Paul DAMPFHOEFFER
commited on
Commit
·
cbe05b5
1
Parent(s):
cd782ec
fix: small fix
Browse files
app.py
CHANGED
@@ -13,17 +13,23 @@ def greet_json():
|
|
13 |
|
14 |
@app.post("/")
|
15 |
async def aria_image_to_text(request: Request):
|
|
|
16 |
data = await request.json()
|
|
|
17 |
image_url = data.get("image_url")
|
|
|
|
|
|
|
18 |
image = Image.open(requests.get(image_url, stream=True).raw)
|
19 |
-
|
20 |
model_id_or_path = "rhymes-ai/Aria"
|
|
|
21 |
model = AriaForConditionalGeneration.from_pretrained(
|
22 |
model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16
|
23 |
)
|
24 |
-
|
25 |
processor = AriaProcessor.from_pretrained(model_id_or_path)
|
26 |
-
|
27 |
messages = [
|
28 |
{
|
29 |
"role": "user",
|
@@ -34,11 +40,15 @@ async def aria_image_to_text(request: Request):
|
|
34 |
}
|
35 |
]
|
36 |
|
|
|
37 |
text = processor.apply_chat_template(messages, add_generation_prompt=True)
|
|
|
38 |
inputs = processor(text=text, images=image, return_tensors="pt")
|
|
|
39 |
inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
|
|
|
40 |
inputs.to(model.device)
|
41 |
-
|
42 |
output = model.generate(
|
43 |
**inputs,
|
44 |
max_new_tokens=15,
|
@@ -47,21 +57,26 @@ async def aria_image_to_text(request: Request):
|
|
47 |
do_sample=True,
|
48 |
temperature=0.9,
|
49 |
)
|
|
|
50 |
output_ids = output[0][inputs["input_ids"].shape[1]:]
|
|
|
51 |
response = processor.decode(output_ids, skip_special_tokens=True)
|
|
|
52 |
return {"response": response}
|
53 |
|
54 |
@app.get("/aria-test")
|
55 |
def aria_test():
|
|
|
56 |
model_id_or_path = "rhymes-ai/Aria"
|
|
|
57 |
model = AriaForConditionalGeneration.from_pretrained(
|
58 |
model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16
|
59 |
)
|
60 |
-
|
61 |
processor = AriaProcessor.from_pretrained(model_id_or_path)
|
62 |
-
|
63 |
image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
|
64 |
-
|
65 |
messages = [
|
66 |
{
|
67 |
"role": "user",
|
@@ -71,12 +86,15 @@ def aria_test():
|
|
71 |
],
|
72 |
}
|
73 |
]
|
74 |
-
|
75 |
text = processor.apply_chat_template(messages, add_generation_prompt=True)
|
|
|
76 |
inputs = processor(text=text, images=image, return_tensors="pt")
|
|
|
77 |
inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
|
|
|
78 |
inputs.to(model.device)
|
79 |
-
|
80 |
output = model.generate(
|
81 |
**inputs,
|
82 |
max_new_tokens=15,
|
@@ -86,5 +104,7 @@ def aria_test():
|
|
86 |
temperature=0.9,
|
87 |
)
|
88 |
output_ids = output[0][inputs["input_ids"].shape[1]:]
|
|
|
89 |
response = processor.decode(output_ids, skip_special_tokens=True)
|
90 |
-
|
|
|
|
13 |
|
14 |
@app.post("/")
|
15 |
async def aria_image_to_text(request: Request):
|
16 |
+
print(1)
|
17 |
data = await request.json()
|
18 |
+
print(2)
|
19 |
image_url = data.get("image_url")
|
20 |
+
print(3)
|
21 |
+
print('image_url')
|
22 |
+
print(image_url)
|
23 |
image = Image.open(requests.get(image_url, stream=True).raw)
|
24 |
+
print(4)
|
25 |
model_id_or_path = "rhymes-ai/Aria"
|
26 |
+
print(5)
|
27 |
model = AriaForConditionalGeneration.from_pretrained(
|
28 |
model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16
|
29 |
)
|
30 |
+
print(6)
|
31 |
processor = AriaProcessor.from_pretrained(model_id_or_path)
|
32 |
+
print(7)
|
33 |
messages = [
|
34 |
{
|
35 |
"role": "user",
|
|
|
40 |
}
|
41 |
]
|
42 |
|
43 |
+
print(8)
|
44 |
text = processor.apply_chat_template(messages, add_generation_prompt=True)
|
45 |
+
print(9)
|
46 |
inputs = processor(text=text, images=image, return_tensors="pt")
|
47 |
+
print(10)
|
48 |
inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
|
49 |
+
print(11)
|
50 |
inputs.to(model.device)
|
51 |
+
print(12)
|
52 |
output = model.generate(
|
53 |
**inputs,
|
54 |
max_new_tokens=15,
|
|
|
57 |
do_sample=True,
|
58 |
temperature=0.9,
|
59 |
)
|
60 |
+
print(13)
|
61 |
output_ids = output[0][inputs["input_ids"].shape[1]:]
|
62 |
+
print(14)
|
63 |
response = processor.decode(output_ids, skip_special_tokens=True)
|
64 |
+
print(15)
|
65 |
return {"response": response}
|
66 |
|
67 |
@app.get("/aria-test")
|
68 |
def aria_test():
|
69 |
+
print(1)
|
70 |
model_id_or_path = "rhymes-ai/Aria"
|
71 |
+
print(2)
|
72 |
model = AriaForConditionalGeneration.from_pretrained(
|
73 |
model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16
|
74 |
)
|
75 |
+
print(3)
|
76 |
processor = AriaProcessor.from_pretrained(model_id_or_path)
|
77 |
+
print(4)
|
78 |
image = Image.open(requests.get("http://images.cocodataset.org/val2017/000000039769.jpg", stream=True).raw)
|
79 |
+
print(5)
|
80 |
messages = [
|
81 |
{
|
82 |
"role": "user",
|
|
|
86 |
],
|
87 |
}
|
88 |
]
|
89 |
+
print(6)
|
90 |
text = processor.apply_chat_template(messages, add_generation_prompt=True)
|
91 |
+
print(7)
|
92 |
inputs = processor(text=text, images=image, return_tensors="pt")
|
93 |
+
print(8)
|
94 |
inputs['pixel_values'] = inputs['pixel_values'].to(torch.bfloat16)
|
95 |
+
print(9)
|
96 |
inputs.to(model.device)
|
97 |
+
print(10)
|
98 |
output = model.generate(
|
99 |
**inputs,
|
100 |
max_new_tokens=15,
|
|
|
104 |
temperature=0.9,
|
105 |
)
|
106 |
output_ids = output[0][inputs["input_ids"].shape[1]:]
|
107 |
+
print(11)
|
108 |
response = processor.decode(output_ids, skip_special_tokens=True)
|
109 |
+
print(12)
|
110 |
+
return {"response": response}
|