Upload 3 files

README.md

@@ -26,23 +26,99 @@ pretty_name: Jomleh
 
 ## Dataset Summary
 
-"Jomleh" is a high-quality Farsi language dataset consisting of sentences that have been carefully preprocessed to ensure they contain only Farsi characters and no contamination from other languages. The data has been sourced from multiple sources and undergone a deduplication process to ensure that each sentence is unique. While the text in the dataset is not original, the focus on quality over quantity ensures that each sentence is useful and informative. Each sample in "Jomleh" is a sentence, making it a valuable resource for natural language processing tasks and language modeling.
+"Jomleh" is a high-quality Farsi language dataset consisting of sentences that have been carefully preprocessed to ensure they contain only Farsi characters, without any contamination from other languages. The data has been sourced from multiple sources and undergone a deduplication process to ensure that each sentence is unique. While the text in the dataset is not original, the focus on quality over quantity ensures that each sentence is useful and informative. Each sample in "Jomleh" is a sentence, making it a valuable resource for natural language processing tasks and language modeling.
 
 ## Source Data
 
 The data used to curate Jomleh is taken from the following sources:
 
-- OSCAR (fa)
-- CommonCrawl
-- Leipzig
-- VOA Persian
-- Persian poems corpus
-- Web to Corpus
-- TEP: Tehran English-Persian parallel corpus
+- [OSCAR](https://huggingface.co/datasets/oscar) (fa):
+  * [OSCAR-2109](https://huggingface.co/datasets/oscar-corpus/OSCAR-2109)
+  * [OSCAR-2201](https://huggingface.co/datasets/oscar-corpus/OSCAR-2201)
+  * [OSCAR-2301](https://huggingface.co/datasets/oscar-corpus/OSCAR-2301)
+- [CommonCrawl](https://storage.googleapis.com/danielk-files/farsi-text/merged_files/commoncrawl_fa_merged.txt)
+- [Leipzig](https://wortschatz.uni-leipzig.de/en/download/Iranian%20Persian):
+  * Community:
+      - Year: 2017 -> Alle
+  * Web
+      - Year: 2011, Country: Iran -> 10K, 30K, 100K
+      - Year: 2015, Country: Iran -> 10K, 30K, 100K
+      - Year: 2019, Country: Iran -> 10K, 30K, 100K, 300K, 1M
+  *  Web-public
+      - Year: 2019, Country: Iran -> 10K, 30K, 100K, 300K, 1M
+  *  Web.public
+      - Year: 2019, Country: Iran -> 10K, 30K, 100K, 300K, 1M
+  * Wikipedia
+      - Year: 2016, Country: Iran -> 10K, 30K, 100K, 300K, 1M
+      - Year: 2021, Country: Iran -> 10K, 30K, 100K, 300K, 1M
+- [VOA Persian](https://jon.dehdari.org/corpora/)
+- [Persian poems corpus](https://github.com/amnghd/Persian_poems_corpus)
+- [Web to Corpus](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0022-6133-9)
+- [TEP](https://opus.nlpl.eu/TEP.php): Tehran English-Persian parallel corpus
+
+### Number of samples contributed by each source
+
+| Source | Code | Number of samples |
+|----|----|-----:|
+| OSCAR | oscar_2109 | 3,628,547 |
+| OSCAR | oscar_2201 | 2,679,904 |
+| OSCAR | oscar_2301 | 3,604,914 |
+| CommonCrawl | cc | 1,127,690 |
+| Leipzig | web-2019_1M | 19,203 |
+| Leipzig | web-2019_10K | 160 |
+| Leipzig | web-2019_30K | 494 |
+| Leipzig | web-2019_100K | 1,782 |
+| Leipzig | web-2019_300K | 5,355 |
+| Leipzig | news_2019_10K | 171 |
+| Leipzig | news_2019_30K | 499 |
+| Leipzig | news_2019_100K | 1,614 |
+| Leipzig | news_2019_300K | 3,818 |
+| Leipzig | news_2020_10K | 117 |
+| Leipzig | news_2020_30K | 392 |
+| Leipzig | news_2020_100K | 1,287 |
+| Leipzig | news_2020_300K | 3,277 |
+| Leipzig | newscrawl_2011_1M | 21,285 |
+| Leipzig | newscrawl_2015_1M | 21,061 |
+| Leipzig | newscrawl_2015_10K | 167 |
+| Leipzig | newscrawl_2015_30K | 529 |
+| Leipzig | newscrawl_2015_100K | 1,743 |
+| Leipzig | newscrawl_2015_300K | 5,286 |
+| Leipzig | newscrawl_2016_1M | 16,779 |
+| Leipzig | newscrawl_2016_10K | 96 |
+| Leipzig | newscrawl_2016_30K | 337 |
+| Leipzig | newscrawl_2016_100K | 1,065 |
+| Leipzig | newscrawl_2016_300K | 3,105 |
+| Leipzig | newscrawl_2017_1M | 12,222 |
+| Leipzig | newscrawl_2017_10K | 69 |
+| Leipzig | newscrawl_2017_30K | 187 |
+| Leipzig | newscrawl_2017_100K | 712 |
+| Leipzig | newscrawl_2017_300K | 1,968 |
+| Leipzig | newscrawl_2019_1M | 14,805 |
+| Leipzig | newscrawl_2019_10K | 96 |
+| Leipzig | newscrawl_2019_30K | 272 |
+| Leipzig | newscrawl_2019_100K | 916 |
+| Leipzig | newscrawl_2019_300K | 2,674 |
+| Leipzig | wikipedia_2010_10K | 115 |
+| Leipzig | wikipedia_2010_30K | 323 |
+| Leipzig | wikipedia_2010_100K | 984 |
+| Leipzig | wikipedia_2010_300K | 2,415 |
+| Leipzig | wikipedia_2012_10K | 81 |
+| Leipzig | wikipedia_2012_30K | 244 |
+| Leipzig | wikipedia_2012_100K | 732 |
+| Leipzig | wikipedia_2012_300K | 1,929 |
+| Leipzig | wikipedia_2014_1M | 6,999 |
+| Leipzig | wikipedia_2014_10K | 25 |
+| Leipzig | wikipedia_2014_30K | 101 |
+| Leipzig | wikipedia_2014_100K | 307 |
+| Leipzig | wikipedia_2014_300K | 857 |
+| VOA Persian | voa | 5,836 |
+| Persian poems corpus | poems | 51,189 |
+| Web to Corpus| w2c | 88,899 |
+| TEP | tep | 24,602 |
 
 ## Layout and Structure
 
-The dataset is composed of 60 json-line files. As the samples are spread across these files randomly (uniform), the number of samples per each file is not an exact number but generally speaking, there are roughly an equal number of samples per each file.
+The dataset is composed of 60 JSON-line files. As the samples are spread across these files randomly (using a uniform distribution), the number of samples per each file is not an exact number but generally speaking, there are roughly an equal number of samples per each file (roughly 190,000 samples per file).
 
 Each line of a file is a sample formatted in JSON with the following layout:
 
@@ -50,7 +126,7 @@ Each line of a file is a sample formatted in JSON with the following layout:
 {
   "id": <A sequential integer>,
   "text": "<A Farsi sentence>",
-  "source": "<One of: []>"
+  "source": "<One of codes mentioned on the table above>"
 }
 ```
 
@@ -58,9 +134,9 @@ Each line of a file is a sample formatted in JSON with the following layout:
 
 ### 1. Preprocessing
 
-The value of this dataset is its preprocessing of the text. The main struggle working with Farsi text is the fact that due to some historical challenges, there are so many different codings out there used to save Farsi text. On top of that, you can add the complexity of dealing with multiple character codes for the same letter. In Farsi, the look of a character depends on its neighbouring characters. For example, consider the final letter of Farsi alphabet "Ye":
+The value of this dataset is its preprocessing step. The main struggle working with Farsi text is the fact that due to some historical challenges, there are so many different codings out there used to save Farsi text. On top of that, you can add the complexity of dealing with multiple character codes for the same letter. In Farsi, the look of a character depends on its neighbouring characters. For example, consider the very last letter of Farsi alphabet "Ye":
 
-It has a standalone shape:
+It has a standalone form:
 
 <pre><font size="7">&#64508;</font></pre>
 
@@ -68,29 +144,29 @@ But when surronded with other characters, its middle form is used:
 
 <pre><font size="7">&#64511;</font></pre>
 
-This requirement is usually taken care of by "substitution table" feature of the fonts. Which will help show the correct form of the letters. But at the same time, some text don't rely on the fonts and use the specific code designed for the specific form of the letters directly. From the reader point of the view, both will look identical but printing the code, you'll have different numbers. This complicates text processing in Farsi since we need to identify each character with a unique code regardless of their position in the word. On top of that, add the problem of using Arabic characters which some times are used to type Farsi text. Again, since the two languages share very similar alphabets (visually speaking), one can successfully read a text in Farsi while it's been typed using Arabic characters.
+This requirement is usually taken care of by "substitution table" which is a feature of the fonts. This will help show the correct form of the letters according to its positioning in the word. But at the same time, some text don't rely on the fonts and use the specific code designed for the specific form of the letters directly. From the readers' point of the view, both will look identical but printing the code, you'll have different numbers. This complicates text processing in Farsi since we need to identify each character with a unique code regardless of their position in the word. On top of that, add the problem of using Arabic characters which some times are used to type Farsi text. Again, since the two languages share very similar alphabets (visually speaking), one can successfully read a text in Farsi while it's been typed using Arabic characters.
 
 To address these problems, the preprocessing used in Jomle tries its best to map all the different characters that look alike to their Farsi counterpart. This is not an exact science but based on the best effort. For instance, if a sentence is actually an Arabic sentence, the preprocessing script used here will make things worse. But assuming that all the text used here as source are 100% Farsi, this script should help make them uniform.
 
-The same cleaning process also includes digits and puncuations.
+The same cleaning process is also applied to digits and puncuations.
 
-At the end, any character that can be found in the Jomleh dataset is either:
+At the end, any character that can be found in the Jomleh dataset is one of the following:
 
-- a Farsi alphabet letter (`ا` to `ی`) or
-- one of the: `آ`, `أ`, `ؤ`, `ئ` or
-- a Farsi digit (`۹` to `۰`) or
-- a zero-width non-joiner (`\u200c`) or
-- a space or
+- a Farsi alphabet letter (`ا` to `ی`)
+- one of the: `آ`, `أ`, `ؤ`, `ئ`
+- a Farsi digit (`۹` to `۰`)
+- a zero-width non-joiner (`\u200c`)
+- a space
 - one of the Farsi punctuations (`.`, `!`, `؟`, `،`, `؛`)
 
-Any other character found in the text is eliminated based on best effort and if the elimination of such characters could harm the integrity and the meaning of the sentence, then that sentence is removed from the dataset altogether.
+Any other character found in the text is eliminated based on best effort and if the elimination of such characters could harm the integrity of the sentence, then that sentence is removed from the dataset altogether.
 
 The script used for the preprocessing can be found [here](/datasets/mlengineer-ai/jomleh/blob/main/preprocess.py).
 
-It's also worth mentioning that the preprocessing script will convert the text into vertical format which is expected by the third step (deduplication). Simply put, vertical format replaces spaces with a line feed. And also surround it with a `<doc>` tag. Here's an example sample converted into vertical format:
+It's also worth mentioning that the preprocessing script will convert the text into vertical format which is expected by the third step (deduplication). Simply put, in vertical format spaces are replaced with a line feed. And also they are surrounded with a `<doc>` tag. Here's an example sample converted into vertical format:
 
 ```
-<doc id="poems_merged.txt_3">
+<doc id="poems_merged.txt">
 این
 درگه
 ما
@@ -100,12 +176,12 @@ It's also worth mentioning that the preprocessing script will convert the text i
 </doc>
 ```
 
-The `id` attribute of the `<doc>` tag points to the file where the sample is coming from.
+In this example, the `id` attribute of the `<doc>` tag points to the file where the sample is coming from.
 
 This is the command that executes the preprocessing script:
 
 ```
-find 1_prepared -name "*.txt" | parallel 'python ./processing/preprocess.py $(basename {}) < {} > ./2_cleaned_vertical/$(basename {})'
+find 1_prepared -name "*.txt" | parallel 'python ./preprocess.py $(basename {}) < {} > ./2_cleaned_vertical/$(basename {})'
 ```
 
 ### 2. Merging into one text file
@@ -118,7 +194,7 @@ cat ./2_cleaned_vertical/* > ./3_temp/clean_merged.vert
 
 ### 3. Deduplication
 
-Once all the text is transformed into vertical format and saved in a single text file, `onion` program is used to eliminate any duplicate samples. You can find the onion program from [this website](https://corpus.tools/wiki/Onion) and it is used here like this:
+Once all the text is transformed into vertical format and saved into a single text file, the `onion` program is used to eliminate any duplicate samples. You can find the onion program from [this website](https://corpus.tools/wiki/Onion) and it is used here like this:
 
 ```
 onion -sm -n 5 -t 0.5 ./3_temp/clean_merged.vert > ./3_temp/deduplicated.vert
@@ -128,39 +204,44 @@ onion -sm -n 5 -t 0.5 ./3_temp/clean_merged.vert > ./3_temp/deduplicated.vert
 
 The postprocessing involves:
 
-1. Converting back from vertical format into a single line per sample.
-2. Mapping the file name mentioned in the `id` attribute of the `<doc>` tag into a simpler text which can be found in the [postprocessing script](/datasets/mlengineer-ai/jomleh/blob/main/postprocessing.py).
-3. Formatting each sample as a JSON-line (one json per line)
-4. Distributing and saving the sample unifomrly across 60 files, trying to get relatively same number of samples per file.
-5. Collection some statistics along the way.
+1. Converting back from vertical format into a single line per sample format.
+2. Mapping the file names mentioned in the `id` attribute of the `<doc>` tag into one of the codes mentioned above.
+3. Formatting each sample as a JSON-line (one json per line).
+4. Distributing and saving the samples randomly across 60 files, trying to get relatively same number of samples per file.
 
 These steps are run using the following command:
 
 ```
-python ./postprocess.py ./3_temp < ./3_temp/deduplicated.vert | parallel "echo '{}' | python ./processing/add_id.py ./3_temp ./jomleh/files"
+python ./postprocess.py ./3_temp < ./3_temp/deduplicated.vert | parallel "echo '{}' | python ./add_id.py ./3_temp ./jomleh/files"
 ```
 
 ### 5. Compressing the files
 
-This can be done using the following command:
+The generated JSON-line files are compressed using Zstandard - Real-time data compression algorithm:
 
 ```
-find ./jomleh/files/*.jsonl -type f | parallel 'zstd {}'
+find ./jomleh/files/*.jsonl -type f | parallel 'zstd --rm {}'
 ```
 
 ### 6. Generating the checksum file
 
+The checksum file plays a dual role. Firstly, it keeps the checksum for each of 60 files for future verification. And also, it plays the role of index so the script can list and load the files. This is how the checksum file is generated:
+
 ```
 ls ./jomleh/files/*.zst | sort -t _ -k 2 -n | xargs sha256sum > ./jomleh/files/checksum.sha256
 ```
 
-After applying all these steps, we are left a dataset with these characteristics:
+## Statistics
 
-| | Some statistics on the collected sentences |
+After applying all the steps mentioned above, the curated dataset has the following statistics:
+
+| | Statistics on the collected sentences |
 |---:|:---|
-| Total number of sentences: | 123 | 
-| Average number of characters in a sentence: | 123 |
-| Average number of words in a sentence: | 123 |
-| Standard devitaion for the number of words in a sentence: | 123 |
-| Average number of letters in a word: | 123 |
-| Standard devitaion for the number of letters in a word: | 123 |
+| Total number of sentences: | 11,370,236 |
+| Average number of characters in a sentence: | 101.17 |
+| Standard deviation of the number of characters in a sentence: | 88.16 |
+| Average number of words in a sentence: | 19.93 |
+| Standard devitaion of the number of words in a sentence: | 17.45 |
+| Average number of characters in a word: | 4.12 |
+| Standard devitaion of the number of characters in a word: | 1.99 |
+

postprocess.py

@@ -0,0 +1,193 @@
+import sys
+import math
+import re
+import random
+import json
+from pathlib import Path
+
+
+__FILE_COUNT__ = 60
+doc_regex = re.compile("<doc id=\"([^\"]+)_\\d+\">")
+
+file_names = []
+file_pointers = {}
+record_counter = {}
+
+line_counter = 0
+sum_token_count = 0
+sum_token_sq = 0
+sum_char_count = 0
+sum_char_sq = 0
+source_dist = {}
+dataset_names = {
+    "2109_0.txt": "oscar_2109",
+    "2109_1.txt": "oscar_2109",
+    "2109_2.txt": "oscar_2109",
+    "2109_3.txt": "oscar_2109",
+    "2109_4.txt": "oscar_2109",
+    "2109_5.txt": "oscar_2109",
+    "2109_6.txt": "oscar_2109",
+    "2109_7.txt": "oscar_2109",
+    "2109_8.txt": "oscar_2109",
+    "2109_9.txt": "oscar_2109",
+    "2201_0.txt": "oscar_2201",
+    "2201_1.txt": "oscar_2201",
+    "2201_2.txt": "oscar_2201",
+    "2201_3.txt": "oscar_2201",
+    "2201_4.txt": "oscar_2201",
+    "2201_5.txt": "oscar_2201",
+    "2201_6.txt": "oscar_2201",
+    "2201_7.txt": "oscar_2201",
+    "2301_0.txt": "oscar_2301",
+    "2301_10.txt": "oscar_2301",
+    "2301_11.txt": "oscar_2301",
+    "2301_1.txt": "oscar_2301",
+    "2301_2.txt": "oscar_2301",
+    "2301_3.txt": "oscar_2301",
+    "2301_4.txt": "oscar_2301",
+    "2301_5.txt": "oscar_2301",
+    "2301_6.txt": "oscar_2301",
+    "2301_7.txt": "oscar_2301",
+    "2301_8.txt": "oscar_2301",
+    "2301_9.txt": "oscar_2301",
+    "commoncrawl_fa_merged_aa.txt": "cc",
+    "commoncrawl_fa_merged_ab.txt": "cc",
+    "commoncrawl_fa_merged_ac.txt": "cc",
+    "commoncrawl_fa_merged_ad.txt": "cc",
+    "commoncrawl_fa_merged_ae.txt": "cc",
+    "commoncrawl_fa_merged_af.txt": "cc",
+    "commoncrawl_fa_merged_ag.txt": "cc",
+    "commoncrawl_fa_merged_ah.txt": "cc",
+    "commoncrawl_fa_merged_ai.txt": "cc",
+    "commoncrawl_fa_merged_aj.txt": "cc",
+    "fas-ir_web-public_2019_100K-sentences.txt": "web-2019_100K",
+    "fas-ir_web-public_2019_10K-sentences.txt": "web-2019_10K",
+    "fas-ir_web-public_2019_1M-sentences.txt": "web-2019_1M",
+    "fas-ir_web-public_2019_300K-sentences.txt": "web-2019_300K",
+    "fas-ir_web-public_2019_30K-sentences.txt": "web-2019_30K",
+    "fas_news_2019_100K-sentences.txt": "news_2019_100K",
+    "fas_news_2019_10K-sentences.txt": "news_2019_10K",
+    "fas_news_2019_300K-sentences.txt": "news_2019_300K",
+    "fas_news_2019_30K-sentences.txt": "news_2019_30K",
+    "fas_news_2020_100K-sentences.txt": "news_2020_100K",
+    "fas_news_2020_10K-sentences.txt": "news_2020_10K",
+    "fas_news_2020_300K-sentences.txt": "news_2020_300K",
+    "fas_news_2020_30K-sentences.txt": "news_2020_30K",
+    "fas_newscrawl_2011_100K-sentences.txt": "newscrawl_2011_100K",
+    "fas_newscrawl_2011_10K-sentences.txt": "newscrawl_2011_10K",
+    "fas_newscrawl_2011_1M-sentences.txt": "newscrawl_2011_1M",
+    "fas_newscrawl_2011_300K-sentences.txt": "newscrawl_2011_300K",
+    "fas_newscrawl_2011_30K-sentences.txt": "newscrawl_2011_30K",
+    "fas_newscrawl_2015_100K-sentences.txt": "newscrawl_2015_100K",
+    "fas_newscrawl_2015_10K-sentences.txt": "newscrawl_2015_10K",
+    "fas_newscrawl_2015_1M-sentences.txt": "newscrawl_2015_1M",
+    "fas_newscrawl_2015_300K-sentences.txt": "newscrawl_2015_300K",
+    "fas_newscrawl_2015_30K-sentences.txt": "newscrawl_2015_30K",
+    "fas_newscrawl_2016_100K-sentences.txt": "newscrawl_2016_100K",
+    "fas_newscrawl_2016_10K-sentences.txt": "newscrawl_2016_10K",
+    "fas_newscrawl_2016_1M-sentences.txt": "newscrawl_2016_1M",
+    "fas_newscrawl_2016_300K-sentences.txt": "newscrawl_2016_300K",
+    "fas_newscrawl_2016_30K-sentences.txt": "newscrawl_2016_30K",
+    "fas_newscrawl_2017_100K-sentences.txt": "newscrawl_2017_100K",
+    "fas_newscrawl_2017_10K-sentences.txt": "newscrawl_2017_10K",
+    "fas_newscrawl_2017_1M-sentences.txt": "newscrawl_2017_1M",
+    "fas_newscrawl_2017_300K-sentences.txt": "newscrawl_2017_300K",
+    "fas_newscrawl_2017_30K-sentences.txt": "newscrawl_2017_30K",
+    "fas_newscrawl_2019_100K-sentences.txt": "newscrawl_2019_100K",
+    "fas_newscrawl_2019_10K-sentences.txt": "newscrawl_2019_10K",
+    "fas_newscrawl_2019_1M-sentences.txt": "newscrawl_2019_1M",
+    "fas_newscrawl_2019_300K-sentences.txt": "newscrawl_2019_300K",
+    "fas_newscrawl_2019_30K-sentences.txt": "newscrawl_2019_30K",
+    "fas_wikipedia_2010_100K-sentences.txt": "wikipedia_2010_100K",
+    "fas_wikipedia_2010_10K-sentences.txt": "wikipedia_2010_10K",
+    "fas_wikipedia_2010_300K-sentences.txt": "wikipedia_2010_300K",
+    "fas_wikipedia_2010_30K-sentences.txt": "wikipedia_2010_30K",
+    "fas_wikipedia_2012_100K-sentences.txt": "wikipedia_2012_100K",
+    "fas_wikipedia_2012_10K-sentences.txt": "wikipedia_2012_10K",
+    "fas_wikipedia_2012_300K-sentences.txt": "wikipedia_2012_300K",
+    "fas_wikipedia_2012_30K-sentences.txt": "wikipedia_2012_30K",
+    "fas_wikipedia_2014_100K-sentences.txt": "wikipedia_2014_100K",
+    "fas_wikipedia_2014_10K-sentences.txt": "wikipedia_2014_10K",
+    "fas_wikipedia_2014_1M-sentences.txt": "wikipedia_2014_1M",
+    "fas_wikipedia_2014_300K-sentences.txt": "wikipedia_2014_300K",
+    "fas_wikipedia_2014_30K-sentences.txt": "wikipedia_2014_30K",
+    "poems_merged.txt": "poems",
+    "TEP_fa.txt": "tep",
+    "voa_persian_2003_2008_cleaned.txt": "voa",
+    "w2c_merged.txt": "w2c",
+}
+
+
+def stats(tokens):
+    global line_counter, sum_token_count, sum_token_sq, sum_char_count, sum_char_sq
+    line_counter = line_counter + 1
+    sum_token_count = sum_token_count + len(tokens)
+    sum_token_sq = sum_token_sq + len(tokens) * len(tokens)
+    sum_char = sum([len(t) for t in tokens])
+    sum_char_count = sum_char_count + sum_char
+    sum_char_sq = sum_char_sq + sum_char * sum_char
+
+
+output_folder = sys.argv[1]
+Path(output_folder).mkdir(parents=True, exist_ok=True)
+
+for i in range(__FILE_COUNT__):
+    fn = f"jomleh_{i+1}.jsonl"
+    file_names.append(fn)
+    # file_pointers[fn] = open(f'{output_folder}/jomleh_{i+1}.jsonl', 'w')
+    record_counter[fn] = 0
+
+seen = set()
+tokens = []
+for token in sys.stdin:
+    token = token.strip()
+    if token.startswith("<doc"):
+        tokens = []
+        doc_id = doc_regex.match(token).groups()[0]
+        ds_name = dataset_names[doc_id] if doc_id in dataset_names else doc_id
+        source_dist[ds_name] = source_dist.get(ds_name, 0) + 1
+        continue
+    if token == "</doc>":
+        sentence = " ".join(tokens)
+        if len(tokens) >= 10:
+            stats(tokens)
+            jsonl = json.dumps({"source": ds_name, "text": sentence}, ensure_ascii=False)
+            fn = random.sample(file_names, 1)[0]
+            # file_pointers[fn].write(jsonl + "\n")
+            record_counter[fn] += 1
+        elif sentence not in seen:
+            seen.add(sentence)
+            stats(tokens)
+            jsonl = json.dumps({"source": ds_name, "text": sentence}, ensure_ascii=False)
+            fn = random.sample(file_names, 1)[0]
+            # file_pointers[fn].write(jsonl + "\n")
+            record_counter[fn] += 1
+        continue
+    tokens.append(token)
+
+# for i in range(__FILE_COUNT__):
+#     file_pointers[file_names[i]].close()
+
+avg_tokens = sum_token_count / line_counter
+stddev_tokens = math.sqrt((sum_token_sq / line_counter) - avg_tokens * avg_tokens)
+avg_char = sum_char_count / sum_token_count
+stddev_chars = math.sqrt((sum_char_sq / sum_token_count) - avg_char * avg_char)
+
+results = {
+    "Number of records per each file": record_counter,
+    "Number of samples from each source": source_dist,
+    "Number of lines": line_counter,
+    "Total number of words": sum_token_count,
+    "Average number of tokens per line": avg_tokens,
+    "Standard deviation for the number of tokens per line": stddev_tokens,
+    "Average number of characters per token": avg_char,
+    "Standard deviation for the number of characters per token": stddev_chars,
+}
+
+print(json.dumps(results))
+# print(json.dumps(results), sys.stderr)
+
+# offset = 1
+# for fn in file_names:
+#     print(json.dumps({"filename": fn, "first_id": offset}))
+#     offset += record_counter[fn]