Update README.md
Browse files
README.md
CHANGED
@@ -118,7 +118,7 @@ Recently, IBM has introduced GneissWeb; a large dataset yielding around 10 trill
|
|
118 |
or with [IBM Data Prep Kit](https://github.com/IBM/data-prep-kit/) (DPK) (please refer to the [example notebook](https://github.com/IBM/data-prep-kit/blob/dev/transforms/language/gneissweb_classification/gneissweb_classification.ipynb) for using a fastText model with DPK).
|
119 |
|
120 |
The GneissWeb ensemble filter uses the confidence score given to `__label__hq` for filtering documents based on an appropriately chosen threshold.
|
121 |
-
The fastText
|
122 |
|
123 |
2. [Bloom filter](https://huggingface.co/ibm-granite/GneissWeb.bloom) built on the document ids of GneissWeb documents. This can be used to recreat GneissWeb using the document ids from FineWeb 1.1.0 or any version of Common Crawl
|
124 |
|
|
|
118 |
or with [IBM Data Prep Kit](https://github.com/IBM/data-prep-kit/) (DPK) (please refer to the [example notebook](https://github.com/IBM/data-prep-kit/blob/dev/transforms/language/gneissweb_classification/gneissweb_classification.ipynb) for using a fastText model with DPK).
|
119 |
|
120 |
The GneissWeb ensemble filter uses the confidence score given to `__label__hq` for filtering documents based on an appropriately chosen threshold.
|
121 |
+
The fastText models are used together along with other quality annotators.
|
122 |
|
123 |
2. [Bloom filter](https://huggingface.co/ibm-granite/GneissWeb.bloom) built on the document ids of GneissWeb documents. This can be used to recreat GneissWeb using the document ids from FineWeb 1.1.0 or any version of Common Crawl
|
124 |
|