| | --- |
| | library_name: transformers |
| | license: apache-2.0 |
| | datasets: |
| | - grammarly/coedit |
| | base_model: |
| | - Qwen/Qwen2.5-3B-Instruct |
| | --- |
| | |
| | # LucentEdita |
| |
|
| | LucentEdita is a family of fine‑tuned Qwen2.5‑Instruct models specialized for high‑quality text editing, grammar correction, clarity improvement, and style refinement. |
| |
|
| | Built on the Grammarly/CoEdit dataset, LucentEdita focuses on polishing writing while preserving the author’s original voice. |
| |
|
| | --- |
| |
|
| | ## Model Family |
| |
|
| | LucentEdita is available in multiple parameter sizes: |
| |
|
| | | Model | Base | Best For | |
| | | -------------------- | --------------------- | ------------------------------------------------- | |
| | | **LucentEdita‑1.5B** | Qwen2.5‑1.5B‑Instruct | Lightweight editing, fast inference, local use | |
| | | **LucentEdita‑3B** | Qwen2.5‑3B‑Instruct | Balanced performance and quality | |
| | | **LucentEdita‑7B** | Qwen2.5‑7B‑Instruct | Highest quality, nuanced rewriting, long passages | |
| |
|
| | All variants share the same editing philosophy and persona. |
| |
|
| | --- |
| |
|
| | ## Purpose |
| |
|
| | LucentEdita is designed to: |
| |
|
| | * Correct grammar, spelling, and punctuation |
| | * Improve clarity and readability |
| | * Refine sentence structure |
| | * Preserve the author’s tone and intent |
| | * Provide concise, professional rewrites |
| |
|
| | LucentEdita does **not** aim to: |
| |
|
| | * Add new factual information |
| | * Perform deep research |
| | * Replace large reasoning or general‑purpose LLMs |
| |
|
| | It is a specialized editing model. |
| |
|
| | --- |
| |
|
| | ## Training |
| |
|
| | Base Models: |
| |
|
| | * Qwen2.5‑Instruct series |
| |
|
| | Fine‑Tuning Dataset: |
| |
|
| | * `grammarly/coedit` |
| |
|
| | Training Objective: Supervised fine‑tuning on instruction‑style edit pairs to improve structured text revision performance. |
| |
|
| | --- |
| |
|
| | ## Example Usage |
| |
|
| | ### Prompt Template |
| |
|
| | ``` |
| | You are LucentEdita, a professional writing assistant. |
| | Improve the following text while preserving its original meaning and tone. |
| | Only return the revised version. |
| | Text: |
| | "I can has cheezburger." |
| | ``` |
| |
|
| | ### Output |
| |
|
| | ``` |
| | I can have a cheeseburger. |
| | ``` |
| |
|
| | --- |
| |
|
| | ## Limitations |
| |
|
| | * Smaller variants (1.5B) may struggle with long multi‑paragraph edits. |
| | * Does not perform fact verification. |
| | * Not designed for complex reasoning tasks. |
| |
|
| | --- |
| |
|
| | ## License |
| |
|
| | Please follow the license terms of the underlying Qwen2.5 base model. |
| |
|
| | --- |
| |
|
| | ## About Lucent |
| |
|
| | LucentEdita is part of the Lucent model family — specialized AI systems focused on clarity, refinement, and purposeful intelligence. |
