Update all files for BitDance-14B-64x-diffusers
Browse files
README.md
CHANGED
|
@@ -23,16 +23,16 @@ Diffusers-converted checkpoint for BitDance-14B-64x with bundled custom pipeline
|
|
| 23 |
import torch
|
| 24 |
from diffusers import DiffusionPipeline
|
| 25 |
|
| 26 |
-
|
| 27 |
-
|
| 28 |
pipe = DiffusionPipeline.from_pretrained(
|
| 29 |
-
|
| 30 |
-
|
| 31 |
torch_dtype=torch.bfloat16,
|
| 32 |
).to("cuda")
|
| 33 |
|
| 34 |
result = pipe(
|
| 35 |
-
prompt="A cinematic
|
| 36 |
height=1024,
|
| 37 |
width=1024,
|
| 38 |
num_inference_steps=50,
|
|
@@ -41,6 +41,32 @@ result = pipe(
|
|
| 41 |
result.images[0].save("bitdance_14b_64x.png")
|
| 42 |
```
|
| 43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
## Model Metadata
|
| 45 |
|
| 46 |
- Pipeline class: `BitDanceDiffusionPipeline`
|
|
|
|
| 23 |
import torch
|
| 24 |
from diffusers import DiffusionPipeline
|
| 25 |
|
| 26 |
+
# Local path (recommended - no trust_remote_code needed)
|
| 27 |
+
model_path = "path/to/BitDance-14B-64x-diffusers"
|
| 28 |
pipe = DiffusionPipeline.from_pretrained(
|
| 29 |
+
model_path,
|
| 30 |
+
custom_pipeline=model_path,
|
| 31 |
torch_dtype=torch.bfloat16,
|
| 32 |
).to("cuda")
|
| 33 |
|
| 34 |
result = pipe(
|
| 35 |
+
prompt = "A close-up portrait in a cinematic photography style, capturing a girl-next-door look on a sunny daytime urban street. She wears a khaki sweater, with long, flowing hair gently draped over her shoulders. Her head is turned slightly, revealing soft facial features illuminated by realistic, delicate sunlight coming from the left. The sunlight subtly highlights individual strands of her hair. The image has a Canon film-like color tone, evoking a warm nostalgic atmosphere.",
|
| 36 |
height=1024,
|
| 37 |
width=1024,
|
| 38 |
num_inference_steps=50,
|
|
|
|
| 41 |
result.images[0].save("bitdance_14b_64x.png")
|
| 42 |
```
|
| 43 |
|
| 44 |
+
## Test Running
|
| 45 |
+
|
| 46 |
+
Run tests from the model directory in your active Python environment:
|
| 47 |
+
|
| 48 |
+
```bash
|
| 49 |
+
python test_bitdance.py
|
| 50 |
+
```
|
| 51 |
+
|
| 52 |
+
## VRAM Usage by Resolution
|
| 53 |
+
|
| 54 |
+
Measured on `NVIDIA A100-SXM4-80GB` using:
|
| 55 |
+
|
| 56 |
+
- `dtype=torch.bfloat16`
|
| 57 |
+
- `num_inference_steps=30`
|
| 58 |
+
- `guidance_scale=7.5`
|
| 59 |
+
- prompt: `A close-up portrait in a cinematic photography style, capturing a girl-next-door look on a sunny daytime urban street. She wears a khaki sweater, with long, flowing hair gently draped over her shoulders. Her head is turned slightly, revealing soft facial features illuminated by realistic, delicate sunlight coming from the left. The sunlight subtly highlights individual strands of her hair. The image has a Canon film-like color tone, evoking a warm nostalgic atmosphere.`
|
| 60 |
+
|
| 61 |
+
| Resolution | Peak Allocated VRAM (GiB) | Peak Reserved VRAM (GiB) | Time (s) | Status |
|
| 62 |
+
| --- | ---: | ---: | ---: | --- |
|
| 63 |
+
| 512x512 | 39.60 | 40.62 | 4.08 | ok |
|
| 64 |
+
| 1024x1024 | 41.21 | 50.15 | 15.79 | ok |
|
| 65 |
+
| 1280x768 | 40.88 | 49.52 | 14.78 | ok |
|
| 66 |
+
| 768x1280 | 40.88 | 49.52 | 14.75 | ok |
|
| 67 |
+
| 1536x640 | 40.88 | 49.52 | 14.76 | ok |
|
| 68 |
+
| 2048x512 | 41.21 | 50.15 | 15.85 | ok |
|
| 69 |
+
|
| 70 |
## Model Metadata
|
| 71 |
|
| 72 |
- Pipeline class: `BitDanceDiffusionPipeline`
|