Update README.md
Browse files
README.md
CHANGED
@@ -65,6 +65,68 @@ We evaluated the model using the [VLMEvalKit](https://github.com/open-compass/VL
|
|
65 |
|
66 |
For comparison models, evaluations were conducted in a local environment, so the scores may differ slightly from those reported in papers or on the official VLMEvalKit leaderboard.
|
67 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
# Future Plan
|
69 |
|
70 |
* We plan to train models of various sizes.
|
|
|
65 |
|
66 |
For comparison models, evaluations were conducted in a local environment, so the scores may differ slightly from those reported in papers or on the official VLMEvalKit leaderboard.
|
67 |
|
68 |
+
# How to use
|
69 |
+
|
70 |
+
```python
|
71 |
+
# pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
|
72 |
+
from llava.model.builder import load_pretrained_model
|
73 |
+
from llava.mm_utils import process_images, tokenizer_image_token
|
74 |
+
from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
|
75 |
+
from llava.conversation import conv_templates
|
76 |
+
from PIL import Image
|
77 |
+
import requests
|
78 |
+
import copy
|
79 |
+
import torch
|
80 |
+
import warnings
|
81 |
+
|
82 |
+
warnings.filterwarnings("ignore")
|
83 |
+
|
84 |
+
pretrained = "BAAI/Aquila-VL-2B-llava-qwen"
|
85 |
+
|
86 |
+
model_name = "llava_qwen"
|
87 |
+
device = "cuda"
|
88 |
+
device_map = "auto"
|
89 |
+
tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map) # Add any other thing you want to pass in llava_model_args
|
90 |
+
|
91 |
+
model.eval()
|
92 |
+
|
93 |
+
# load image from url
|
94 |
+
url = "https://github.com/haotian-liu/LLaVA/blob/1a91fc274d7c35a9b50b3cb29c4247ae5837ce39/images/llava_v1_5_radar.jpg?raw=true"
|
95 |
+
image = Image.open(requests.get(url, stream=True).raw)
|
96 |
+
|
97 |
+
# load image from local environment
|
98 |
+
# url = "./local_image.jpg"
|
99 |
+
# image = Image.open(url)
|
100 |
+
|
101 |
+
image_tensor = process_images([image], image_processor, model.config)
|
102 |
+
image_tensor = [_image.to(dtype=torch.float16, device=device) for _image in image_tensor]
|
103 |
+
|
104 |
+
conv_template = "qwen_1_5" # Make sure you use correct chat template for different models
|
105 |
+
question = DEFAULT_IMAGE_TOKEN + "\nWhat is shown in this image?"
|
106 |
+
conv = copy.deepcopy(conv_templates[conv_template])
|
107 |
+
conv.append_message(conv.roles[0], question)
|
108 |
+
conv.append_message(conv.roles[1], None)
|
109 |
+
prompt_question = conv.get_prompt()
|
110 |
+
|
111 |
+
input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device)
|
112 |
+
image_sizes = [image.size]
|
113 |
+
|
114 |
+
cont = model.generate(
|
115 |
+
input_ids,
|
116 |
+
images=image_tensor,
|
117 |
+
image_sizes=image_sizes,
|
118 |
+
do_sample=False,
|
119 |
+
temperature=0,
|
120 |
+
max_new_tokens=4096,
|
121 |
+
)
|
122 |
+
|
123 |
+
text_outputs = tokenizer.batch_decode(cont, skip_special_tokens=True)
|
124 |
+
|
125 |
+
print(text_outputs)
|
126 |
+
```
|
127 |
+
|
128 |
+
|
129 |
+
|
130 |
# Future Plan
|
131 |
|
132 |
* We plan to train models of various sizes.
|