>
O
Model Details
Introduction
We introduce OpenFlamingo, a family of autoregressive vision-language models ranging from 3B to 9B parameters. OpenFlamingo is an ongoing effort to produce an open-source replication of DeepMind's Flamingo models. On seven vision-language datasets, OpenFlamingo models average between 80 - 89% of corresponding Flamingo performance. This technical report describes our models, training data, hyperparameters, and evaluation suite. We share our models and code at this https URL.
Benchmarking
Notes: 9B model is trained using 64 A100-80GBs in 16bf. Length of training not stated. Might be possible to calculate training compute using operations counting method.
Training
Training Code Accessibility
MIT. code and weights:
https://github.com/mlfoundations/open_flamingo?tab=readme-ov-fileHardware
NVIDIA A100 SXM4 80 GBHardware Quantity
64Size Notes: "OpenFlamingo models were trained for 60M interleaved (MMC4) examples1 and 120M LAION-2B examples." Table 2 and Figure 4 provide details on the length of each example. MMC4 has median of 2 images and 256 text tokens per sample, while LAION-2B has 1 and 17, respectively. Suggests a total of around 240M images and 17.4B text tokens. Prediction task is next (text) token prediction; images are only used for conditioning.
Parameters
Parameters
9000000000Notes: "We introduce OpenFlamingo, a family of autoregressive vision-language models ranging from 3B to 9B parameters."
Authors
Anas Awadalla, Irena Gao, Josh Gardner, Jack Hessel, Yusuf Hanafy, Wanrong Zhu, Kalyani Marathe, Yonatan Bitton, Samir Gadre, Shiori Sagawa, Jenia Jitsev, Simon Kornblith, Pang Wei Koh, Gabriel Ilharco, Mitchell Wortsman, Ludwig Schmidt