Ah, I reverted FlashAttention in this run because it made code messier. Will look into incorporating it back, but yes not sure how nicely it plays with torch.compile. The usual problem with taking on large dependencies you don't understand ;(
AI
-
PyTorch Compile Optimization Reduces Iteration Time by 36%
By
–
To follow up, I had a chance to try it btw:
before: 212ms / iter
>>> model = torch.compile(model)
after: 135ms / iter
nice 🙂 -
Stable Diffusion v3: Future Without Prompts
By
–
"Where we're going, we're not going to need prompts." – Emad on the future of Stable Diffusion (v3 coming in a few months! )
-
Chain-of-Thought Prompting Demo on Poe Chatbots
By
–
Another demo — chain-of-thought prompting on Poe (“Dragonfly” — there’s multiple chatbots) using the “format trick”: https://
poe.quora.com/s/Tycgjooa3Aio
SfOfLZsN
… -
New AI Chat Interface Offers Natural Speech and Reduced Latency
By
–
Okay, I’m cleared to say more. It’s like ChatGPT, but the speech is more natural and terse, like texting. Much better latency (hope it lasts). Social sharing good enough to replace sharing screenshots on Twitter. Here’s me talking to Poe: https://
poe.quora.com/s/YPMAqBPuDShG
g3kl1mbt
… -
Interior AI: Plain Stable Diffusion No Training
By
–
Interior AI is plain Stable Diffusion no training
-
AI Photo Pack Pricing Guide: $15 Without Video
By
–
around $15 for AI Photo pack without video make sure you buy the right one
-
Experiment-Based Goals Over Outcome Targets in Tech
By
–
A management lesson we learned early on in Dota. We'd initially set goals of the form "beating X player by Y date", but when we fell behind, wasn't clear to anyone how to speed it up since we didn't know what would work. Great results from instead "try X experiment by Y date".
-
Prompt Engineering Remains, But Five New More Valuable Things Will Emerge
By
–
prompt eng will always be "a thing" but this time next year we'll have found 5 more new "things" that are more valuable to build on
-
AI Model Conditioning: Prompting vs. Fine-Tuning Techniques
By
–
Generally I’d call that “conditioning the model”, which can take the form of either prompting or fine-tuning, but not sure if that’s exactly the right term. They’re two different techniques to make it respond how you want.