Spaces:
Running
Running
Commit
·
d7e7825
1
Parent(s):
a2b6d64
trying new model
Browse files
main.py
CHANGED
@@ -5,17 +5,17 @@ import torch
|
|
5 |
from fastapi import FastAPI, Query
|
6 |
from PIL import Image
|
7 |
from qwen_vl_utils import process_vision_info
|
8 |
-
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
|
9 |
|
10 |
app = FastAPI()
|
11 |
|
12 |
-
checkpoint = "Qwen/Qwen2
|
13 |
min_pixels = 256 * 28 * 28
|
14 |
max_pixels = 1280 * 28 * 28
|
15 |
processor = AutoProcessor.from_pretrained(
|
16 |
checkpoint, min_pixels=min_pixels, max_pixels=max_pixels
|
17 |
)
|
18 |
-
model =
|
19 |
checkpoint,
|
20 |
torch_dtype=torch.bfloat16,
|
21 |
device_map="auto",
|
@@ -58,25 +58,34 @@ def encode_image(image_path, max_size=(800, 800), quality=85):
|
|
58 |
print(f"❌ Error encoding image {image_path}: {e}")
|
59 |
return None
|
60 |
|
61 |
-
|
62 |
@app.get("/predict")
|
63 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
image = encode_image(image_url)
|
66 |
|
|
|
|
|
67 |
messages = [
|
68 |
-
{
|
69 |
-
"role": "system",
|
70 |
-
"content": "You are a helpful assistant with vision abilities.",
|
71 |
-
},
|
72 |
{
|
73 |
"role": "user",
|
74 |
"content": [
|
75 |
{"type": "image", "image": f"data:image;base64,{image}"},
|
76 |
{"type": "text", "text": prompt},
|
77 |
],
|
78 |
-
}
|
79 |
]
|
|
|
|
|
80 |
text = processor.apply_chat_template(
|
81 |
messages, tokenize=False, add_generation_prompt=True
|
82 |
)
|
@@ -87,16 +96,60 @@ def predict(image_url: str = Query(...), prompt: str = Query(...)):
|
|
87 |
videos=video_inputs,
|
88 |
padding=True,
|
89 |
return_tensors="pt",
|
90 |
-
).to(
|
91 |
-
|
92 |
-
|
|
|
93 |
generated_ids_trimmed = [
|
94 |
out_ids[len(in_ids) :]
|
95 |
for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
96 |
]
|
97 |
-
|
98 |
generated_ids_trimmed,
|
99 |
skip_special_tokens=True,
|
100 |
clean_up_tokenization_spaces=False,
|
101 |
)
|
102 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
from fastapi import FastAPI, Query
|
6 |
from PIL import Image
|
7 |
from qwen_vl_utils import process_vision_info
|
8 |
+
from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration, Qwen2VLForConditionalGeneration
|
9 |
|
10 |
app = FastAPI()
|
11 |
|
12 |
+
checkpoint = "Qwen/Qwen2-VL-3B-Instruct"
|
13 |
min_pixels = 256 * 28 * 28
|
14 |
max_pixels = 1280 * 28 * 28
|
15 |
processor = AutoProcessor.from_pretrained(
|
16 |
checkpoint, min_pixels=min_pixels, max_pixels=max_pixels
|
17 |
)
|
18 |
+
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
19 |
checkpoint,
|
20 |
torch_dtype=torch.bfloat16,
|
21 |
device_map="auto",
|
|
|
58 |
print(f"❌ Error encoding image {image_path}: {e}")
|
59 |
return None
|
60 |
|
|
|
61 |
@app.get("/predict")
|
62 |
+
def describe_image_with_qwen2_vl(image_url: str = Query(...), prompt: str = Query(...)):
|
63 |
+
"""
|
64 |
+
Generates a description for an image using the Qwen-2-VL model.
|
65 |
+
|
66 |
+
Args:
|
67 |
+
image_url (str): The URL of the image to describe.
|
68 |
+
prompt (str): The text prompt to guide the model's response.
|
69 |
+
|
70 |
+
Returns:
|
71 |
+
str: The generated description of the image.
|
72 |
+
"""
|
73 |
|
74 |
image = encode_image(image_url)
|
75 |
|
76 |
+
|
77 |
+
# Create the input message structure
|
78 |
messages = [
|
|
|
|
|
|
|
|
|
79 |
{
|
80 |
"role": "user",
|
81 |
"content": [
|
82 |
{"type": "image", "image": f"data:image;base64,{image}"},
|
83 |
{"type": "text", "text": prompt},
|
84 |
],
|
85 |
+
}
|
86 |
]
|
87 |
+
|
88 |
+
# Prepare inputs for the model
|
89 |
text = processor.apply_chat_template(
|
90 |
messages, tokenize=False, add_generation_prompt=True
|
91 |
)
|
|
|
96 |
videos=video_inputs,
|
97 |
padding=True,
|
98 |
return_tensors="pt",
|
99 |
+
).to("cuda:0")
|
100 |
+
|
101 |
+
# Generate the output
|
102 |
+
generated_ids = model.generate(**inputs, max_new_tokens=2056)
|
103 |
generated_ids_trimmed = [
|
104 |
out_ids[len(in_ids) :]
|
105 |
for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
106 |
]
|
107 |
+
output_text = processor.batch_decode(
|
108 |
generated_ids_trimmed,
|
109 |
skip_special_tokens=True,
|
110 |
clean_up_tokenization_spaces=False,
|
111 |
)
|
112 |
+
|
113 |
+
return {"response": output_text[0] if output_text else "No description generated."}
|
114 |
+
|
115 |
+
# @app.get("/predict")
|
116 |
+
# def predict(image_url: str = Query(...), prompt: str = Query(...)):
|
117 |
+
|
118 |
+
# image = encode_image(image_url)
|
119 |
+
|
120 |
+
# messages = [
|
121 |
+
# {
|
122 |
+
# "role": "system",
|
123 |
+
# "content": "You are a helpful assistant with vision abilities.",
|
124 |
+
# },
|
125 |
+
# {
|
126 |
+
# "role": "user",
|
127 |
+
# "content": [
|
128 |
+
# {"type": "image", "image": f"data:image;base64,{image}"},
|
129 |
+
# {"type": "text", "text": prompt},
|
130 |
+
# ],
|
131 |
+
# },
|
132 |
+
# ]
|
133 |
+
# text = processor.apply_chat_template(
|
134 |
+
# messages, tokenize=False, add_generation_prompt=True
|
135 |
+
# )
|
136 |
+
# image_inputs, video_inputs = process_vision_info(messages)
|
137 |
+
# inputs = processor(
|
138 |
+
# text=[text],
|
139 |
+
# images=image_inputs,
|
140 |
+
# videos=video_inputs,
|
141 |
+
# padding=True,
|
142 |
+
# return_tensors="pt",
|
143 |
+
# ).to(model.device)
|
144 |
+
# with torch.no_grad():
|
145 |
+
# generated_ids = model.generate(**inputs, max_new_tokens=128)
|
146 |
+
# generated_ids_trimmed = [
|
147 |
+
# out_ids[len(in_ids) :]
|
148 |
+
# for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
149 |
+
# ]
|
150 |
+
# output_texts = processor.batch_decode(
|
151 |
+
# generated_ids_trimmed,
|
152 |
+
# skip_special_tokens=True,
|
153 |
+
# clean_up_tokenization_spaces=False,
|
154 |
+
# )
|
155 |
+
# return {"response": output_texts[0]}
|