Add model card metadata: base_model, license, pipeline_tag, tags
Browse files
README.md
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: Qwen/Qwen-Image
|
| 3 |
+
library_name: diffusers
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
pipeline_tag: text-to-image
|
| 6 |
+
tags:
|
| 7 |
+
- qwen-image
|
| 8 |
+
- text-to-image
|
| 9 |
+
- bf16
|
| 10 |
+
- diffusers
|
| 11 |
+
---
|
| 12 |
+
# Qwen-Image-bf16
|
| 13 |
+
|
| 14 |
+
An unmodified bf16 copy of [Qwen/Qwen-Image](https://huggingface.co/Qwen/Qwen-Image),
|
| 15 |
+
resharded for WaveSpeed's loader.
|
| 16 |
+
|
| 17 |
+
## What was changed
|
| 18 |
+
|
| 19 |
+
Nothing numerically. Tensor names, shapes, dtypes (bf16) and total parameter
|
| 20 |
+
bytes are identical to upstream; the transformer is simply written as 5 shards
|
| 21 |
+
instead of 9, which suits how our serving nodes stream weights from cache.
|
| 22 |
+
|
| 23 |
+
**If you are not us, use [`Qwen/Qwen-Image`](https://huggingface.co/Qwen/Qwen-Image)
|
| 24 |
+
instead.** This repository exists as a pinned serving artifact and offers no
|
| 25 |
+
advantage over the original.
|
| 26 |
+
|
| 27 |
+
The quantized variants derived from this baseline are
|
| 28 |
+
[`wavespeed/Qwen-Image-e4m3`](https://huggingface.co/wavespeed/Qwen-Image-e4m3)
|
| 29 |
+
(fp8) — those do differ from upstream.
|
| 30 |
+
|
| 31 |
+
## License
|
| 32 |
+
|
| 33 |
+
Apache-2.0, inherited from Qwen-Image.
|