Changed the model VRAM requirements to what you'd need to comfortably run the model rather than barely (Like with the manual). Will probably revise this in a later commit.
More importantly, it now supports models that use </s> which will be required to support XGLM and Fairseq models.
My last attempt at fixing this caused GPT2 to break, since the other fix is an edge case we assume that the GPT2 method should be used, and if that fails we try the other one to catch rare errors with bad model config's.
More settings reordering so similar settings are on the same rows now that we have more settings for the repetition penalty. Amount to generate is now top left so some muscle memory may be lost with the temp. But the settings that control AI randomness are on the same row now, and repetition related settings are next to each other as well.
Turns out model_config does not work on models that have no model_type defined. In case this happens we now fall back to the old .json loading method. This will not work in --colab mode if its not already a local model, but since almost all modern models define a model type and to my knowledge all models on huggingface do that should not be an issue. If it is we can always ask the model creator to either update it, distribute the model differently or load that model with --remote instead of --colab.
This allows Colab developers to first get the correct folder structure on drive, before placing a configuration file for the model. That way we can quickly add the settings for a model without maintaining the init settings in two different lines. Its a substitute to the common --init only and --init skip approach from before.
henk717
Gnome Ann
jojorne
ebolam