Update handler.py
Browse files- handler.py +16 -9
handler.py
CHANGED
@@ -19,17 +19,17 @@ class EndpointHandler:
|
|
19 |
# Extract image and text from the input data
|
20 |
image_data = data.get("inputs", {}).get("image", "")
|
21 |
text_prompt = data.get("inputs", {}).get("text", "")
|
22 |
-
|
23 |
if not image_data or not text_prompt:
|
24 |
return {"error": "Both 'image' and 'text' must be provided in the input data."}
|
25 |
-
|
26 |
# Process the image data
|
27 |
try:
|
28 |
image_bytes = base64.b64decode(image_data)
|
29 |
image = Image.open(BytesIO(image_bytes)).convert("RGB")
|
30 |
except Exception as e:
|
31 |
return {"error": f"Failed to process image data: {e}"}
|
32 |
-
|
33 |
# Prepare the input in the format expected by the model
|
34 |
messages = [
|
35 |
{
|
@@ -40,7 +40,7 @@ class EndpointHandler:
|
|
40 |
],
|
41 |
}
|
42 |
]
|
43 |
-
|
44 |
# Process the input
|
45 |
text = self.processor.apply_chat_template(
|
46 |
messages, tokenize=False, add_generation_prompt=True
|
@@ -53,17 +53,24 @@ class EndpointHandler:
|
|
53 |
padding=True,
|
54 |
return_tensors="pt",
|
55 |
)
|
56 |
-
|
57 |
# Move inputs to the appropriate device
|
58 |
inputs = {k: v.to(self.device) for k, v in inputs.items()}
|
59 |
-
|
60 |
# Generate output
|
61 |
with torch.no_grad():
|
62 |
-
output_ids = self.model.generate(
|
63 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
# Decode the output
|
65 |
output_text = self.processor.batch_decode(
|
66 |
output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
67 |
)[0]
|
68 |
-
|
69 |
return {"generated_text": output_text}
|
|
|
19 |
# Extract image and text from the input data
|
20 |
image_data = data.get("inputs", {}).get("image", "")
|
21 |
text_prompt = data.get("inputs", {}).get("text", "")
|
22 |
+
|
23 |
if not image_data or not text_prompt:
|
24 |
return {"error": "Both 'image' and 'text' must be provided in the input data."}
|
25 |
+
|
26 |
# Process the image data
|
27 |
try:
|
28 |
image_bytes = base64.b64decode(image_data)
|
29 |
image = Image.open(BytesIO(image_bytes)).convert("RGB")
|
30 |
except Exception as e:
|
31 |
return {"error": f"Failed to process image data: {e}"}
|
32 |
+
|
33 |
# Prepare the input in the format expected by the model
|
34 |
messages = [
|
35 |
{
|
|
|
40 |
],
|
41 |
}
|
42 |
]
|
43 |
+
|
44 |
# Process the input
|
45 |
text = self.processor.apply_chat_template(
|
46 |
messages, tokenize=False, add_generation_prompt=True
|
|
|
53 |
padding=True,
|
54 |
return_tensors="pt",
|
55 |
)
|
56 |
+
|
57 |
# Move inputs to the appropriate device
|
58 |
inputs = {k: v.to(self.device) for k, v in inputs.items()}
|
59 |
+
|
60 |
# Generate output
|
61 |
with torch.no_grad():
|
62 |
+
output_ids = self.model.generate(
|
63 |
+
**inputs,
|
64 |
+
max_new_tokens=2000, # Increased from 128 to 2000
|
65 |
+
num_return_sequences=1,
|
66 |
+
do_sample=True,
|
67 |
+
temperature=0.7,
|
68 |
+
top_p=0.95
|
69 |
+
)
|
70 |
+
|
71 |
# Decode the output
|
72 |
output_text = self.processor.batch_decode(
|
73 |
output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
74 |
)[0]
|
75 |
+
|
76 |
return {"generated_text": output_text}
|