Publish validated code corpus

README.md

@@ -1,5 +1,5 @@
 ---
-pretty_name: Personal Code Corpus (raw-max)
+pretty_name: Personal Codex Model Training Corpus
 license: other
 language:
 - code
@@ -13,12 +13,28 @@ size_categories:
 - 10K<n<100K
 task_categories:
 - text-generation
+task_ids:
+- language-modeling
 tags:
 - code
+- text
 - source-code
 - code-completion
+- code-generation
+- causal-language-modeling
+- continued-pretraining
+- fine-tuning
+- coding-assistant
+- software-engineering
+- repository-level
+- multilingual-code
 - deduplicated
+- provenance-aware
+- parquet
 - datasets
+- typescript
+- python
+- javascript
 configs:
 - config_name: default
   data_files:
@@ -44,93 +60,218 @@ dataset_info:
     dtype: int32
   splits:
   - name: train
-    num_examples: 15157
-    num_bytes: 46331486
+    num_examples: 15226
+    num_bytes: 46393834
   - name: valid
-    num_examples: 3122
-    num_bytes: 9533576
+    num_examples: 3135
+    num_bytes: 9543076
 ---
 
-# Personal code corpus: raw-max
+# Personal Codex Model Training Corpus
 
-This is a repo-walked style/volume corpus, **not a curated accepted-state dataset**. It captures
-files present in local repository checkouts when the builder ran. It does not establish that every
-chunk is correct, reviewed, authored exclusively by one person, or suitable as a preferred answer.
+## Overview
 
-The corpus is intended for code-language modeling, code completion, and controlled personalization
-experiments. It must not be used to infer repository ownership, software quality, security, or the
-identity of an author.
+Personal Codex Model Training Corpus is a provenance-aware, repository-level dataset for causal
+language modeling, code completion, continued pretraining, and coding assistant adaptation. It is
+built from source files present in local Git repository checkouts at a defined collection point.
 
-## Dataset summary
+The dataset prioritizes broad, authentic software-engineering coverage while retaining enough
+metadata to audit every emitted chunk. It is not an instruction dataset, benchmark, or collection
+of verified solutions. Each record represents source text as it existed in a repository checkout,
+after filtering, chunking, secret screening, and global deduplication.
 
-- Rows: 18,279
-- Emitted lines: 1,301,863 (1,150,304 nonblank)
-- UTF-8 source text: 50.5 MiB
-- Source files with retained chunks: 7,836
-- Repositories with retained rows: 58
-- Completion field: `text`
-- Provenance fields: `repo`, `path`, `language`, `sha`, `chunk_index`, `n_tokens`
-- Split: deterministic repository-level train/valid assignment; a repository never crosses splits
-- Hash: SHA-256 of the emitted chunk text
-- Token counts: deterministic tokenizer-independent lexical estimates used for approximate chunking
+## Dataset profile
+
+| Metric | Value |
+| --- | ---: |
+| Dataset variant | `raw-max` |
+| Total examples | 18,361 |
+| Training examples | 15,226 |
+| Validation examples | 3,135 |
+| Emitted lines | 1,305,187 |
+| Nonblank emitted lines | 1,153,093 |
+| Approximate lexical tokens | 12,472,126 |
+| UTF-8 source text | 50.5 MiB |
+| Source files with retained chunks | 7,913 |
+| Repositories with retained rows | 58 |
+
+Line, byte, and token totals measure emitted training chunks. The configured chunk overlap can
+repeat text at chunk boundaries. These figures describe training volume, not unique repository
+lines of code or model-tokenizer counts.
+
+## Language and format distribution
+
+The `language` value is assigned from a controlled extension and exact-filename mapping. Markdown,
+configuration, schema, and build-system files are retained because they are part of real software
+engineering workflows and frequently contain executable examples or machine-consumed structure.
+
+| Language or format | Examples | Share |
+| --- | ---: | ---: |
+| Markdown | 6,559 | 35.7% |
+| TypeScript | 6,170 | 33.6% |
+| JSON | 2,828 | 15.4% |
+| Python | 1,532 | 8.3% |
+| XML Schema | 354 | 1.9% |
+| JavaScript | 218 | 1.2% |
+| CSS | 144 | 0.8% |
+| YAML | 112 | 0.6% |
+| MDX | 93 | 0.5% |
+| SQL | 69 | 0.4% |
+| Shell | 60 | 0.3% |
+| HTML | 53 | 0.3% |
+| Git Ignore | 44 | 0.2% |
+| Text | 31 | 0.2% |
+| Swift | 15 | 0.1% |
+| TeX | 14 | 0.1% |
+| Java | 13 | 0.1% |
+| TOML | 9 | 0.0% |
+| Dockerfile | 6 | 0.0% |
+| Git Attributes | 4 | 0.0% |
+| Web Manifest | 4 | 0.0% |
+| Prettier Ignore | 4 | 0.0% |
+| Prettier | 4 | 0.0% |
+| Docker Ignore | 3 | 0.0% |
+| XML | 3 | 0.0% |
+| EditorConfig | 2 | 0.0% |
+| INI | 2 | 0.0% |
+| Handlebars | 2 | 0.0% |
+| Batch | 2 | 0.0% |
+| Procfile | 1 | 0.0% |
+| JSON Lines | 1 | 0.0% |
+| Runpod Ignore | 1 | 0.0% |
+| Prisma | 1 | 0.0% |
+| ESLint Ignore | 1 | 0.0% |
+| Makefile | 1 | 0.0% |
+| SCSS | 1 | 0.0% |
+
+## Intended uses
+
+Appropriate uses include:
+
+- continued pretraining or domain adaptation of causal language models
+- code completion and repository-aware coding assistant experiments
+- tokenizer, chunking, deduplication, and corpus composition research
+- retrieval and provenance experiments using repository and path metadata
+- controlled studies of personalization on repository-disjoint validation data
+
+The dataset is not suitable as a correctness benchmark, a secure-code reference, a software
+license classifier, or evidence of authorship and repository ownership.
 
 ## Load the dataset
 
+Install a compatible version of `datasets`, then load the full corpus:
+
 ```python
 from datasets import load_dataset
 
 dataset = load_dataset("JulianAT/personal-codex-model")
+print(dataset)
+print(dataset["train"].features)
 ```
 
-Each row has the following stable schema:
+Stream examples without downloading the complete dataset:
 
-- `text`: source-code chunk used as the completion field
-- `repo`, `path`: source provenance within the local corpus
-- `language`: language inferred from the file extension
-- `sha`: SHA-256 of the emitted `text`
-- `chunk_index`: zero-based chunk position within the source file
-- `n_tokens`: tokenizer-independent lexical token estimate
+```python
+from datasets import load_dataset
 
-Line and byte totals describe emitted chunks. Chunk overlap can repeat boundary text, so these are
-training-corpus volume measurements rather than unique repository lines of code.
+stream = load_dataset("JulianAT/personal-codex-model", split="train", streaming=True)
+first_example = next(iter(stream))
+```
+
+## Schema
+
+| Field | Type | Description |
+| --- | --- | --- |
+| `text` | string | Source-code or repository-text chunk used as the modeling target. |
+| `repo` | string | Source repository name at collection time. |
+| `path` | string | Repository-relative source path. |
+| `language` | string | Language or format inferred from the configured mapping. |
+| `sha` | string | SHA-256 digest of the emitted `text`. |
+| `chunk_index` | int32 | Zero-based chunk position within the source file. |
+| `n_tokens` | int32 | Tokenizer-independent lexical token estimate. |
+
+## Dataset construction
+
+The builder applies the following deterministic pipeline:
+
+1. Discover configured Git repository checkouts.
+2. Walk supported source, documentation, schema, configuration, and build files.
+3. Exclude ignored, sensitive, generated, vendored, binary, oversized, and unsupported content.
+4. Decode retained files as UTF-8 and reject unreadable or empty payloads.
+5. Reject complete files containing high-confidence credential signatures.
+6. Chunk source text to approximately 896 lexical tokens with an overlap of
+   64 lexical tokens.
+7. Remove exact duplicate chunks by SHA-256.
+8. Remove near-duplicate chunks with MinHash LSH.
+9. Assign repositories, rather than individual rows, to deterministic train and validation splits.
 
-`raw-max` is bounded by code that actually exists after filtering. A claim of several million rows
-is only credible when the source repositories contain enough retained code; `--max-rows` is a
-ceiling, not a promised row count. This implementation supports at most 1,000,000 rows per variant.
+This repository-level split prevents a source repository from appearing in both splits. It reduces
+direct leakage from repeated project structure and repository-specific conventions.
 
-## Deduplication
+## Deduplication and quality controls
 
-Exact duplicate chunks are removed by SHA-256. Near duplicates are removed online with
-`datasketch.MinHashLSH`, 128 permutations, token
-5-grams, and a Jaccard threshold of 0.85.
+Near-duplicate detection uses `datasketch.MinHashLSH` with
+128 permutations, token 5-grams, and a Jaccard
+threshold of 0.85. The current build retained
+18,361 chunks after dropping
+1,470 exact duplicates and
+1,027 near duplicates.
 
-## Exclusions and limitations
+The file walk excludes Git metadata, ignored paths, dependency and environment directories, build
+outputs, vendored and generated directories, lockfiles, minified files, symlinks, binary or
+non-UTF-8 payloads, files above 1,048,576 bytes, and unsupported formats.
+Credential screening covers high-confidence private-key, platform-token, cloud-key, API-key, and
+JWT patterns.
 
-The walk excludes gitignored paths, `.git`, dependency/environment directories, build outputs,
-vendored/generated directories and filename patterns, lockfiles, minified files, symlinks, binaries,
-non-UTF-8 files, unsupported extensions, and files above 1,048,576 bytes.
-The quality variant may additionally exclude tests, fixtures, and low-history files.
+These controls reduce common leakage and duplication risks. They do not constitute a formal proof
+that every row is safe, original, correct, or free of sensitive information.
 
-Repository licenses and obligations still apply to the source code. The dataset-level `other`
-license value does not replace per-repository licenses. Paths and code can contain sensitive data;
-inspect the artifacts before publishing. MinHash is approximate, lexical token counts are not model
-token counts, current checkouts omit deleted historical code, and repository-level splitting can
-produce an empty validation split when fewer than two repositories contribute rows.
+## Provenance, privacy, and licensing
 
-This corpus can contain insecure, incomplete, duplicated, outdated, or otherwise low-quality code.
-It is not a benchmark and has no correctness labels. The secret scanner uses a conservative set of
-high-confidence patterns and is not proof that the corpus is free of private or identifying data.
+Every record retains repository, path, language, chunk position, and content-hash metadata. This
+supports traceability inside the published corpus without publishing local checkout locations or
+builder credentials.
 
-## Reproducibility and audit
+Some contributing repositories were private at collection time. Public publication was explicitly
+enabled by the dataset maintainer. Users should still treat repository names, paths, comments, and
+source text as potentially identifying information.
 
-`statistics.json` records row counts, split assignments, language distribution, filter decisions,
-deduplication totals, and build parameters. Splits are deterministic for the recorded seed. Source
-repository checkouts and the builder configuration are deliberately not included in the Hub repo.
+The dataset uses the Hugging Face `other` license classification because no single dataset-wide
+software license supersedes the licenses and obligations of the contributing repositories. Users
+are responsible for reviewing source-specific rights and restrictions before redistribution,
+commercial use, model release, or generated-code reuse.
 
-## Hub files
+## Limitations
 
-- `data/*.parquet`: Hugging Face loader source
-- `dataset_infos.json`, `statistics.json`: schema, counts, filters, and dedup audit
+- Source is collected from working-tree snapshots, not from deleted Git history.
+- Repository contents can include incomplete, insecure, outdated, experimental, or generated-like
+  code that survives the configured filters.
+- Extension-based language labels do not perform parser-level language verification.
+- Lexical token estimates are not equivalent to tokens from a production model tokenizer.
+- MinHash is approximate and can retain related text or remove independently written similar text.
+- Chunk overlap increases emitted volume and can repeat boundary lines.
+- The dataset contains no correctness, security, quality, preference, or authorship labels.
+- Repository-disjoint validation measures transfer across included repositories, not general coding
+  ability across unrelated ecosystems.
 
-The local build also creates Arrow and JSONL representations for training. They are intentionally
-excluded from Hub publication because they duplicate the Parquet data.
+## Reproducibility and audit artifacts
+
+`statistics.json` records build parameters, split assignments, row and token counts, language
+distribution, filter decisions, and deduplication totals. `dataset_infos.json` records the feature
+schema and split sizes. The Parquet shards are the canonical Hub loader source.
+
+The Hub publication intentionally omits local Arrow and JSONL copies because they duplicate the
+Parquet payload. It also omits source checkouts, local filesystem paths, author-email configuration,
+and training artifacts.
+
+## Citation
+
+```bibtex
+@misc{personal_codex_model_training_corpus,
+  author       = {JulianAT},
+  title        = {Personal Codex Model Training Corpus},
+  year         = {2026},
+  howpublished = {Hugging Face Datasets},
+  url          = {https://huggingface.co/datasets/JulianAT/personal-codex-model}
+}
+```

data/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08f42133ddb13188997a3b512d556a4ca0f3cd26dbb4bc9644b788421fce46bc
-size 15767661
+oid sha256:fcdb4a9d58f6217ab1a142586c5a756befd69b3427650b332deb851d3953bd56
+size 15813974

data/valid-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3eb5fffa54a9a5f1a6136c79f4ab261168b7acc4fa7f2ca07bc68720e005cbd8
-size 3721175
+oid sha256:e6b2f45d80f690bfbf7563346779ff0e6c7b113d1648032868bf30b082bda879
+size 3730829

dataset_infos.json

@@ -3,9 +3,9 @@
     "builder_name": "parquet",
     "config_name": "default",
     "dataset_name": "raw-max",
-    "dataset_size": 55865062,
+    "dataset_size": 55936910,
     "description": "Repository-walked personal code chunks for completion training.",
-    "download_size": 19488836,
+    "download_size": 19544803,
     "features": {
       "chunk_index": {
         "_type": "Value",
@@ -37,17 +37,17 @@
       }
     },
     "license": "other",
-    "size_in_bytes": 75353898,
+    "size_in_bytes": 75481713,
     "splits": {
       "train": {
         "name": "train",
-        "num_bytes": 46331486,
-        "num_examples": 15157
+        "num_bytes": 46393834,
+        "num_examples": 15226
       },
       "valid": {
         "name": "valid",
-        "num_bytes": 9533576,
-        "num_examples": 3122
+        "num_bytes": 9543076,
+        "num_examples": 3135
       }
     }
   }

statistics.json

@@ -1,17 +1,30 @@
 {
   "by_language": {
+    "Batch": 2,
     "CSS": 144,
+    "Docker Ignore": 3,
     "Dockerfile": 6,
+    "ESLint Ignore": 1,
+    "EditorConfig": 2,
+    "Git Attributes": 4,
+    "Git Ignore": 44,
     "HTML": 53,
+    "Handlebars": 2,
+    "INI": 2,
     "JSON": 2828,
+    "JSON Lines": 1,
     "Java": 13,
     "JavaScript": 218,
     "MDX": 93,
     "Makefile": 1,
-    "Markdown": 6554,
+    "Markdown": 6559,
+    "Prettier": 4,
+    "Prettier Ignore": 4,
     "Prisma": 1,
     "Procfile": 1,
-    "Python": 1530,
+    "Python": 1532,
+    "Runpod Ignore": 1,
+    "SCSS": 1,
     "SQL": 69,
     "Shell": 60,
     "Swift": 15,
@@ -19,6 +32,7 @@
     "TeX": 14,
     "Text": 31,
     "TypeScript": 6170,
+    "Web Manifest": 4,
     "XML": 3,
     "XML Schema": 354,
     "YAML": 112
@@ -27,76 +41,76 @@
     "1AHIF-BOT": 3,
     "Illegale_Strassenrennen-Webprojekt": 8,
     "Kriegsmarine_Webprojekt": 19,
-    "POS-23-24-Java": 15,
+    "POS-23-24-Java": 17,
     "Project-CB": 1,
     "Valorant-Webprojekt": 30,
-    "adapter-lab": 1301,
-    "ai-video": 196,
+    "adapter-lab": 1304,
+    "ai-video": 197,
     "asai": 50,
-    "aucta": 398,
-    "brz-chatbot": 366,
-    "catchi": 239,
-    "clawguard": 732,
-    "codex-hackathon": 1081,
+    "aucta": 399,
+    "brz-chatbot": 368,
+    "catchi": 240,
+    "clawguard": 734,
+    "codex-hackathon": 1082,
     "cursor-austria": 41,
-    "cursor-vienna-discord": 147,
-    "datagvat-mcp": 2107,
-    "daytona-generative-ui": 383,
-    "diplomarbeit-ideen": 387,
-    "diplomarbeit-os": 2737,
+    "cursor-vienna-discord": 149,
+    "datagvat-mcp": 2111,
+    "daytona-generative-ui": 389,
+    "diplomarbeit-ideen": 389,
+    "diplomarbeit-os": 2740,
     "ewor-skill": 276,
-    "hacknation": 624,
-    "hacknation-app": 68,
-    "hacknation-landing": 211,
+    "hacknation": 626,
+    "hacknation-app": 69,
+    "hacknation-landing": 212,
     "hacknation-venture": 17,
     "hacknation-web": 37,
-    "hands-on-ai": 269,
-    "htl-donaustadt-chatbot": 42,
+    "hands-on-ai": 272,
+    "htl-donaustadt-chatbot": 44,
     "html-starter": 4,
-    "ifg-agentic": 267,
-    "interior-3d": 35,
-    "interiorly": 457,
+    "ifg-agentic": 268,
+    "interior-3d": 38,
+    "interiorly": 463,
     "interiorly-dashboard": 70,
-    "interiorly-pitch": 101,
+    "interiorly-pitch": 103,
     "interiorly-preview": 41,
-    "interiorly-promo-fake": 263,
-    "interiorly_changelog": 36,
+    "interiorly-promo-fake": 266,
+    "interiorly_changelog": 37,
     "interiorly_landingpage": 56,
-    "jku-assignments": 511,
-    "jku-current-topics-report": 29,
-    "jku-encyclopedia": 1216,
+    "jku-assignments": 512,
+    "jku-current-topics-report": 31,
+    "jku-encyclopedia": 1220,
     "jku-python": 1,
     "julian-at": 2,
     "mlpc-2026-task3": 92,
     "mlpc-exam": 403,
-    "mlpc_task_4": 57,
+    "mlpc_task_4": 58,
     "mystahd": 66,
     "next-todo": 14,
-    "openai-build-week": 762,
+    "openai-build-week": 764,
     "portfolio": 142,
-    "qdrant-hackathon": 191,
-    "start-hackathon": 78,
-    "synth-ui": 525,
-    "synth-ui-training": 20,
+    "qdrant-hackathon": 193,
+    "start-hackathon": 81,
+    "synth-ui": 528,
+    "synth-ui-training": 22,
     "tdot_chatbot": 146,
     "untis-lehrer-extension": 29,
-    "young_ai_leaders_linz": 70,
-    "zero_one_hack_01": 810
+    "young_ai_leaders_linz": 74,
+    "zero_one_hack_01": 813
   },
   "cap_reached": false,
-  "chunks_considered": 20700,
+  "chunks_considered": 20858,
   "dedup_report": {
-    "exact_duplicates_dropped": 1409,
-    "near_duplicates_dropped": 1012,
-    "rows_kept": 18279
+    "exact_duplicates_dropped": 1470,
+    "near_duplicates_dropped": 1027,
+    "rows_kept": 18361
   },
-  "files_considered": 8825,
-  "files_selected": 8779,
+  "files_considered": 8970,
+  "files_selected": 8924,
   "filter_report": {
     "empty": 45,
     "excluded_directory": 65,
-    "extension": 1283,
-    "gitignored": 25,
+    "extension": 1131,
+    "gitignored": 32,
     "lockfile": 55,
     "secret_openai_key": 1,
     "sensitive_path": 31,
@@ -104,10 +118,10 @@
     "symlink_directory": 2,
     "too_large": 121
   },
-  "generated_at": "2026-07-16T15:57:59+00:00",
+  "generated_at": "2026-07-16T16:10:46+00:00",
   "lines": {
-    "nonblank": 1150304,
-    "total": 1301863
+    "nonblank": 1153093,
+    "total": 1305187
   },
   "max_rows": 1000000,
   "parameters": {
@@ -185,31 +199,31 @@
   "repositories_discovered": 60,
   "repositories_with_rows": 58,
   "rows": {
-    "total": 18279,
-    "train": 15157,
-    "valid": 3122
+    "total": 18361,
+    "train": 15226,
+    "valid": 3135
   },
   "self_test": {
     "loader": "datasets.load_dataset(\"JulianAT/personal-codex-model\")",
     "splits": {
-      "train": 15157,
-      "valid": 3122
+      "train": 15226,
+      "valid": 3135
     },
     "status": "passed"
   },
-  "source_files": 7836,
+  "source_files": 7913,
   "sources": {
     "config": "local config.yaml (not published)",
     "repository_root": "local repository checkouts (not published)"
   },
   "tokens": {
     "histogram": {
-      "0-255": 2780,
-      "256-511": 2511,
-      "512-1023": 12988
+      "0-255": 2847,
+      "256-511": 2515,
+      "512-1023": 12999
     },
-    "total": 12456402
+    "total": 12472126
   },
-  "utf8_bytes": 52915839,
+  "utf8_bytes": 52976638,
   "variant": "raw-max"
 }