Upload README.md from Drive bundle
Browse files
README.md
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Stable Diffusion v1.5 converted to LiteRT
|
| 2 |
+
|
| 3 |
+
This repository contains a LiteRT/TFLite export of the Hugging Face model `stable-diffusion-v1-5/stable-diffusion-v1-5`.
|
| 4 |
+
|
| 5 |
+
## Variants
|
| 6 |
+
|
| 7 |
+
- `fp32/`: reference export
|
| 8 |
+
- `int8/`: mixed bundle with fp32 text encoder fallback, PT2E dynamic int8 UNet, and fp32 VAE fallback
|
| 9 |
+
|
| 10 |
+
## Files per variant
|
| 11 |
+
|
| 12 |
+
- `text_encoder.tflite`
|
| 13 |
+
- `unet.tflite`
|
| 14 |
+
- `vae_decoder.tflite`
|
| 15 |
+
|
| 16 |
+
## Shared assets
|
| 17 |
+
|
| 18 |
+
- `tokenizer/`
|
| 19 |
+
- `scheduler/`
|
| 20 |
+
- `configs/`
|
| 21 |
+
- `conversion_manifest.json`
|
| 22 |
+
|
| 23 |
+
## Notes
|
| 24 |
+
|
| 25 |
+
- Stable Diffusion v1.5 is a multi-stage pipeline, so this export is split into submodels.
|
| 26 |
+
- Both the `fp32` and `int8` variants are smoke-tested by reloading the serialized LiteRT models and executing inference.
|
| 27 |
+
- The preview images in `preview/` are decoder smoke tests, not final text-to-image samples.
|