Create README.md

README.md

@@ -0,0 +1,27 @@
+**Stats about bigcode dataset:**
+
+* Permissive licenses only :
+
+
+ |Language |Raw (only exact dedup)  | Near dedup | Near dedup + content filters| Near dedup + more near-dedup (1)|Near dedup + comments filter (1)| Near dedup + more near-dedup + comments(1)|
+|-------|--------|-------|--------|--------|--------|--------|
+|Python | 200 GB| [80 GB](https://huggingface.co/datasets/bigcode/the-stack-dedup-pjj/tree/v1.1.a1) | [75.61 GB](https://huggingface.co/datasets/bigcode/the-stack-pjjs-no-pii-filtered) | [61.97 GB](https://huggingface.co/datasets/bigcode/stack-dedup-alt-filter-no-pii) | [65 GB](https://huggingface.co/datasets/bigcode/the-stack-comments-filter) | (?) less than 60 |
+|Java | 266 GB |112 GB |110 GB | 88 GB | 92 GB |(?) less than 80 |
+|JavaScript | 496 GB | 166 GB |83 GB| 65 GB | 75 GB |(?) less than 60|
+|C | 255 GB | 75 GB |73 GB (2)| - | - |-|
+|C++ | 215 GB | 65 GB |61 GB (2)|- | - |-|
+
+* Non permissive data, the number are higher than the  e.g 240GB of python I mentionned (it was on old dump)
+|Language |Raw (only exact dedup)  | Near dedup |
+|-------|--------|-------|
+| Python (3)| 737 GB | - |
+|Java | 1.3 TB | - |
+|JavaScript | 5.8 TB | -|
+|C | 1.64 TB | - |
+|C++ | 644 GB | - |
+
+(1) all these runs have content filtering, notice that it removes a lot of data from JavaScript (you could try filtering with less strict thresholds, the [script](https://github.com/bigcode-project/bigcode-dataset/tree/main/preprocessing) is very easy to run)
+
+(2) I don't have the data but I found these numbers from an old run on the forst version of The Stack (it uses the same content filtering thresholds as for python, java and js)
+
+(3) this went through content filters