Latent consistency model: make very fast Stable diffusion images with just 4 to 8 steps (~1 second per image) https://
replicate.com/luosiallen/lat
ent-consistency-model
…
OPEN SOURCE
-
Latent Consistency Model: Ultra-Fast Stable Diffusion Image Generation
By
–
-
SDXL Canny ControlNet Now Supports LoRA Integration
By
–
SDXL Canny controlnet with LoRA support: Use your Replicate trained loras with Canny controlnet
-
Regulation Strategy: Product Safety vs Research Freedom
By
–
Incidentally:
– I am not anti-regulation.
– regulating the deployment of *product* is useful to guarantee safety. – but I'm very much against regulating research and development.
– and I'm very much against regulations that would make open source AI illegal or difficult. -
ALiBi Models Performance Extension Without Fine-Tuning
By
–
Unlike existing techniques – this method requires no additional fine tuning. By setting a few parameters in your HuggingFace config.json, you can instantly extend the performance of ALiBi models such as BTLM-3B-8K by ~2x: https://
huggingface.co/cerebras/btlm-
3b-8k-base#during-inference-without-fine-tuning
… -
Position Interpolation Doubles Context Length for ALiBi Models
By
–
Paper drop: Position Interpolation Improves ALiBi Extrapolation We found a simple method to 2x the context length of models that use ALiBi. This lets models like BTLM-3B-8K and MPT-7B-8K run high quality inference at up to 16K with no additional fine tuning.
-
OSS VAT Compliance for European Creators
By
–
If your a creator in Europe, how do you deal with OSS VAT? Seems like a jungle, somebody must have solved this?
-
Uncensored AI, essential fight of the next 10 years
By
–
In 10 years, we will look back and realize we should have been fighting for uncensored AI this whole time.
-
Open Research Communities Drive Innovation and Success
By
–
Which is why open research communities wins.
-
Military R&D Lags Behind Due to Closed Innovation Models
By
–
Yup.
And that's why military R&D is years behind.
Closed R&D organisations inevitably trail behind open ones because the set of potential contributors is smaller and less diverse.