True. But it really becomes a pain to observe the changes for a single speaker TTS training runs.
@reach_vb
-
Diffusion Models Brittleness and Non-Determinism Issues
By
–
The more I deal with diffusion models – the more brittle they appear. The non-determinism is low-key pissing me off specially for TTS.
-
Hugging Face Inference Endpoints Scale Well One Click Deploy
By
–
Hugging Face Inference endpoints scale quite well – https://
huggingface.co/inference-endp
oints
… Most of these already have one click deploy available 🙂 -
Kingmaker Strategy and Underground Innovation in Tech
By
–
Two recent but strongly held believes: 1. Being king maker is far more comfier than being a king – often nets way more benefits too! 2. Fuck the hype, useful stuff is often found in the trenches. Find and nurture it!
-
42-Hour Work Days: AI Automation and Labor Ethics
By
–
I mean.. realistically 42 hour work days is all you need!
-
CamemBERTa Model Weights Released on Hugging Face
By
–
Always love to see the model weights being released along with the code: https://
huggingface.co/almanach/camem
berta-base
… -
Network optimization breakthrough: mysterious changes deliver smooth performance
By
–
Woke up, and saw that the changes I made actually worked and the network is way more smooth now Me: *I have no fkn clue what changes I made* In the words of Feeki – the model made a comeback – https://
open.spotify.com/track/0M1bJPyl
KSYGyNU3v1HhtQ
… -
Dataset sharing on Hugging Face Hub for AI community
By
–
Wohoo! This is fantastic! @gentaiscool – would you be open to share the dataset over on Hugging Face Hub, I believe this can have a massive impact! Happy to help ofc!
-
Things You Do to Save the Planet
By
–
Yes! Things you do to save the planet. pic.twitter.com/8Rdv9HMNGJ
— Vaibhav (VB) Srivastav (@reach_vb) 4 mai 2023Yes! Things you do to save the planet.
-
Learning ML by Dissecting Code and Dummy Tensors
By
–
I usually read the experimental details and then straight up dive into the code.
Pick it apart till I’m down to bare metal. Create dummy tensors and then play around with individual modules to understand what is happening. Colabs are helpful here too, to isolate the code.