Update handler.py
Browse files- handler.py +15 -3
handler.py
CHANGED
@@ -1,16 +1,23 @@
|
|
1 |
import torch
|
2 |
-
from diffusers import StableDiffusionXLPipeline, DDIMScheduler
|
3 |
import base64
|
4 |
from io import BytesIO
|
5 |
import os
|
6 |
|
7 |
class InferenceHandler:
|
8 |
def __init__(self):
|
|
|
9 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
10 |
-
model_name = "./" # Use the current directory
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
self.pipe = StableDiffusionXLPipeline.from_pretrained(
|
13 |
-
|
14 |
torch_dtype=torch.float16,
|
15 |
use_safetensors=True,
|
16 |
use_auth_token=os.getenv("HUGGINGFACE_TOKEN")
|
@@ -20,12 +27,14 @@ class InferenceHandler:
|
|
20 |
self.pipe.scheduler = DDIMScheduler.from_config(self.pipe.scheduler.config)
|
21 |
|
22 |
def __call__(self, inputs):
|
|
|
23 |
prompt = inputs.get("prompt", "")
|
24 |
if not prompt:
|
25 |
raise ValueError("A prompt must be provided")
|
26 |
|
27 |
negative_prompt = inputs.get("negative_prompt", "")
|
28 |
|
|
|
29 |
image = self.pipe(
|
30 |
prompt=prompt,
|
31 |
negative_prompt=negative_prompt,
|
@@ -33,10 +42,13 @@ class InferenceHandler:
|
|
33 |
guidance_scale=7.5
|
34 |
).images[0]
|
35 |
|
|
|
36 |
buffered = BytesIO()
|
37 |
image.save(buffered, format="PNG")
|
38 |
image_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
39 |
|
|
|
40 |
return {"image_base64": image_base64}
|
41 |
|
|
|
42 |
handler = InferenceHandler()
|
|
|
1 |
import torch
|
2 |
+
from diffusers import StableDiffusionXLPipeline, DDIMScheduler
|
3 |
import base64
|
4 |
from io import BytesIO
|
5 |
import os
|
6 |
|
7 |
class InferenceHandler:
|
8 |
def __init__(self):
|
9 |
+
# Determine the device to run on
|
10 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
11 |
|
12 |
+
# Get the directory where this script is located
|
13 |
+
model_dir = os.path.dirname(os.path.abspath(__file__))
|
14 |
+
|
15 |
+
# Print the model directory for debugging purposes
|
16 |
+
print("Loading model from directory:", model_dir)
|
17 |
+
|
18 |
+
# Load the pipeline with authentication
|
19 |
self.pipe = StableDiffusionXLPipeline.from_pretrained(
|
20 |
+
model_dir,
|
21 |
torch_dtype=torch.float16,
|
22 |
use_safetensors=True,
|
23 |
use_auth_token=os.getenv("HUGGINGFACE_TOKEN")
|
|
|
27 |
self.pipe.scheduler = DDIMScheduler.from_config(self.pipe.scheduler.config)
|
28 |
|
29 |
def __call__(self, inputs):
|
30 |
+
# Extract the prompt from inputs
|
31 |
prompt = inputs.get("prompt", "")
|
32 |
if not prompt:
|
33 |
raise ValueError("A prompt must be provided")
|
34 |
|
35 |
negative_prompt = inputs.get("negative_prompt", "")
|
36 |
|
37 |
+
# Generate the image using the pipeline
|
38 |
image = self.pipe(
|
39 |
prompt=prompt,
|
40 |
negative_prompt=negative_prompt,
|
|
|
42 |
guidance_scale=7.5
|
43 |
).images[0]
|
44 |
|
45 |
+
# Convert the image to base64 encoding
|
46 |
buffered = BytesIO()
|
47 |
image.save(buffered, format="PNG")
|
48 |
image_base64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
|
49 |
|
50 |
+
# Return the base64 image
|
51 |
return {"image_base64": image_base64}
|
52 |
|
53 |
+
# Instantiate the handler
|
54 |
handler = InferenceHandler()
|