File size: 3,371 Bytes
9e3db82 cf9bf4a 9e3db82 cf9bf4a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 | ---
license: other
license_name: open-data-attribution-training-disclosure-license-odatl-1.0
license_link: LICENSE
language:
- en
tags:
- tiny
- english
---
# LLM-English-100MB — Compact & Dense English Teaching Corpus
A 100MB, extremely clean CSV designed to teach an LLM English from scratch via instruction-tuning. No noise, no HTML, no duplicates — just pure grammar, vocabulary, and syntax transformations.
Generated with a single paste-and-run Python script in Google Colab.
### Why this teaches English
Instead of raw text, the dataset is **instruction -> input -> output** pairs that force the model to learn rules:
- **Grammar Mechanics:** tense conversion, negation, question formation, active/passive, subject-verb agreement, article & preposition usage, contraction
- **Vocabulary:** pluralization, synonyms, antonyms, definitions, comparatives
- **Error Correction:** common ESL mistakes with corrected form + rule
Each row includes an explicit `rule` column so the model learns the *pattern*, not just memorizes.
### Dataset Schema
Extremely clean: UTF-8, LF (`\n`), `QUOTE_MINIMAL`, no empty fields, no newlines inside fields.
| Column | Type | Description |
| :--- | :--- | :--- |
| `id` | int | Unique row ID |
| `level` | string | CEFR level: A1, A2, B1, B2 |
| `category` | string | `grammar`, `vocabulary`, `syntax` |
| `task_type` | string | e.g. `tense_past`, `pluralization`, `negation`, `question_formation`, `comparative`, `active_passive`, `article_usage`, `synonym`, `antonym` |
| `instruction` | string | What the model must do |
| `input_text` | string | Input sentence / prompt |
| `output_text` | string | Correct target |
| `rule` | string | Short linguistic rule |
Example:
```
instruction: Convert to simple past tense.
input_text: We build the house.
output_text: We built the house.
rule: build -> built
```
### Stats
- **Size:** 100 MB (configurable)
- **Rows:** ~650k - 750k (avg ~150 bytes/row)
- **Format:** CSV
- **Encoding:** UTF-8
- **Generation Time:** ~90-120 seconds on Colab
### How to Use
**Pandas:**
```python
import pandas as pd
df = pd.read_csv("/content/llm_english_100MB.csv")
df.head()
```
**Hugging Face Datasets:**
```python
from datasets import load_dataset
ds = load_dataset("csv", data_files="/content/llm_english_100MB.csv")["train"]
# Format for instruction tuning: instruction + input -> output
def format_prompt(ex):
return {"text": f"### Instruction: {ex['instruction']}\n### Input: {ex['input_text']}\n### Output: {ex['output_text']}"}
ds = ds.map(format_prompt)
```
**Training Prompt Template:**
```
Below is an instruction that teaches English.
### Instruction:
Convert to simple past tense.
### Input:
She eats the apple.
### Output:
She ate the apple.
```
### Quality Guarantees — EXTREMELY CLEAN
- No HTML, no URLs, no emojis
- No nulls / NaNs
- Stripped whitespace, no `\r`, no `\n` inside fields
- Deterministic seed (42) for reproducibility
- Validated with `csv.DictReader`
- All rows are synthetic and license-free
### Customize
Edit vocab lists at top of script:
```python
NOUNS = [...]
VERBS = [...]
```
Add new generators to `GENERATORS` list to add new task types.
### License
Open Data Attribution Training Disclosure License (ODATL‑1.0)
### File Structure
```
/content/
llm_english_100MB.csv # 100MB corpus
README.md # this file
``` |