Add README

README.md

@@ -0,0 +1,90 @@
+# Parametric Floorplan Generator
+
+A model that generates 2D construction/home floor plans from parametric input (room count, total area, room types, adjacency constraints).
+
+## Approach
+
+This model is based on the **DStruct2Design** paper ([arXiv:2407.15723](https://arxiv.org/abs/2407.15723)):
+> *DStruct2Design: Data and Benchmarks for Data Structure Driven Generative Floor Plan Design*
+
+The approach fine-tunes **Qwen2.5-1.5B-Instruct** with LoRA on the [ludolara/DStruct2Design](https://huggingface.co/datasets/ludolara/DStruct2Design) dataset, learning to output structured JSON floorplans from natural-language parametric constraints.
+
+## Dataset
+
+- **Source**: `ludolara/DStruct2Design`
+- **Train**: 10,000 examples
+- **Validation**: 1,000 examples
+- **Test**: 1,000 examples
+- **Format**: Each example contains:
+  - `room_count`: number of rooms
+  - `total_area`: total floor area in m²
+  - `room_types`: list of room type strings
+  - `rooms`: list of rooms with polygon vertices, area, width, height
+  - `edges`: adjacency graph (room index pairs)
+
+## Usage
+
+### Training
+
+```bash
+pip install transformers trl torch datasets peft accelerate trackio
+python train.py
+```
+
+### Inference
+
+```bash
+python generate.py --room_count 4 --total_area 100 --room_types Bedroom Bathroom Kitchen LivingRoom
+```
+
+## Example Output
+
+For input:
+```
+Generate a floor plan with 4 rooms and a total area of 100 square meters.
+The room types are: Bedroom, Bathroom, Kitchen, LivingRoom.
+```
+
+The model outputs JSON like:
+```json
+{
+  "rooms": [
+    {
+      "room_type": "Bedroom",
+      "area": 25.2,
+      "width": 6.1,
+      "height": 4.1,
+      "floor_polygon": [
+        {"x": 6.1, "z": 4.1},
+        {"x": 6.1, "z": 10.2},
+        {"x": 10.2, "z": 10.2},
+        {"x": 10.2, "z": 4.1}
+      ],
+      "is_regular": 1
+    },
+    ...
+  ],
+  "edges": [[0,1], [0,2], [0,3], [1,3], [2,3]],
+  "room_count": 4,
+  "total_area": 100.0,
+  "room_types": ["Bedroom", "Bathroom", "Kitchen", "LivingRoom"]
+}
+```
+
+## Architecture
+
+| Component | Value |
+|-----------|-------|
+| Base Model | Qwen2.5-1.5B-Instruct |
+| Fine-tuning | LoRA (r=16, alpha=32) |
+| Target Modules | q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj |
+| Epochs | 5 |
+| Batch Size | 4 (accumulation=4) |
+| Learning Rate | 1e-4 |
+| Max Sequence Length | 2048 |
+| Precision | bf16 |
+
+## Links
+
+- Paper: [DStruct2Design (arXiv:2407.15723)](https://arxiv.org/abs/2407.15723)
+- Dataset: [ludolara/DStruct2Design](https://huggingface.co/datasets/ludolara/DStruct2Design)