Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering
@jxmnop
-
Unified Decoder Models Gaining Adoption in Modern AI Systems
By
–
yeah, you're right; i think some people are doing everything in one big decoder model these days though
-
Are Encoder-Decoder Models Like T5 and BART Still Relevant?
By
–
is anyone still using encoder-decoder models (T5, BART, etc.)? if so — for what?
-
Debugging Pattern: Fix Error Without Understanding Root Cause
By
–
fun coding pattern
> get an error message that i don't understand
> go to code
> spot something obviously wrong
> fix it
> run code again
> error goes away
> move on, never fully understanding the twisted complex incorrect state i got myself into -
Intellectual Growth Through Academic Research and Talks
By
–
Research: finally, intellectually, i'm doing the best work of my life. the talks here are also phenomenal. it's kind of like with your favorite bands, all the academics you want to meet will come through new york city every year or two. this semester we've had talks on campus i
-
Cornell’s NLP/ML Culture: Academic Excellence Meets Work-Life Balance
By
–
People: cornell obviously is very strong academically but i have noticed compared to other similar-caliber schools people here are remarkably chill. people don't really work nights/weekends and encourage you to have work-life balance. everyone in nlp/ML is fun to hang out with
-
Fixing Multiprocessing Hangs in Large-Scale Model Inference
By
–
yeah the multiprocessing in this method isn't properly handled and hangs if you scale to millions of sentences. i (and others) have struggled with it a lot. you can just wrap your model in DataParallel and do a regular inference loop, no need for explicit multiprocessing
-
Model Interpretability: How Neural Networks Scatter and Refine Information
By
–
the clarity in these examples is a little startling to me; this is such a good way of doing interpretability; it's interesting that the model seems to scatter details like what Einstein looks like across multiply layers, and it is able to gradually refine throughout the fwd pass
-
New Research Paper from Boknilev Group on AI
By
–
this is the paper, it's quite new, from @boknilev
's group! https://
arxiv.org/pdf/2403.05846
.pdf
… -
Diffusion Lens reveals abstract-to-concrete representation layers in image encoders
By
–
Diffusion Lens is a pretty neat new paper, you can see a text-to-image encoder's representation of a giraffe getting less and less abstract with every layer