Upload README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# IRIS: Iterative Refinement Image Synthesizer
|
| 2 |
+
|
| 3 |
+
A mobile-first image generation architecture designed from recent research (2025-2026).
|
| 4 |
+
|
| 5 |
+
**17/17 tests pass** — all modules verified for shape correctness, gradient flow, weight sharing, numerical stability, and actual training convergence.
|
| 6 |
+
|
| 7 |
+
See `iris/README.md` for full documentation.
|
| 8 |
+
|
| 9 |
+
## Quick Start
|
| 10 |
+
|
| 11 |
+
```python
|
| 12 |
+
from iris import IRIS, get_model_config, flow_matching_loss, euler_sample
|
| 13 |
+
import torch
|
| 14 |
+
|
| 15 |
+
model = IRIS(**get_model_config("iris-small")) # 40M params
|
| 16 |
+
z_0 = torch.randn(4, 32, 16, 16) * 2.5
|
| 17 |
+
text_emb = torch.randn(4, 16, 512)
|
| 18 |
+
losses = flow_matching_loss(model, z_0, text_emb, num_iterations=4)
|
| 19 |
+
losses["loss"].backward()
|
| 20 |
+
```
|
| 21 |
+
|
| 22 |
+
## Model Variants
|
| 23 |
+
|
| 24 |
+
| Config | Params | Tokens | FP16 Memory |
|
| 25 |
+
|--------|--------|--------|-------------|
|
| 26 |
+
| iris-tiny | 10.3M | 16 | 21 MB |
|
| 27 |
+
| iris-small | 40.0M | 16 | 80 MB |
|
| 28 |
+
| iris-base | 53.4M | 64 | 107 MB |
|
| 29 |
+
| iris-medium | 181.2M | 64 | 362 MB |
|
| 30 |
+
| iris-large | 430.9M | 64 | 862 MB |
|