Update README: document humanities+wikitext corpus, 4000 BPE tokenizer

README.md

@@ -7,59 +7,71 @@ language:
 tags:
 - philosophy
 - classical-texts
-- character-level
+- humanities
+- wikitext
+- bpe-tokenizer
 - gpt-training
 size_categories:
-- 10K<n<100K
-dataset_info:
-  features:
-  - name: text
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 20542969
-    num_examples: 284485
-  - name: validation
-    num_bytes: 2273279
-    num_examples: 31610
-  download_size: 14763402
-  dataset_size: 22816248
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
+- 1M<n<10M
 ---
 
-# Philosophy Corpus
+# Philosophy & Humanities Corpus
 
-54 classical philosophy texts for training character-level language models.
+Combined humanities and Wikipedia corpus for training small language models.
 
-## Contents
+## Dataset
 
-| Category | Works | Count |
-|---|---|---|
-| **Plato** | Republic, Apology, Symposium, Phaedo, Crito, Meno, Phaedrus, Timaeus, Laws, Gorgias, Protagoras, Theaetetus | 12 |
-| **Aristotle** | Categories, Ethics, Rhetoric, Physics, Metaphysics, Poetics, Politics, On the Soul, On the Heavens, Prior/Posterior Analytics, Topics, On Generation & Corruption | 12 |
-| **Stoics** | Marcus Aurelius Meditations, Epictetus Discourses & Enchiridion, Seneca Moral Essays | 4 |
-| **Roman** | Lucretius On the Nature of Things, Cicero (On Duties, Nature of Gods, On Friendship) | 4 |
-| **Early Modern** | Descartes (Method, Meditations), Kant Critique of Pure Reason, Spinoza Ethics, Hobbes Leviathan, Locke Second Treatise, Bacon Essays | 6 |
-| **Enlightenment/19th c.** | Hume, Rousseau, Nietzsche (x2), Mill (x2), Machiavelli, Emerson, Thoreau, Montaigne, Schopenhauer | 10 |
-| **Other** | Boethius Consolation, Diogenes/Epicurus, Latin Grammar | 6 |
+| Split | Lines | Size | Description |
+|-------|-------|------|-------------|
+| **train.txt** | 3.0M | 549 MB | Humanities (368K lines) + WikiText-103 (2.6M lines) |
+| **val.txt** | 315K | 57 MB | Matching validation split |
 
-## Files
+## Sources
 
-- **data/*.txt** - Individual cleaned text files (54 files)
-- **data/training_data.txt** - Combined paragraphs, lowercased, a-z + space + period only (85K paragraphs, ~25 MB)
-- **data/full_corpus.txt** - Full concatenated corpus with original casing (~26 MB)
+### Humanities (368K lines, 66 MB)
+54 classical philosophy and humanities texts:
 
-## Sources
+| Category | Works |
+|---|---|
+| **Plato** | Republic, Apology, Symposium, Phaedo, Crito, Meno, Phaedrus, Timaeus, Laws, Gorgias, Protagoras, Theaetetus |
+| **Aristotle** | Categories, Ethics, Rhetoric, Physics, Metaphysics, Poetics, Politics, On the Soul, On the Heavens, Prior/Posterior Analytics, Topics, On Generation & Corruption |
+| **Stoics** | Marcus Aurelius Meditations, Epictetus Discourses & Enchiridion, Seneca Moral Essays |
+| **Roman** | Lucretius, Cicero (On Duties, Nature of Gods, On Friendship) |
+| **Early Modern** | Descartes, Kant, Spinoza, Hobbes, Locke, Bacon |
+| **Enlightenment/19th c.** | Hume, Rousseau, Nietzsche, Mill, Machiavelli, Emerson, Thoreau, Montaigne, Schopenhauer |
+| **Other** | Boethius, Diogenes/Epicurus, Aeschylus, Latin Grammar, Euclid Elements |
+
+Sources: [Project Gutenberg](https://www.gutenberg.org/), [MIT Internet Classics Archive](http://classics.mit.edu/)
+
+### WikiText-103 (2.6M lines, 481 MB)
+Wikipedia articles from [Salesforce/wikitext](https://huggingface.co/datasets/Salesforce/wikitext) (wikitext-103-v1). Cleaned, chunked, and deduplicated.
 
-- [Project Gutenberg](https://www.gutenberg.org/) (public domain)
-- [MIT Internet Classics Archive](http://classics.mit.edu/) (public domain translations)
+## Tokenizer
+
+**tokenizer.json** — BPE tokenizer (4000 vocab) trained on the combined corpus.
+- Format: HuggingFace tokenizers JSON (GPT-2 ByteLevel BPE)
+- Special tokens: `<|pad|>` (id=0), `<|eos|>` (id=1)
+
+## Files
+
+| File | Description |
+|------|-------------|
+| `train.txt` | Combined training data (one chunk per line) |
+| `val.txt` | Combined validation data |
+| `tokenizer.json` | BPE tokenizer (vocab_size=4000) |
+| `data/*.txt` | Individual source text files |
+| `train_enriched.jsonl` | Enriched training data with metadata |
+| `train_trivium.txt` | Trivium-phase subset |
+| `train_quadrivium.txt` | Quadrivium-phase subset |
+| `train_philosophy.txt` | Philosophy-phase subset |
 
 ## Usage
 
-Used as training data for [JuliaGPT](https://github.com/DavinciDreams/JuliaGPT) and [JuliaFluxGPT](https://github.com/DavinciDreams/JuliaGPT) character-level transformer models.
+Training data for [JuliaGPT](https://github.com/DavinciDreams/JuliaGPT) — small transformer language models in Julia (Flux.jl).
+
+```julia
+# Auto-downloaded by juliadistill.ipynb
+hf_download("LisaMegaWatts/philosophy-corpus", "train.txt"; repo_type="dataset")
+hf_download("LisaMegaWatts/philosophy-corpus", "val.txt"; repo_type="dataset")
+hf_download("LisaMegaWatts/philosophy-corpus", "tokenizer.json"; repo_type="dataset")
+```