Downloading script

README.md

@@ -264,6 +264,79 @@ wget https://data.hplt-project.org/three/sorted/eng_Latn.map
 To speed up large downloads, it can be beneficial to use multiple parallel connections, for example using the `--max-threads` option in `wget`. 
 We recommend to limit download parallelization to 16–32 threads, to avoid server-side rate limitations, which should allow download rates of around 250 gigabytes per hour.
 
+## Downloading with HuggingFace Datasets
+To load a monolingual portion of the **HPLT v3.0** dataset in the _Huggingface Datasets_ format, you can run the following code to download the dataset files map and then load the `.jsonl.zst` files directly using `load_datasets()`. 
+The `Datasets` package will then handle downloading of the files. 
+If you would like to **stream** the files instead of downloading them all at once, set `streaming=True` within the `load_dataset()` function.
+
+```
+from datasets import load_dataset, Features, Value, Sequence, List
+import requests
+
+lang_code = "yor_Latn"  # Define your language-script code here, or "multilingual" for full multilingual portion minus English
+
+r = requests.get(f"https://data.hplt-project.org/three/sorted/{lang_code}.map")
+
+source_urls = r.text.strip().split("\n")
+
+features = Features(
+    {
+        "f": Value("string"),
+        "o": Value("int64"),
+        "s": Value("int64"),
+        "rs": Value("int64"),
+        "u": Value("string"),
+        "c": Value("string"),
+        "ts": Value("timestamp[s]"),
+        "de": Value("string"),
+        "crawl_id": Value("string"),
+        "lang": List(Value("string")),
+        "prob": List(Value("float64")),
+        "text": Value("string"),
+        "xml": Value("string"),
+        "html_lang": List(Value("string")),
+        "cluster_size": Value("int64"),
+        "seg_langs": List(Value("string")),
+        "id": Value("string"),
+        "filter": Value("string"),
+        "pii": List(List(Value("int64"))),
+        "doc_scores": List(Value("float64")),
+        "web-register": {
+            "MT": Value("float64"),
+            "LY": Value("float64"),
+            "SP": Value("float64"),
+            "ID": Value("float64"),
+            "NA": Value("float64"),
+            "HI": Value("float64"),
+            "IN": Value("float64"),
+            "OP": Value("float64"),
+            "IP": Value("float64"),
+            "it": Value("float64"),
+            "ne": Value("float64"),
+            "sr": Value("float64"),
+            "nb": Value("float64"),
+            "re": Value("float64"),
+            "en": Value("float64"),
+            "ra": Value("float64"),
+            "dtp": Value("float64"),
+            "fi": Value("float64"),
+            "lt": Value("float64"),
+            "rv": Value("float64"),
+            "ob": Value("float64"),
+            "rs": Value("float64"),
+            "av": Value("float64"),
+            "ds": Value("float64"),
+            "ed": Value("float64"),
+        },
+    }
+)
+
+ds = load_dataset("json", data_files=source_urls, features=features)
+
+print(ds)
+```
+
+
 ## New in this release compared to HPLT v2
  - Reflects substantially more raw web data, primarily from the Common Crawl
  - Additional metadata, including more information from the underlying crawl