-
https://git.ecker.tech/ aims to provide a place to share my efforts while maintaining true ownership of my code, as I do not trust GitHub.
XMR: 4B9TQdkAkBFYrbj5ztvTx89e5LpucPeTSPzemCihdDi9EBnx7btn8RDNZTBz2zihWsjMnDkzn5As1LU6gLv3KQy8BLsZ8SG
- Joined on
Oct 10, 2022
mrq commented on issue mrq/ai-voice-cloning#484
So is this abandoned?> So is this abandoned? Pretty much. The *extremely short* of it, is that this repo itself is in dire need of a rewrite. > If it is dead, anyone have suggestions on replacements with a UI…
mrq pushed to master at mrq/vall-e
- 7075c2a5f0 added an option to allow injecting embeddings from another model, because it dawned upon me how valuable embeddings from a good model can be for subsequent trainings (defined under cfg.models._embeddings as a relative path to the yaml)
- 91062361af tweaks
- f3c59c3e7e cleaner replacement code (because I realized BitNet had an implementation for it too), added calculating gradient norm and performing gradient clipping in local trainer (non-deepspeed)
- 47435207f7 Added cfg.bitsandbytes.replace as a less intrusive alternative to cfg.bitsandbytes.inject to replace all Linear modules in a model
- 0427d8d076 logger broke for some reason, added flag to just tqdm.write instead, make cfg.bitsandbytes.bitnet==True yamls denoted since I'm sure they're not interoperable
- Compare 13 commits »
mrq pushed to master at mrq/vee-speedrun-ratings
- f4227ef673 Another grim GDQ
mrq pushed to master at mrq/vall-e
- 6c51a629cc resetting step count resets the samples processed and other metrics
mrq pushed to master at mrq/vall-e
- 0aa2a3cc07 evaluation/validation passes language ID during training (oops)
mrq pushed to dreamcast at mrq/reactphysics3d
- f61c6c5713 Dreamcast fixes, also that experimental float16/quantized uint16_t triangle mesh support
mrq pushed to dreamcast at mrq/reactphysics3d
- 2958d57725 Dreamcast fixes, also that experimental float16/quantized uint16_t triangle mesh support
mrq pushed to master at mrq/reactphysics3d
- 1332aa7990 Agnostified CMakeList to deduce whether to inject KOS's GCC flags; support for using float16/bfloat16 vertices for triangle mesh
mrq pushed to master at mrq/reactphysics3d
- 4738a640b0 One of these flags fixes the '__gnu_cxx::__concurrence_lock_error' error on Dreamcast
mrq pushed to master at mrq/reactphysics3d
- bd94c476a3 Added fallback quantized-float-as-uint16_t for envs without float16 (like KOS's dc-chain)
mrq pushed to master at mrq/reactphysics3d
- dccb584c82 Agnostified CMakeList to deduce whether to inject KOS's GCC flags; support for using float16/bfloat16 vertices for triangle mesh (experimental)
mrq pushed to master at mrq/reactphysics3d
- 8ca1798ef9 Agnostified CMakeList to deduce whether to inject KOS's GCC flags; support for using float16/bfloat16 vertices for triangle mesh (experimental)
mrq pushed to master at mrq/ai-voice-cloning
- 94f88886b0 updated colab notebook
mrq commented on issue mrq/ai-voice-cloning#427
Google Colab Notebook Not Working> I'm not sure how I would edit the notebook to make it work properly, would I have to update the commands in the repo files or just run the commands inside the colab? If I recall right, there…
mrq commented on issue mrq/ai-voice-cloning#428
Discussions Tab?You'll need to re-create your training YAML and specifying the `Resume State Path`. It's a bit of a bother with how DLAS requires it to be defined, rather it be automatically deduced.
mrq commented on issue mrq/ai-voice-cloning#428
Discussions Tab?> because I can't find a discussions tab. Gitea still doesn't really have an analogue to Github's discussion tabs, so the best is just opening issues here and hoping someone else (or I) will…