Meta releases OPT-175B

Cognitive said:

Unfortunately..."License: OPT-175B and the smaller baseline models are made available through a non-commercial use license agreement provided in our model license." As far as I know (so I'm not sure about this) one needs to request access to download the weights.

But the good news is that BigScience (not related to Meta) is training an 176B AI open-source AI model on the French Jean Zay supercomputer (28 petaFLOPS from multiple Nvidia A100 and V100 GPUs). The model is multilingual (trained on 1.5 TB of text, in 46 languages) and it should be ready this summer
https://huggingface.co/bigscience/tr11-176B-ml-logs

Click to expand...

Cognitive said:

Unfortunately..."License: OPT-175B and the smaller baseline models are made available through a non-commercial use license agreement provided in our model license." As far as I know (so I'm not sure about this) one needs to request access to download the weights.

But the good news is that BigScience (not related to Meta) is training an 176B AI open-source AI model on the French Jean Zay supercomputer (28 petaFLOPS from multiple Nvidia A100 and V100 GPUs). The model is multilingual (trained on 1.5 TB of text, in 46 languages) and it should be ready this summer
https://huggingface.co/bigscience/tr11-176B-ml-logs

Click to expand...

Cognitive said:

As far as I know (so I'm not sure about this) one needs to request access to download the weights.

Click to expand...

Steptoe said:

Smaller weights are here: https://github.com/facebookresearch/metaseq/tree/main/projects/OPT#pretrained-model-weights
For research purposes only of course

Click to expand...

Cognitive said:

Unfortunately..."License: OPT-175B and the smaller baseline models are made available through a non-commercial use license agreement provided in our model license." As far as I know (so I'm not sure about this) one needs to request access to download the weights.

But the good news is that BigScience (not related to Meta) is training an 176B AI open-source AI model on the French Jean Zay supercomputer (28 petaFLOPS from multiple Nvidia A100 and V100 GPUs). The model is multilingual (trained on 1.5 TB of text, in 46 languages) and it should be ready this summer
https://huggingface.co/bigscience/tr11-176B-ml-logs

Click to expand...

takitaki007 said:

Crap!

Missed that part. All current AI's I have found uses GPT-3 and it's against their ToS to use it for gambling content. So for me that operates in that niche is depsperatly looking at alternatives and I didn't see anything about that in Meta's ToS for the OPT-175B.

Click to expand...