@esfrankel Profile picture

Eric Frankel

@esfrankel

f-divergence enthusiast | phd @uwcse | math + stats @stanford

Similar User
Mihir Patel photo

@mvpatel2000

Vivek Myers photo

@vivek_myers

Sumer Sao photo

@sumersao

Rohan Deshpande photo

@rohankalvade

Tyler Shibata photo

@ShibataTyler

Adi Sidapara photo

@adisidapara

ankush swarnakar photo

@ankushswar1

Dhruvik Parikh 🌎 photo

@dhruvik

Jack Yuan photo

@JackYuan35

Yasmeena Khan photo

@yasmeena_khan

Lena Han photo

@_Lena_Han

Shuvom Sadhuka photo

@shuvom_s

izzy photo

@occamsrrazor

SGA hater photo

@lahtekboy

Connor Hogan photo

@co__ho

Bello! I got 100% on GSM8K @Bobisminion

Tweet Image 1

Eric Frankel Reposted

🐪🚨 I recently added Tulu 2 Llama 3.1 finetunes to the Llama 3 Tulu 2 suite (huggingface.co/collections/ha…). We get a small boost from using the updated base models! Lots way more exciting updates coming soon hopefully :)

Tweet Image 1

Eric Frankel Reposted

🔥We release the first open-source 1.4T-token RAG datastore and present a scaling study for RAG on perplexity and downstream tasks! We show LM+RAG scales better than LM alone, with better performance for the same training compute (pretraining+indexing) retrievalscaling.github.io 🧵


Eric Frankel Reposted

1/ Excited to share our new paper “Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making”! Link: arxiv.org/abs/2406.17098… Code: github.com/vivekmyers/con… #ICML2024

Tweet Image 1

Eric Frankel Reposted

Happy to share LightSeq is accepted by @COLM_conf 🥳 LightSeq supports efficient long-context Transformer training, where the supported context length grows with the number of nodes. We are excited about the innovative applications it will enable, such as long-context LLM/VLM! 🚀

Introduce LightSeq for long-context LLM training: - Highly optimized for decoder models - smarter checkpointing - better support for fewer heads models up to 2x faster, 2-8x longer sequences vs Megatron-LM. arxiv.org/abs/2310.03294



Eric Frankel Reposted

Will training on AI-generated synthetic data lead to the next frontier of vision models?🤔 Our new paper suggests NO—for now. Synthetic data doesn't magically enable generalization beyond the generator's original training set. 📜: arxiv.org/abs/2406.05184 Details below🧵(1/n)


Eric Frankel Reposted

Can we computationally “read the room”👀? Our work ValueScope🔍can help you get a digital vibe check🌟! We measure normative behaviors and track the evolution of norms and values in response to world events 📈 to inform moderation and engagement 🌐 👉arxiv.org/abs/2407.02472 🧵

Tweet Image 1

Eric Frankel Reposted
Tweet Image 1

A dictatorial, unelected majority in the Supreme Court has just rendered America a dictatorial president above the law. Thank you Hillary Clinton, whose blundering campaign let the dictatorial Trump become president and led to a rightwing dictatorial majority on the Supreme…



Eric Frankel Reposted

BREAKING: in a 9-0 decision, the black-robed Supreme Injustices of Mordor have ruled that the Dark Lord has immunity for official acts undertaken while wielding the One ring to rule them all.

Tweet Image 1

Eric Frankel Reposted

For those of you who refused to vote for Hillary, this is the SCOTUS you get.


Eric Frankel Reposted

I can’t believe the Supreme Court just ruled that the president can sleep outdoors in a public space.


Eric Frankel Reposted

You're voting for the chance to replace these guys

Tweet Image 1

Eric Frankel Reposted

Biden has the opportunity to do the funniest thing ever...


Eric Frankel Reposted

Happy to share some small updates to the Tulu suite! We’re releasing ~Llama 3 Tulu 2~ update: llama 3 finetunes + DPO tunes that outperform the original Tulu V2 suite. Reward models included! 🐪find the models here: huggingface.co/collections/ha… ⬇️some results below!

Tweet Image 1

Eric Frankel Reposted

Tweet thread tomorrow! 🤗 A few models here: huggingface.co/collections/al…

Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback abs: arxiv.org/abs/2406.09279 "In this work, we aim to systematically investigate these key components of learning from preferences" Exploration of Tulu 2 13B (Llama-2-13B SFT) with 14…

Tweet Image 1


Eric Frankel Reposted

Civilization in Phoenix and its surrounding valley depends on solving the problem of water. Solving the problem of water depends on solving the problem of democracy. In The Atlantic’s latest cover story, George Packer explores if America is up to the task. theatln.tc/owos9R46

Tweet Image 1

Eric Frankel Reposted

Aw shit the model forecasted avg Passer rating of 112 this season so we’re in violation of our covenants and Apollo is gonna skin Zach Wilson alive

Tweet Image 1

Eric Frankel Reposted

We going to Wembley. Innit.


Eric Frankel Reposted

🥹😊😇

Tweet Image 1

Loading...

Something went wrong.


Something went wrong.