everything GPU @mixedbreadai | Research, ColBERT, DSPy @Stanford

Stanford
Joined December 2013
DSRs, @DSPyOSS for Rust is here馃殌 Happy to finally announce the stable release of DSRs. Over the past few months, I鈥檝e been building DSRs with incredible support and contributions from folks Maguire Papay, @tech_optimist, and @joshmo_dev. A big shout out to @lateinteraction and @ChenMoneyQ who were the first people to hear my frequent rants on this!! Couldn't have done this without all of them. DSRs originally started as a passion project to explore true compilation and as it progressed I saw it becoming more. I can鈥檛 wait to see what the community builds with it. DSRs is a 3 phase project: 1. API Stabilization. We are nearly done with this and it was mostly implementing the API design. We kept the DSPy style in mind and tried to keep it close to it so it's easier to onboard and while at it we tried to improve it and make it a bit more idiomatic and intuitive! 2. Performance Optimisation with benchmarking vs DSPy. We want to benchmark LLMs performance vs DSPy, with API design finalized we want to improve performance in every front. We'll improve the latency and improve the templates and optimizers in DSRs. 3. True Module Compilation. Why should you optimize signature when you can optimize and fuse much more? This is the idea of the final phase of DSRs. A true LLM workflow compiler. More on this after Phase 2. Really grateful for @PrimeIntellect offering compute to drive Phase 2 and 3 experimentation for this! Big shoutout to them and @johannes_hage for this!!! But what is DSRs? What does it offer? Let's see.
12
24
8
210
41,018
Herumb Shandilya 馃 retweeted
And if you think that鈥檚 even close to a ceiling, stay tuned
Some strong embeddings models have been released recently. Ours is still better. We benchmarked Embed 5 Fast against Google's EmbeddingGemma 2 across 163 datasets using RCP-nDCG@10. Embed 5 Fast leads on over 95% of them.
3
66
5,334
I really miss coding by hand 馃ゲ
5
1
36
1,433
Working from SF tower is something else 馃檹馃徎
1
27
727
Open Owl really cooked with this one ngl
2
272
Make friends that bring you gifts like this. Thanks @halcyonrayes! Great book @philipkiely 馃挴
25
660
Herumb Shandilya 馃 retweeted
I'm joining Mixedbread to build better information retrieval for AI agents
33
2
3
172
12,260
This @gepa_ai interface is soooo gooooodddd, Fable really cooks UI from vague ideas
1
2
46
2,671
Finally some signal after two weeks... Win is not just on winner prompt size馃槸
1
21
1,086
It鈥檚 still GEPA with some changes :)
1
1
8
927
Using GEPA to optimize @mixedbreadai's toast-1 harness! Wrote up the recipe. Coming in a few days.
2
9
1
69
6,747
Thinking to give writing another shot
1
12
592
Finally a @DSPyOSS meetup that doesn鈥檛 clash with exams :)
1
3
30
1,364
all this meme knowledge and no one to get it
4
1
104
3,821
Repos are huge these days. Not every question should mean digging through a thousand lines of code yourself or burning $20 on an agent to do it for you. Introducing Repofetch Talk to any GitHub repo, built on @mixedbreadai's toast-1 model.
11
11
4
70
7,057
Every answer points back to the code. Click a citation and it opens the file directly at those lines.
1
15
198
Chats stay in your browser. Mixedbread handles retrieval, toast-1 answers by default, and you can switch answers to OpenRouter if you want. Try it on your own repo: repofetch.vercel.app/
1
15
205