Upload infer_qwen2_vl.py with huggingface_hub
Browse files- infer_qwen2_vl.py +127 -0
infer_qwen2_vl.py
ADDED
@@ -0,0 +1,127 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
|
2 |
+
from qwen_vl_utils import process_vision_info
|
3 |
+
import json
|
4 |
+
|
5 |
+
def read_json(file_path):
|
6 |
+
with open(file_path, 'r', encoding='utf-8') as file:
|
7 |
+
data = json.load(file)
|
8 |
+
return data
|
9 |
+
|
10 |
+
def write_json(file_path, data):
|
11 |
+
with open(file_path, 'w', encoding='utf-8') as file:
|
12 |
+
json.dump(data, file, ensure_ascii=False, indent=4)
|
13 |
+
|
14 |
+
# default: Load the model on the available device(s)
|
15 |
+
model_path = '/inspire/hdd/ws-ba572160-47f8-4ca1-984e-d6bcdeb95dbb/a100-maybe/albus/ICCV_2025/qvq/models/QVQ-72B-Preview'
|
16 |
+
model = Qwen2VLForConditionalGeneration.from_pretrained(
|
17 |
+
model_path, torch_dtype="auto", device_map="auto"
|
18 |
+
)
|
19 |
+
|
20 |
+
# default processer
|
21 |
+
processor = AutoProcessor.from_pretrained(model_path)
|
22 |
+
|
23 |
+
# The default range for the number of visual tokens per image in the model is 4-16384. You can set min_pixels and max_pixels according to your needs, such as a token count range of 256-1280, to balance speed and memory usage.
|
24 |
+
# min_pixels = 256*28*28
|
25 |
+
# max_pixels = 1280*28*28
|
26 |
+
#processor = AutoProcessor.from_pretrained("Qwen/QVQ-72B-Preview", min_pixels=min_pixels, max_pixels=max_pixels)
|
27 |
+
|
28 |
+
import glob
|
29 |
+
from PIL import Image
|
30 |
+
import argparse
|
31 |
+
import os
|
32 |
+
|
33 |
+
# parser = argparse.ArgumentParser(description="Process a dataset with specific index range.")
|
34 |
+
# parser.add_argument("--batch_size", type=int, default = 1,help="batch size")
|
35 |
+
# #parser.add_argument("--index", type=int, default = 0,help="index")
|
36 |
+
# args = parser.parse_args()
|
37 |
+
|
38 |
+
|
39 |
+
folder = "/inspire/hdd/ws-ba572160-47f8-4ca1-984e-d6bcdeb95dbb/a100-maybe/albus/ICCV_2025/qvq/dataset"
|
40 |
+
|
41 |
+
file_names = os.listdir(folder)
|
42 |
+
|
43 |
+
num_image = len(file_names)
|
44 |
+
|
45 |
+
begin, end, batch_size= 0, num_image, 6
|
46 |
+
print(f"beigin : {begin}, end : {end}, batch_size : {batch_size}")
|
47 |
+
messages = [
|
48 |
+
{
|
49 |
+
"role": "system",
|
50 |
+
"content": [
|
51 |
+
{"type": "text", "text": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."}
|
52 |
+
],
|
53 |
+
},
|
54 |
+
{
|
55 |
+
"role": "user",
|
56 |
+
"content": [
|
57 |
+
{
|
58 |
+
"type": "image",
|
59 |
+
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/QVQ/demo.png",
|
60 |
+
},
|
61 |
+
{"type": "text", "text": "Please describe in detail the content of the picture."},
|
62 |
+
],
|
63 |
+
}
|
64 |
+
]
|
65 |
+
|
66 |
+
from tqdm import tqdm
|
67 |
+
# Preparation for inference
|
68 |
+
ans = []
|
69 |
+
counter = 0
|
70 |
+
for batch_idx in tqdm(range(begin, end, batch_size)):
|
71 |
+
up = min(batch_idx + batch_size, end)
|
72 |
+
batch = file_names[batch_idx: up]
|
73 |
+
print(f"data index range : {batch_idx} ~ {up}")
|
74 |
+
image_inputs_batch, video_inputs_batch,text_batch = [], [], []
|
75 |
+
for idx,i in enumerate(batch):
|
76 |
+
#img = batch[i]
|
77 |
+
#print('gain image successfully !')
|
78 |
+
img_path = '/inspire/hdd/ws-ba572160-47f8-4ca1-984e-d6bcdeb95dbb/a100-maybe/albus/ICCV_2025/qvq/dataset/' + i
|
79 |
+
#print(img_path)
|
80 |
+
messages[1]["content"][0]["image"] = '/inspire/hdd/ws-ba572160-47f8-4ca1-984e-d6bcdeb95dbb/a100-maybe/albus/ICCV_2025/qvq/dataset/' + i
|
81 |
+
text = processor.apply_chat_template(
|
82 |
+
messages, tokenize=False, add_generation_prompt=True
|
83 |
+
)
|
84 |
+
text_batch.append(text)
|
85 |
+
image_inputs, video_inputs = process_vision_info(messages)
|
86 |
+
print(video_inputs)
|
87 |
+
image_inputs_batch.append(image_inputs)
|
88 |
+
video_inputs_batch.append(video_inputs)
|
89 |
+
inputs = processor(
|
90 |
+
text=text_batch, # [text]
|
91 |
+
images=image_inputs_batch,
|
92 |
+
videos=None,
|
93 |
+
padding=True,
|
94 |
+
return_tensors="pt",
|
95 |
+
)
|
96 |
+
inputs = inputs.to("cuda")
|
97 |
+
|
98 |
+
# Inference: Generation of the output
|
99 |
+
|
100 |
+
#print(inputs)
|
101 |
+
|
102 |
+
# for x in range(len(inputs)):
|
103 |
+
# print(f"Generating {x}th image")
|
104 |
+
# generated_ids = model.generate(**x, max_new_tokens=8192)
|
105 |
+
# generated_ids_trimmed = [
|
106 |
+
# out_ids[len(in_ids) :] for in_ids, out_ids in zip(x.input_ids, generated_ids)
|
107 |
+
# ]
|
108 |
+
# output_text = processor.batch_decode(
|
109 |
+
# generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=True
|
110 |
+
# )
|
111 |
+
# ans.append(output_text)
|
112 |
+
|
113 |
+
generated_ids = model.generate(**inputs, max_new_tokens=8192)
|
114 |
+
generated_ids_trimmed = [
|
115 |
+
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
|
116 |
+
]
|
117 |
+
output_text = processor.batch_decode(
|
118 |
+
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
119 |
+
)
|
120 |
+
ans.append(output_text)
|
121 |
+
save_path = "output_final.json"
|
122 |
+
counter = counter + 1
|
123 |
+
if counter % 10 == 0 or up + 10 >= end:
|
124 |
+
print(f"Saving data at iteration {idx + 1}")
|
125 |
+
write_json(save_path, ans)
|
126 |
+
|
127 |
+
|