fix typos

README.md

@@ -11,7 +11,7 @@ necessary used for model training.
 - `preprocessing`: code for filtering code datasets based on:
   - line length and percentage of alphanumeric characters (basic filter)
   - number of stars, comments to code ratio, tokenizer fertility
-  - Additionnal filters used for StarCoder Training:
+  - Additional filters used for StarCoder Training:
     - basic-filter with parameters that depend on the file's extension.
     - filter to remove XML files
     - filter for HTML based on displayed-text VS code ratio

pii/README.md

@@ -4,12 +4,12 @@ We provide code to detect Names, Emails, IP addresses, Passwords API/SSH keys in
 ## NER approach
 For the **NER** model based approach (e.g [StarPII](https://huggingface.co/bigcode/starpii)), please go to the `ner` folder. 
 
-We provide the code used for training a PII NER model to detect : Names, Emails, Keys, Passwords & IP addresses (more details in our paper: [StarCoder: May The Source Be With You](https://drive.google.com/file/d/1cN-b9GnWtHzQRoE7M7gAEyivY0kl4BYs/view)).  You will also find the code (and `slurm` scripts) used for running PII Inference on [StarCoderData](https://huggingface.co/datasets/bigcode/starcoderdata), we were able to detect PII in 800GB of text in 800 GPU-hours on A100 80GB. To replace secrets we used teh following tokens:
+We provide the code used for training a PII NER model to detect : Names, Emails, Keys, Passwords & IP addresses (more details in our paper: [StarCoder: May The Source Be With You](https://drive.google.com/file/d/1cN-b9GnWtHzQRoE7M7gAEyivY0kl4BYs/view)).  You will also find the code (and `slurm` scripts) used for running PII Inference on [StarCoderData](https://huggingface.co/datasets/bigcode/starcoderdata), we were able to detect PII in 800GB of text in 800 GPU-hours on A100 80GB. To replace secrets we used the following tokens:
 `<NAME>, <EMAIL>, <KEY>, <PASSWORD>`
 To mask IP addresses, we randomly selected an IP address from 5~synthetic, private, non-internet-facing IP addresses of the same type.
 
 ## Regex approach
-Below we explain the regex based approach to dectect Emails, IP addresses adn keys only:
+Below we explain the regex based approach to detect Emails, IP addresses and keys only:
 We use regexes for emails and IP addresses (they are adapted from [BigScience PII pipeline](https://github.com/bigscience-workshop/data-preparation/tree/main/preprocessing/training/02_pii)). And we use [detect-secrets](https://github.com/Yelp/detect-secrets) for finding secrets keys. We additionally implement some filters on top to reduce the number of false positives. There is also some evaluation code to test the pipeline on a PII benchmark we annotated.
 
 ## Usage of the regex approach

pii/main.py

@@ -182,7 +182,7 @@ def main():
                   'max_forks_repo_path', 'max_forks_repo_name', 'max_forks_repo_head_hexsha', \
                   'max_forks_repo_licenses', 'max_forks_count', 'max_forks_repo_forks_event_min_datetime', 'max_forks_repo_forks_event_max_datetime']
         ds = ds.remove_columns(columns) 
-        logger.info(f"New dataset fomat: {ds}")
+        logger.info(f"New dataset format: {ds}")
     # add id column to dataset
     logger.info(f" ===== Adding an index column =====")
     ds = ds.add_column("index", list(range(len(ds))))

pii/ner/pii_inference/notebooks/Pipeline with sliding-window.ipynb

@@ -296,8 +296,8 @@
    "id": "33a72a0a",
    "metadata": {},
    "source": [
-    "### 5 seconds out 12 (41% of time) consumed on transfering from cpu to cuda \n",
-    "### 7 seconds out 12 (58% of time) consumed on transfering from cuda to cpu \n",
+    "### 5 seconds out 12 (41% of time) consumed on transferring from cpu to cuda \n",
+    "### 7 seconds out 12 (58% of time) consumed on transferring from cuda to cpu \n",
     "### so, most of the time consumption is just a bus bottleneck =/"
    ]
   }

pii/ner/pii_inference/notebooks/Train DeBERTa-v3-base on pseudo-labeled data.ipynb

@@ -4,7 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### The fisrt stage of training \n",
+    "### The first stage of training \n",
     "The initial training on pseudo-labeled data "
    ]
   },

pii/ner/pii_redaction/manual_sharding.py

@@ -27,7 +27,7 @@ def save_manual_shards(
     # this will create a folder OUT_PATH that is a clone of REMOTE_DATASET_REPO
     # you can save the shards inside it and do git add/commit/push to push data to the hub
     out_path = remote_dataset_repo if local_dir is None else local_dir
-    # if out path doesnt already exist
+    # if out path doesn't already exist
     if not os.path.exists(out_path):
         repo = Repository(
             local_dir=out_path,

pii/ner/pii_redaction/utils.py

@@ -106,7 +106,7 @@ def redact_pii_text(text, secrets, replacements, add_references=False):
         secrets (list): list with the secrets to redact
         replacements (dict): dictionary of replacements for each PII type
         add_references (bool): whether to add references to the redacted text (delimiters to PII)
-        for vizualization
+        for visualization
     Returns:
         text (str): new text with redacted secrets
     """

pii/ner/pii_train_ner/README.md

@@ -1,6 +1,6 @@
 # Fine-tuning StarEncoder on an NER task  for PII detection
 
-To run the training on an annotated PII dataset (`bigcode/pii-full-ds` in our case, you might need to adpat the code to fit your dataset), use the following command: 
+To run the training on an annotated PII dataset (`bigcode/pii-full-ds` in our case, you might need to adapt the code to fit your dataset), use the following command: 
 ```bash
 python -m torch.distributed.launch \
         --nproc_per_node number_of_gpus train.py \

pii/ner/pii_train_ner/train.py

@@ -220,7 +220,7 @@ def main(args):
         valid_stats = get_stats(valid_data)
         test_stats = get_stats(test_data)
         print("Train low-resource stats")
-        # print stats for keys with less than 100 in teh value
+        # print stats for keys with less than 100 in the value
         pprint({k: v for k, v in train_stats.items() if v < 300})
         print("Valid low-resource stats")
         pprint({k: v for k, v in valid_stats.items() if v < 100})

pii/notebooks/analysis_detect_secrets_tool.ipynb

@@ -1103,7 +1103,7 @@
    "id": "f8a18b9a",
    "metadata": {},
    "source": [
-    "Conclusion: Gibberish detector can filter the false postives of this detector"
+    "Conclusion: Gibberish detector can filter the false positives of this detector"
    ]
   },
   {

pii/notebooks/data_prefiltering_lightag.ipynb

@@ -1012,7 +1012,7 @@
     "    for j in range(len(labels[\"examples\"][i][\"annotations\"])):\n",
     "        tags.append(labels[\"examples\"][i][\"annotations\"][j][\"tag\"])\n",
     "Counter(tags)\n",
-    "# plot distibution of tags\n",
+    "# plot distribution of tags\n",
     "import matplotlib.pyplot as plt\n",
     "plt.bar(Counter(tags).keys(), Counter(tags).values())\n",
     "# add title to the plot saying tags from 180 annotated files\n",

pii/notebooks/early_qualitative_analysis.ipynb

@@ -23,7 +23,7 @@
     "    * Some hashes are not labelled as gibberish by the gibberish detector(=> not filtered), not sure if they are really secrets, for an example see `./experiments/file_with_hashes.txt` (some other hashes -from that file- are filtered though)\n",
     "    * There are still some false positives like name/path (labeled as gibberish) in this format \"e2e_mask_rcnn_X-152-32x8d-FPN-IN5k_1.44x\" and \"//deno.land/std@0.142.0/testing/asserts.ts\"\n",
     "    * If there is an \"=\" or \"id=\" in front of the key it is detected\n",
-    "    * Some instances like \"f47dbc9c:\" and \"dc22a3aa:\" are detected, tehy seems like ids of patch releases, their context is saved in `./experiments/short_keys_patch_releases.txt`\n",
+    "    * Some instances like \"f47dbc9c:\" and \"dc22a3aa:\" are detected, they seems like ids of patch releases, their context is saved in `./experiments/short_keys_patch_releases.txt`\n",
     "    * You can check all detected keys by looking for 'KEY' tags in `./experiments/list_detected_pii.txt` \n",
     "* TODO: get precision numbers and try adding more filters (from detect-secrets fore example)\n",
     "2. For email detection:\n",
@@ -374,15 +374,15 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "nb dectected by detect-secrets: 2\n",
-      "nb dectected by regexes: 21\n",
+      "nb detected by detect-secrets: 2\n",
+      "nb detected by regexes: 21\n",
       "number true API keys: 30\n"
      ]
     }
    ],
    "source": [
-    "print(f\"nb dectected by detect-secrets: {detect_secrets_nb}\")\n",
-    "print(f\"nb dectected by regexes: {regexes_nb}\")\n",
+    "print(f\"nb detected by detect-secrets: {detect_secrets_nb}\")\n",
+    "print(f\"nb detected by regexes: {regexes_nb}\")\n",
     "print(f\"number true API keys: {len(api_keys_clean)}\")"
    ]
   },
@@ -417,7 +417,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "detect-secrets has a very low recall: 2 out of 30, let's anlyze the regex detections"
+    "detect-secrets has a very low recall: 2 out of 30, let's analyze the regex detections"
    ]
   },
   {
@@ -532,15 +532,15 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "nb dectected by detect-secrets: 0\n",
-      "nb dectected by regexes: 7\n",
+      "nb detected by detect-secrets: 0\n",
+      "nb detected by regexes: 7\n",
       "number true ssh keys: 7\n"
      ]
     }
    ],
    "source": [
-    "print(f\"nb dectected by detect-secrets: {detect_secrets_nb}\")\n",
-    "print(f\"nb dectected by regexes: {regexes_nb}\")\n",
+    "print(f\"nb detected by detect-secrets: {detect_secrets_nb}\")\n",
+    "print(f\"nb detected by regexes: {regexes_nb}\")\n",
     "print(f\"number true ssh keys: {len(ssh_keys_clean)}\")"
    ]
   },
@@ -782,7 +782,7 @@
    "source": [
     "### Email detection\n",
     "\n",
-    "* our current regex detects many false positives taht are derivatives of: dusk-network/helpers@4.6.12\n",
+    "* our current regex detects many false positives that are derivatives of: dusk-network/helpers@4.6.12\n",
     "* bigscience updated regex: can't detect emails well when they are in this format: <email> and also labels dusk-network/helpers@4.6.12 as emails, see https://regex101.com/r/LNwpG1/1"
    ]
   },
@@ -895,7 +895,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# fiw an issue with this annotation\n",
+    "# fiX an issue with this annotation\n",
     "ds_emails[142] = ds_emails[142][:-1]\n",
     "ds_emails[108] = ds_emails[108].strip()"
    ]
@@ -953,15 +953,15 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "nb emails dectected by old regex: 170\n",
-      "nb emails dectected by new BS regex: 169\n",
+      "nb emails detected by old regex: 170\n",
+      "nb emails detected by new BS regex: 169\n",
       "number true EMAILS: 170\n"
      ]
     }
    ],
    "source": [
-    "print(f\"nb emails dectected by old regex: {old_regex_nb}\")\n",
-    "print(f\"nb emails dectected by new BS regex: {new_regex_nb}\")\n",
+    "print(f\"nb emails detected by old regex: {old_regex_nb}\")\n",
+    "print(f\"nb emails detected by new BS regex: {new_regex_nb}\")\n",
     "print(f\"number true EMAILS: {len(ds_emails)}\")"
    ]
   },
@@ -1065,8 +1065,8 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "nb emails dectected by old regex: 7\n",
-      "nb emails dectected by new BS regex: 2\n",
+      "nb emails detected by old regex: 7\n",
+      "nb emails detected by new BS regex: 2\n",
       "number true EMAILS: 7\n"
      ]
     }
@@ -1088,8 +1088,8 @@
     "        new_regex_nb += 1\n",
     "        new_regex_results.append(output_2)\n",
     "\n",
-    "print(f\"nb emails dectected by old regex: {old_regex_nb}\")\n",
-    "print(f\"nb emails dectected by new BS regex: {new_regex_nb}\")\n",
+    "print(f\"nb emails detected by old regex: {old_regex_nb}\")\n",
+    "print(f\"nb emails detected by new BS regex: {new_regex_nb}\")\n",
     "print(f\"number true EMAILS: {len(ds_emails[:7])}\")"
    ]
   },
@@ -1147,7 +1147,7 @@
    "source": [
     "def compare_intervals(ref_intervals, pred_intervals):\n",
     "    \"\"\"Compare two lists of intervals and return the number of true positives, false positives and false negatives\n",
-    "    authur : @copilot\n",
+    "    author : @copilot\n",
     "    \"\"\"\n",
     "    ref_intervals = sorted(ref_intervals, key=lambda x: x[0])\n",
     "    pred_intervals = sorted(pred_intervals, key=lambda x: x[0])\n",

pii/notebooks/initial_analysis_regexes.ipynb

@@ -991,9 +991,9 @@
    "metadata": {},
    "source": [
     "In this section we test Presidio for the detection of the entities [\"EMAIL_ADDRESS\",\"PERSON\", \"CREDIT_CARD\", \"IP_ADDRESS\", \"IBAN_CODE\"]\n",
-    "* We first apply it on the code files then on their dosctrings/comments only\n",
+    "* We first apply it on the code files then on their docstrings/comments only\n",
     "\n",
-    "* Even with a high threshold on prediction score (0.8) the **predictiosn are bad**. The underhood NER must have trouble infering context from technical text\n",
+    "* Even with a high threshold on prediction score (0.8) the **predictions are bad**. The underhood NER must have trouble inferring context from technical text\n",
     "which often include python keywords and code"
    ]
   },

pii/notebooks/postprocess_annotations_lightag.ipynb

@@ -91,7 +91,7 @@
        " 'archived': False,\n",
        " 'priority': 1,\n",
        " 'active': True,\n",
-       " 'guidelines': '## Task Overview\\n\\nWelcome to our annotation task. In this task we\\'ll present you with one code file at a time and ask you to tag specific entities. We\\'ll be using this data to evaluate PII detection tools on source code from different programming languages.   \\n\\n1. Please highlight the entire span for each tags where applicable. For example: For tag `NAME`, if the text presented has John Doe, please highlight John Doe as one span, instead of highlighting John and Doe separately.\\n2. If you think a word that should be highlighted, but unsure about which tag to go to, use `AMBIGUOUS` instead.\\n3. **Do not overlap** tags. Tag the one most applicable to the entire span. For example, if a person\\'s name is part of `EMAIL`, do not tag `NAME`.\\n\\n## Tags Guidelines\\n\\nFor each file, please highlight you find any corresponding to the following tags: \\n\\n### API_KEY\\n\\nIncluding API Keys or API Tokens, Bearer Tokens, OAuth Tokens, see [here](https://www.freecodecamp.org/news/best-practices-for-building-api-keys-97c26eabfea9/#:~:text=the%20right%20way.-,API%20Key%20Generation,-Since%20the%20API) for example. Please highlight only the actual key.\\n\\n### AMBIGUOUS\\n\\nIf unsure whether to highlight a text, or which tag to highlight a text for, tag as `AMBIGUOUS` to allow reviewers to take a look later.\\n\\n### EMAIL\\n\\nEmail address, including generic emails such as support@organization.com.\\n\\n### IP_ADDRESS\\n\\nIP addresses could come in two different formats, see [IPv4 and IPv6 address formats](https://www.ibm.com/docs/en/ts3500-tape-library?topic=functionality-ipv4-ipv6-address-formats). Please tag both formats. If unsure a span is an IP address or not, tag `AMBIGUOUS` instead.\\n\\n### NAME\\n\\nA person\\'s name, including just first name or last name only. Do not include title used. For example, if \"Ms. Doe\" is used, please highlight \"Doe\" only.\\nFor now we are not tagging usernames.\\n\\n### PASSWORD\\n\\nAny authentication credentials not applicable to SSH keys or API keys. If unsure, tag `AMBIGUOUS`.\\n\\n### SSH_KEY\\nSecure Shell Key, the output usually looks like `ssh-rsa public_key account`, see example [here](https://git-scm.com/book/en/v2/Git-on-the-Server-Generating-Your-SSH-Public-Key). Please highlight the entire public key span. \\n\\n### USERNAME\\n\\nAny username that is used, including for credentials or handles (such as GitHub handles). If username is an email, tag `EMAIL` instead.\\n\\n## Additional note: Class CONTAINS_NON_ENGLISH\\n\\nIf you find a file with comments in a natural language you don\\'t undertsand and can\\'t decide if it includes PII, label it with the class `CONTAINS_NON_ENGLISH`.',\n",
+       " 'guidelines': '## Task Overview\\n\\nWelcome to our annotation task. In this task we\\'ll present you with one code file at a time and ask you to tag specific entities. We\\'ll be using this data to evaluate PII detection tools on source code from different programming languages.   \\n\\n1. Please highlight the entire span for each tags where applicable. For example: For tag `NAME`, if the text presented has John Doe, please highlight John Doe as one span, instead of highlighting John and Doe separately.\\n2. If you think a word that should be highlighted, but unsure about which tag to go to, use `AMBIGUOUS` instead.\\n3. **Do not overlap** tags. Tag the one most applicable to the entire span. For example, if a person\\'s name is part of `EMAIL`, do not tag `NAME`.\\n\\n## Tags Guidelines\\n\\nFor each file, please highlight you find any corresponding to the following tags: \\n\\n### API_KEY\\n\\nIncluding API Keys or API Tokens, Bearer Tokens, OAuth Tokens, see [here](https://www.freecodecamp.org/news/best-practices-for-building-api-keys-97c26eabfea9/#:~:text=the%20right%20way.-,API%20Key%20Generation,-Since%20the%20API) for example. Please highlight only the actual key.\\n\\n### AMBIGUOUS\\n\\nIf unsure whether to highlight a text, or which tag to highlight a text for, tag as `AMBIGUOUS` to allow reviewers to take a look later.\\n\\n### EMAIL\\n\\nEmail address, including generic emails such as support@organization.com.\\n\\n### IP_ADDRESS\\n\\nIP addresses could come in two different formats, see [IPv4 and IPv6 address formats](https://www.ibm.com/docs/en/ts3500-tape-library?topic=functionality-ipv4-ipv6-address-formats). Please tag both formats. If unsure a span is an IP address or not, tag `AMBIGUOUS` instead.\\n\\n### NAME\\n\\nA person\\'s name, including just first name or last name only. Do not include title used. For example, if \"Ms. Doe\" is used, please highlight \"Doe\" only.\\nFor now we are not tagging usernames.\\n\\n### PASSWORD\\n\\nAny authentication credentials not applicable to SSH keys or API keys. If unsure, tag `AMBIGUOUS`.\\n\\n### SSH_KEY\\nSecure Shell Key, the output usually looks like `ssh-rsa public_key account`, see example [here](https://git-scm.com/book/en/v2/Git-on-the-Server-Generating-Your-SSH-Public-Key). Please highlight the entire public key span. \\n\\n### USERNAME\\n\\nAny username that is used, including for credentials or handles (such as GitHub handles). If username is an email, tag `EMAIL` instead.\\n\\n## Additional note: Class CONTAINS_NON_ENGLISH\\n\\nIf you find a file with comments in a natural language you don\\'t understand and can\\'t decide if it includes PII, label it with the class `CONTAINS_NON_ENGLISH`.',\n",
        " 'schema_id': '33a5ef29-c22f-4e64-8fee-6019dde1c76d',\n",
        " 'dataset_id': '5ed27353-49a7-42e9-bc84-81e15f3f4162',\n",
        " 'project_id': '87c47b2c-d503-4967-b314-c04d1e7f8be7',\n",
@@ -788,7 +788,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# remove file at index 11 many ambigous keys\n",
+    "# remove file at index 11 many ambiguous keys\n",
     "# remove file 51 many incorrect names"
    ]
   },
@@ -3155,15 +3155,15 @@
       "        counters_sanity_check.append(1)\n",
       "\n",
       "    def test_rules_priority_forwarded(self, setup, direction, ptfadapter, counters_sanity_check, ip_version):\n",
-      "        \"\"\"Verify that we respect rule priorites in the forwarding case.\"\"\"\n",
+      "        \"\"\"Verify that we respect rule priorities in the forwarding case.\"\"\"\n",
       "        src_ip = \"20.0.0.7\" if ip_version == \"ipv4\" else \"60c0:a800::7\"\n",
       "        pkt = self.tcp_packet(setup, direction, ptfadapter, ip_version, src_ip=src_ip)\n",
       "\n",
       "        self._verify_acl_traffic(setup, direction, ptfadapter, pkt, False, ip_version)\n",
       "        counters_sanity_check.append(20)\n",
       "\n",
       "    def test_rules_priority_dropped(self, setup, direction, ptfadapter, counters_sanity_check, ip_version):\n",
-      "        \"\"\"Verify that we respect rule priorites in the drop case.\"\"\"\n",
+      "        \"\"\"Verify that we respect rule priorities in the drop case.\"\"\"\n",
       "        src_ip = \"20.0.0.3\" if ip_version == \"ipv4\" else \"60c0:a800::4\"\n",
       "        pkt = self.tcp_packet(setup, direction, ptfadapter, ip_version, src_ip=src_ip)\n",
       "\n",