Neither the source code nor the weights are open source... This is actually worse than Stability AI's previous offering, in that regard.
DeepFloyd IF: open-source text-to-image model
11–20 of 237 posts
Re: DeepFloyd IF: open-source text-to-image model
#12The examples on the README are extremely compelling; the state of the art has been raised yet again.
Re: DeepFloyd IF: open-source text-to-image model
#1316GB VRAM minimum is a bit steep. Sadly excludes my 3080 which is annoying because I'd like something better than Stable Diffusion locally.
Once these are quantized (I assume they can be), they should be ~1/4th the size.
Can anyone explain why it needs so much ram in the first place though? 4.3B is only ~9GB at 16bit (I'm not as familiar with image models).
I'm really happy to see that fits under 24GB - that's what I consider the limit for being able to run on "consumer hardware".
Re: DeepFloyd IF: open-source text-to-image model
#14It looks like the model on Hugging Face either hasn't been published yet or was withdrawn. I got this error in their Colab notebook:
OSError: DeepFloyd/IF-I-IF-v1.0 is not a local folder and is not a valid model identifier listed on 'https://huggingface.co/models' If this is a private repository, make sure to pass a token having permission to this repo with `use_auth_token` or log in with `huggingface-cli login` and pass `use_auth_token=True`.
Re: DeepFloyd IF: open-source text-to-image model
#15It looks like the model on Hugging Face either hasn't been published yet or was withdrawn. I got this error in their Colab notebook: OSError: DeepFloyd/IF-I-IF-v1.0 is not a local folder and is not a valid model identifier listed on ' https://huggingface.co/models ' If this is a private repository, make sure to pass a token having permission to this repo with `use_auth_token` or log in with `huggingface-cli login` an…
You need to accept the license on the HuggingFace model card.
Re: DeepFloyd IF: open-source text-to-image model
#16It looks like the model on Hugging Face either hasn't been published yet or was withdrawn. I got this error in their Colab notebook: OSError: DeepFloyd/IF-I-IF-v1.0 is not a local folder and is not a valid model identifier listed on ' https://huggingface.co/models ' If this is a private repository, make sure to pass a token having permission to this repo with `use_auth_token` or log in with `huggingface-cli login` an…
You need to accept the license on the HuggingFace model card.
it doesn't seem like they have anything published https://huggingface.co/DeepFloyd
Re: DeepFloyd IF: open-source text-to-image model
#17Re: DeepFloyd IF: open-source text-to-image model
#18It looks like the model on Hugging Face either hasn't been published yet or was withdrawn. I got this error in their Colab notebook: OSError: DeepFloyd/IF-I-IF-v1.0 is not a local folder and is not a valid model identifier listed on ' https://huggingface.co/models ' If this is a private repository, make sure to pass a token having permission to this repo with `use_auth_token` or log in with `huggingface-cli login` an…
You need to accept the license on the HuggingFace model card.
https://huggingface.co/DeepFloyd/IF-I-IF-v1.0 is a 404 currently.
Re: DeepFloyd IF: open-source text-to-image model
#19New restriction in their License suggests the software can't be modified.
"2. All persons obtaining a copy or substantial portion of the Software, a modified version of the Software (or substantial portion thereof), or a derivative work based upon this Software (or substantial portion thereof) must not delete, remove, disable, diminish, or circumvent any inference filters or inference filter mechanisms in the Software, or any portion of the Software that implements any such filters or filter mechanisms."
Re: DeepFloyd IF: open-source text-to-image model
#2016GB VRAM minimum is a bit steep. Sadly excludes my 3080 which is annoying because I'd like something better than Stable Diffusion locally.
Once these are quantized (I assume they can be), they should be ~1/4th the size. Can anyone explain why it needs so much ram in the first place though? 4.3B is only ~9GB at 16bit (I'm not as familiar with image models). I'm really happy to see that fits under 24GB - that's what I consider the limit for being able to run on "consumer hardware".
>Can anyone explain why it needs so much ram in the first place though?
The T5-XXL text encoder is really large, also we do not quantize the UNets, the UNet outputs 8-bit pixels, so quantizing the UNet to that precision will create pretty bad outputs.