File size: 7,437 Bytes
9d392e7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
---
language: en
license: mit
tags:
- vision
- image-to-text 
- image-captioning
- visual-question-answering
pipeline_tag: image-to-text
---

# BLIP-2, OPT-2.7b, pre-trained only

BLIP-2 model, leveraging [OPT-2.7b](https://huggingface.co/facebook/opt-2.7b) (a large language model with 2.7 billion parameters).
It was introduced in the paper [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://arxiv.org/abs/2301.12597) by Li et al. and first released in [this repository](https://github.com/salesforce/LAVIS/tree/main/projects/blip2).

Disclaimer: The team releasing BLIP-2 did not write a model card for this model so this model card has been written by the Hugging Face team.

## Model description

BLIP-2 consists of 3 models: a CLIP-like image encoder, a Querying Transformer (Q-Former) and a large language model.

The authors initialize the weights of the image encoder and large language model from pre-trained checkpoints and keep them frozen
while training the Querying Transformer, which is a BERT-like Transformer encoder that maps a set of "query tokens" to query embeddings,
which bridge the gap between the embedding space of the image encoder and the large language model.

The goal for the model is simply to predict the next text token, giving the query embeddings and the previous text.

<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/blip2_architecture.jpg"
alt="drawing" width="600"/> 

This allows the model to be used for tasks like:

- image captioning
- visual question answering (VQA)
- chat-like conversations by feeding the image and the previous conversation as prompt to the model

## Direct Use and Downstream Use

You can use the raw model for conditional text generation given an image and optional text. See the [model hub](https://huggingface.co/models?search=Salesforce/blip) to look for
fine-tuned versions on a task that interests you.

## Bias, Risks, Limitations, and Ethical Considerations

BLIP2-OPT uses off-the-shelf OPT as the language model. It inherits the same risks and limitations as mentioned in Meta's model card.

> Like other large language models for which the diversity (or lack thereof) of training
> data induces downstream impact on the quality of our model, OPT-175B has limitations in terms
> of bias and safety. OPT-175B can also have quality issues in terms of generation diversity and
> hallucination. In general, OPT-175B is not immune from the plethora of issues that plague modern
> large language models.
> 
BLIP2 is fine-tuned on image-text datasets (e.g. [LAION](https://laion.ai/blog/laion-400-open-dataset/) ) collected from the internet.  As a result the model itself is potentially vulnerable to generating equivalently inappropriate content or replicating inherent biases in the underlying data.

BLIP2 has not been tested in real world applications. It should not be directly deployed in any applications. Researchers should first carefully assess the safety and fairness of the model in relation to the specific context they’re being deployed within.


### How to use

For code examples, we refer to the [documentation](https://huggingface.co/docs/transformers/main/en/model_doc/blip-2#transformers.Blip2ForConditionalGeneration.forward.example).

### Memory requirements

The memory requirements differ based on the precision one uses. One can use 4-bit inference using [Bitsandbytes](https://huggingface.co/blog/4bit-transformers-bitsandbytes), which greatly reduce the memory requirements.

| dtype             | Largest Layer or Residual Group | Total Size | Training using Adam |
|-------------------|---------------------------------|------------|----------------------|
| float32           | 490.94 MB                       | 14.43 GB   | 57.72 GB             |
| float16/bfloat16  | 245.47 MB                       | 7.21 GB    | 28.86 GB             |
| int8              | 122.73 MB                       | 3.61 GB    | 14.43 GB             |
| int4              | 61.37 MB                        | 1.8 GB     | 7.21 GB              |

#### Running the model on CPU

<details>
<summary> Click to expand </summary>

```python
import requests
from PIL import Image
from transformers import Blip2Processor, Blip2ForConditionalGeneration

processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b")

img_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/demo.jpg' 
raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')

question = "how many dogs are in the picture?"
inputs = processor(raw_image, question, return_tensors="pt")

out = model.generate(**inputs)
print(processor.decode(out[0], skip_special_tokens=True).strip())
```
</details>

#### Running the model on GPU

##### In full precision 

<details>
<summary> Click to expand </summary>

```python
# pip install accelerate
import requests
from PIL import Image
from transformers import Blip2Processor, Blip2ForConditionalGeneration

processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", device_map="auto")

img_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/demo.jpg' 
raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')

question = "how many dogs are in the picture?"
inputs = processor(raw_image, question, return_tensors="pt").to("cuda")

out = model.generate(**inputs)
print(processor.decode(out[0], skip_special_tokens=True).strip())
```
</details>

##### In half precision (`float16`)

<details>
<summary> Click to expand </summary>

```python
# pip install accelerate
import torch
import requests
from PIL import Image
from transformers import Blip2Processor, Blip2ForConditionalGeneration

processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", torch_dtype=torch.float16, device_map="auto")

img_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/demo.jpg' 
raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')

question = "how many dogs are in the picture?"
inputs = processor(raw_image, question, return_tensors="pt").to("cuda", torch.float16)

out = model.generate(**inputs)
print(processor.decode(out[0], skip_special_tokens=True).strip())
```
</details>

##### In 8-bit precision (`int8`)

<details>
<summary> Click to expand </summary>

```python
# pip install accelerate bitsandbytes
import torch
import requests
from PIL import Image
from transformers import Blip2Processor, Blip2ForConditionalGeneration

processor = Blip2Processor.from_pretrained("Salesforce/blip2-opt-2.7b")
model = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-2.7b", load_in_8bit=True, device_map="auto")

img_url = 'https://storage.googleapis.com/sfr-vision-language-research/BLIP/demo.jpg' 
raw_image = Image.open(requests.get(img_url, stream=True).raw).convert('RGB')

question = "how many dogs are in the picture?"
inputs = processor(raw_image, question, return_tensors="pt").to("cuda", torch.float16)

out = model.generate(**inputs)
print(processor.decode(out[0], skip_special_tokens=True).strip())
```
</details>