based classifiers
-The classifiers output probabilities converted to logits log(p). The **quality score** is computed as the difference of logits between the high-quality and low-quality classes.
+Classifiers used:
 
-- [ibm-granite/GneissWeb.Quality_annotator](https://huggingface.co/ibm-granite/GneissWeb.Quality_annotator)
-- [openbmb/Ultra-FineWeb-classifier](https://huggingface.co/openbmb/Ultra-FineWeb-classifier)
+| Label         | Model                                                                 | Method         |
+|---------------|-----------------------------------------------------------------------|----------------|
+| fineweb2hq    | [agentlans/multilingual-e5-small-fineweb2hq-vs-c4-classifier](https://huggingface.co/agentlans/multilingual-e5-small-fineweb2hq-vs-c4-classifier) | Logit difference |
+| finewebedu    | [HuggingFaceFW/fineweb-edu-classifier](https://huggingface.co/HuggingFaceFW/fineweb-edu-classifier)           | Raw logits     |
+| gneiss       | [ibm-granite/GneissWeb.Quality_annotator](https://huggingface.co/ibm-granite/GneissWeb.Quality_annotator)     | FastText       |
+| nemo         | [nvidia/nemocurator-fineweb-nemotron-4-edu-classifier](https://huggingface.co/nvidia/nemocurator-fineweb-nemotron-4-edu-classifier) | Raw logits     |
+| nvidia       | [nvidia/quality-classifier-deberta](https://huggingface.co/nvidia/quality-classifier-deberta)                   | Logit difference |
+| ultrafineweb | [openbmb/Ultra-FineWeb-classifier](https://huggingface.co/openbmb/Ultra-FineWeb-classifier)                     | FastText       |
+| uvp          | [agentlans/bge-small-en-v1.5-ultrafineweb-vs-pile-classifier](https://huggingface.co/agentlans/bge-small-en-v1.5-ultrafineweb-vs-pile-classifier) | Logit difference |
 
-### Transformers-based classifiers
-- [HuggingFaceFW/fineweb-edu-classifier](https://huggingface.co/HuggingFaceFW/fineweb-edu-classifier): raw logits
-- [nvidia/quality-classifier-deberta](https://huggingface.co/nvidia/quality-classifier-deberta): difference of logits (high - low)
-- [agentlans/bge-small-en-v1.5-ultrafineweb-vs-pile-classifier](https://huggingface.co/agentlans/bge-small-en-v1.5-ultrafineweb-vs-pile-classifier): difference of logits
+For FastText-based classifiers, the classifiers' output probabilities were converted to logits log(p).
+The **quality score** for those classifiers were computed as the difference of logits between the high-quality and low-quality classes.
 
 ### Overall score computation
-- Scores from all classifiers were centered and scaled.
+- Scores from all classifiers were centreed and scaled.
 - Principal Components Analysis (PCA) was applied.
-- The **first principal component (PC1)** normalized to z-scores is taken as the overall quality score.
+- The first principal component (PC1) was normalized to z-scores (mean 0, standard deviation 1)
+- The z-score is taken as the overall quality score.
 
 For convenience, the dataset is split into an **80% training set** and a **20% testing set**.
 
@@ -42,10 +47,10 @@ For convenience, the dataset is split into an **80% training set** and a **20% t
 - The **lower triangle** shows pairwise density plots of classifier scores.
 - The **diagonal** presents the distribution of each classifier's scores.
 - The **upper triangle** displays correlations between pairs of classifiers.
-- `uvp` refers to `agentlans/bge-small-en-v1.5-ultrafineweb-vs-pile-classifier`
 
-The classifiers' scores show moderate to strong correlations, except for the Nvidia classifier, which is less correlated.
-The score correlates well with each individual classifier's quality score.
+- The classifiers' scores show moderate to strong correlations, except for the Nvidia classifier, which is less correlated.
+- The custom-trained classifiers often give bimodal distributions instead of smoothly varying values.
+- Despite the above, the overall score correlates well with each individual classifier's quality score.
 
 ## License
 This dataset is licensed under the **[Open Data Commons Attribution License (ODC-BY)](https://opendatacommons.org/licenses/by/)**.