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.

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 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 ) 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.

Ethical Considerations

This release is for research purposes only in support of an academic paper. Our models, datasets, and code are not specifically designed or evaluated for all downstream purposes. We strongly recommend users evaluate and address potential concerns related to accuracy, safety, and fairness before deploying this model. We encourage users to consider the common limitations of AI, comply with applicable laws, and leverage best practices when selecting use cases, particularly for high-risk scenarios where errors or misuse could significantly impact people’s lives, rights, or safety. For further guidance on use cases, refer to our AUP and AI AUP.

How to use

For code examples, we refer to the documentation.

Memory requirements

The memory requirements differ based on the precision one uses. One can use 4-bit inference using 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

Click to expand

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())

Running the model on GPU

In full precision

Click to expand

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())

In half precision (`float16`)

Click to expand

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())

In 8-bit precision (`int8`)

Click to expand

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())

Salesforce/blip2-opt-2.7b

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

Model description

Direct Use and Downstream Use

Bias, Risks, Limitations, and Ethical Considerations

Ethical Considerations

How to use

Memory requirements

Running the model on CPU

Running the model on GPU

In full precision

In half precision (float16)

In 8-bit precision (int8)

Salesforce/
blip2-opt-2.7b

In half precision (`float16`)

In 8-bit precision (`int8`)