Hanna Hajishirzi
@HannaHajishirzi
Followers
6,672
Following
362
Media
6
Statuses
527
Associate professor at @uw_cse ; senior director at @allen_ai co-leading @allenNLP ; AI/NLP researcher at @uw_nlp
Seattle, WA
Joined February 2019
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
North Carolina
• 878067 Tweets
Hurricane Helene
• 689427 Tweets
EP4 U STEAL MY HEART
• 390507 Tweets
FEMA
• 331183 Tweets
#FreeMorara
• 320207 Tweets
Mario
• 205923 Tweets
Dikembe Mutombo
• 132303 Tweets
Valdosta
• 125596 Tweets
Mimi
• 71671 Tweets
Verizon
• 60912 Tweets
Mets
• 57161 Tweets
HAPPY JIMTOBER
• 45806 Tweets
Indigenous
• 45547 Tweets
#النصر_الريان
• 41526 Tweets
#الاهلي_الوصل
• 38846 Tweets
Braves
• 38208 Tweets
Franklin Graham
• 32156 Tweets
MANTRA IS COMING
• 26047 Tweets
#KYSvBJK
• 24403 Tweets
Onana
• 20957 Tweets
Kemp
• 20325 Tweets
Powell
• 18975 Tweets
Semih
• 17004 Tweets
#KızılGoncalar
• 14784 Tweets
Kayseri
• 13414 Tweets
الجيش اللبناني
• 11600 Tweets
My alma mater, Sharif University of Technology, Iran's premier university, was under siege yesterday. Many students were arrested, heavily injured or perhaps killed. Many Iranian scholars and PhD students whom you know have received their BSc degrees from this university.
1
57
391
A bit of outdated news: I am spending more time at AI2
@allen_ai
now, co-leading
@ai2_allennlp
with
@nlpnoah
At
@ai2_allennlp
, we are hiring postdocs and research scientists with strong foundation in NLP.
Links to apply:
4
38
244
Introducing our best OLMo yet. OLMo 1.7-7B outperforms LLaMa2-7B, approaching LLaMa2-13B at MMLU and GSM8k. High-quality data and staged training are key.
I am so proud of our team making such significant improvement in a short period after our first release.
Announcing our latest addition to the OLMo family, OLMo 1.7!🎉Our team's efforts to improve data quality, training procedures and model architecture have led to a leap in performance. See how OLMo 1.7 stacks up against its peers and peek into the technical details on the blog:
13
47
169
2
14
107
Check out these papers from our H2lab at
@uwnlp
and
@allen_ai
and come talk to my amazing students at these EMNLP sessions
@emnlp2020
or about their papers at Findings of EMNLP.
1
12
93
Intuitive, simple inference-time decoding for LLMs: PPO+ Monte Carlo Tree Search.
Just don’t throw out learned values after PPO training. They are important at decoding!
What if we combine PPO with Monte-Carlo Tree Search – the secret sauce for AlphaGo to reach superhuman performance?
Spoiler: MAGIC!! Our inference-time decoding method, PPO-MCTS, achieves impressive results across many text generation tasks.
📜
🧵(1/n)
9
101
455
0
13
86
So proud of you
@sewon__min
! We were so lucky to have you at UW; congrats to
@ucsbNLP
@berkeley_ai
for their great hire; now counting days to your start date at
@allen_ai
📣 After graduating from
@UWCSE
, I am joining
@UCBerkeley
as an Assistant Professor (affiliated w
@berkeley_ai
@BerkeleyNLP
) and
@allen_ai
as a Research Scientist.
I'm looking forward to tackling exciting challenges in NLP & generative AI together with new colleagues! 🐻✨
159
68
2K
2
1
85
Very honored to receive Sloan Fellowship. Thanks
@SloanFoundation
#sloanfellow
Congrats to
#UWAllen
’s
@HannaHajishirzi
of
@uwnlp
&
@allen_ai
and Yin Tat Lee of the Theory of Computation group on being named 2020
@SloanFoundation
Research Fellows for their creativity, leadership, and early-career research achievements:
#UWdiscovers
0
8
52
2
5
83
Happy to announce the 2nd conference on Automated Knowledge Base Construction (AKBC),
#AKBC
#AKBC2020
with
@dipanjand
@andrewmccallum
@sameer_
@nlpmattg
@astent
@Zachives_
Christopher Re Abstract Deadline: Feb 6, Full submission deadline: Feb 13
0
17
76
We just released AI2 Dolma, a 3 Trillion open dataset for LLM pretraining.
Announcing Dolma, the dataset for
@allen_ai
's LLM, OLMo. It's 3+ trillion tokens (web/papers/code/books/wiki). We hope it will facilitate study of LLMs & their behavior!
Released on
@huggingface
w ImpACT license
Overview/datasheet
21
144
557
0
8
74
3/3 papers accepted at
@naacl
"A General Framework for Information Extraction Using Dynamic Span Graphs","Text Generation from Knowledge Graphs with Graph Transformers", "Towards Interpretable Math Word Problem Solving with Operation-Based Formalisms"
0
6
61
Unfortunately, there is not much coverage in US media; in case you are interested, you can read about it here:
I am sending this to inform those of you advising Iranian students, and ask you to keep an eye on them in this difficult time.
2
16
62
Teaching Language Vision Transformer models how to paint: Check out our
@emnlp2020
paper on how to Paint, Caption and Answer Questions with Multi-Modal Transformers
Work by
@jmin__cho
@jiasenlu
@HannaHajishirzi
@anikembhavi
to appear at EMNLP 2020
"GPT-3 trained on an enormous amount of text data. What if the same methods were trained on both text and images?"
Learn about fascinating new work from our
#computervision
team in this piece by
@_KarenHao
via
@techreview
:
0
21
47
0
11
61
Very proud of the entire OLMo team for both OLMo and Dolma projects! OLMo 1.0 shows how our team at
@allen_ai
and
@uwcse
advocates for truly open AI. Stay tuned for better models soon.
Congrats to our team for winning two paper awards at
#ACL2024
!
OLMo won the Best Theme Paper award, and Dolma won a Best Resource Paper award!
All the credit goes to the whole team for the massive group effort 🎉🎉
11
44
247
2
8
62
Check out our latest addition to the OLMo family at
@allen_ai
; this time OLMoE, a fully open MOE model, best in its class. Stay tuned for bigger sizes. Work led by amazing
@Muennighoff
and the
#olmo
team.
Releasing OLMoE - the first good Mixture-of-Experts LLM that's 100% open-source
- 1B active, 7B total params for 5T tokens
- Best small LLM & matches more costly ones like Gemma, Llama
- Open Model/Data/Code/Logs + lots of analysis & experiments
📜
🧵1/9
23
230
957
3
2
58
Check out tülu2-70B-DPO, an open mode, better than GPT3.5 in almost all evaluation benchmarks; very proud of the team.
Check out the Tulu 2 suite 🐪, a set of Llama-2 models finetuned+DPO-trained on a mixture of publicly available datasets! Our best-performing models are competitive with SoTA open models on a range of benchmarks incl. AlpacaEval and MT-Bench.
📜Paper:
3
31
131
0
3
50
So proud of the entire OLMo team. OLMo 1.0 was such a nice team effort, advocating for fully open LMs. Better models and papers coming soon!
🏆 ACL Best Theme Paper Award:
- OLMo: Accelerating the Science of Language Models by Groeneveld et al.
#NLProc
#ACL2024NLP
0
10
115
1
1
49
We are so glad that Tim is joining AI2 for his gap year. Looking forward to it.
After 7 months on the job market, I am happy to announce:
- I joined
@allen_ai
- Professor at
@CarnegieMellon
from Fall 2025
- New bitsandbytes maintainer
@Titus_vK
My main focus will be to strengthen open-source for real-world problems and bring the best AI to laptops 🧵
152
85
2K
1
0
48
Has BERT solved NLP pipelines in IE? Attend Dave’s talk on
“Entity, Relation, and Event Extraction with Contextualized Span Representations” at
#emnlp2019
session 11D at 4pm to learn more.
2
7
43
New work at
@emnlp2020
Findings. UnifiedQA that achieves SOTA in many recent QA datasets by multitask learning and moving beyond format. With
@DanielKhashabi
@sewon__min
and Aristo team at
@allen_ai
📢 New work, to appear in
@emnlp2020
-Findings 📢
UnifiedQA: Crossing Format Boundaries With a Single QA System
Joint with
@sewon__min
@HannaHajishirzi
and other collaborators at
@allen_ai
.
4
5
43
0
5
42
Congratulations to my amazing, ‘genius’ friend, colleague, collaborator, and mentor.
@YejinChoinka
, we are so proud of you and so lucky to have you at uw and AI2.
#UWAllen
@uwnlp
's
@YejinChoinka
aims to develop
#AI
with the ability to reason and communicate about the world in physical and abstract terms, like humans can do. As a 2022
#MacFellow
, she looks forward to taking the “adventurous route” in her research:
9
65
389
0
0
39
AKBC 2020 will happen June 22-25tb, including an amazing set of papers, invited speakers, and workshops. Please register and stay tuned for our final program and virtual website.
@sameer_
@nlpmattg
@dipanjand
#NLProc
#AKBC2020
AKBC 2020 registration has been open for the past week: The main conference will happen over three days (June 22-24) over five and a half hours each day. 1/n
2
11
26
0
12
34
In case this is news to you, this event is a followup to a long line of protests in response to the killing of Mahsa Amini by Iran's morality police about two weeks ago.
2
0
30
So proud of the OLMo team! Sorry I missed the award ceremony.
0
1
28
DataComp benchmark is out, a controlled environment to evaluate dataset design choice for multimodal model. Also, releasing an open model trained on a portion of the data that outcompetes CLIP.
Introducing DataComp, a new benchmark for multimodal datasets!
We release 12.8B image-text pairs, 300+ experiments and a 1.4B subset that outcompetes compute-matched CLIP runs from OpenAI & LAION
📜
🖥️
🌐
8
190
774
0
4
27
Most agent LMs are built on top of proprietary LMs or only target specific domains. Check out this cool work: an open source LM agent that jointly solves numerical, tabular, and knowledge-based reasoning tasks, even outperforming gpt4 in mixed-tool reasoning.
Introducing Husky, a unified open-source language agent for multi-step reasoning with tool-use capabilities. Husky-v1:
- Jointly solves numerical, tabular and knowledge-based reasoning tasks.
- Outperforms GPT-4-Turbo on several OOD tasks via mixed-tool reasoning.
- Features new
5
66
310
0
4
26
#acl2020nlp
#acl2020en
You can reach me at hanna.uw at RocketChat and happy to chat. You can also meet me over zoom at our paper Q&A sessions.
0
3
27
I will be missing EMNLP
#EMNLP2019
, but check out this exciting work from my group (
@seo_minjoon
,
@jmin__cho
) on Nov 6th poster session.
Facing mode collapse with your Seq2Seq?
Seq2Seq is not for a one-to-many relationship.
Check out our poster at
#emnlp2019
on Nov 6th!
"Mixture Content Selection for Diverse Sequence Generation" by Myself,
@seo_minjoon
and
@HannaHajishirzi
.
• Paper:
0
3
6
0
1
26
Congrats to my amazing student
@gabriel_ilharco
for receiving the JP Morgan fellowship.
@jpmorgan
#UWAllen
@UW
's
@gabriel_ilharco
aims to make machine learning models more reliable and versatile;
@sharma_ashish_2
develops
#AI
systems to support mental health. Both earned 2023
@jpmorgan
AI Ph.D. Fellowships for applying AI to solve real-world problems:
0
8
66
0
0
23
Check out our new work on scientific claim verification, particularly verifying COVID-19 claims at
@emnlp2020
by
@davidjwadden
@lucyluwang
@kylelostat
@armancohan
@peter9863
@MadyvanZuylen
Our work on scientific claim verification will appear in
@emnlp2020
! With
@kylelostat
,
@lucyluwang
,
@peter9863
,
@MadyvanZuylen
,
@armancohan
,
@HannaHajishirzi
.
COVID-19 fact-checking demo and SciFact leaderboard: .
Camera-ready: .
3
25
82
0
0
23
Logic-guided data augmentation and regularization to improve consistency of QA model.
#NLProc
#Acl2020
#acl202nlp
I am very excited to announce that new work with
@HannaHajishirzi
has been accepted to
#acl2020nlp
@aclmeeting
.
We investigate the inconsistencies of SOTA QA models' predictions and introduce FOL logic-guided data augmentation & consistency-based training regularization. [1/2]
4
4
141
0
0
22
Congrats to Ali,
@morastegari
, and
@xnor_ai
Exclusive: Apple acquires , edge AI spin-out from Paul Allen’s AI2, for price in $200M range
6
34
186
0
1
22
Check out this work led by
@AkariAsai
on building a new open domain cross lingual QA dataset.
Knowledge is not uniformly distributed across languages so an ideal open-retrieval QA model should search and retrieve multilingual resources.
We propose a new task 𝐗𝐎𝐑 𝐐𝐀 with a new dataset 𝐗𝐎𝐑-𝐓𝐲𝐃𝐢 QA(40k newly annotated Qs in 7 languages)
👇
4
46
178
0
1
21
Tülu models are evolving: Tülu 2.5, now with PPO, outperforming DPO.
How well do DPO and PPO work on public preference datasets? Excited to share some work exploring the effects of data, reward models, and prompts! We also find that PPO generally beats DPO, despite being more challenging engineering-wise.
📜:
More below 👇
5
55
251
0
1
19
Happy to share our recent submission on DiCENets with
@morastegari
@sachmeh
Depth-wise Convolution and Point-wise Convolution showed tremendous benefit for efficient architecture design and now we are presenting a new efficient CNN model DiCENet based on Dimension-wise convolution
#deeplearning
,
#ComputerVision
paper:
1
12
25
1
11
18
When and how retrieval should be augmented with LMs to improve accuracy and factuality, while maintaining efficiency? Introducing Self-RAG to train LMs with on demand retrieval by generating text and self-reflection tokens
Introducing Self-RAG, a new easy-to-train, customizable, and powerful framework for making an LM learn to retrieve, generate, and critique its own outputs and retrieved passages, by using model-predicted reflection tokens.
📜:
🌐:
24
169
744
0
1
16
Looking forward to having you at UW and working together.
I'm happy to share that I'll be a postdoc this fall, working with Prof.
@HannaHajishirzi
at the University of Washington. I will continue research in efficient NLP/AI. Many thanks to everyone who has helped along the journey!
4
5
125
0
0
17
Check put this great work from
@sewon__min
on introducing a new task: AmbigQA
To study natural ambiguity, we construct a new dataset with questions from NQ; we find that ambiguity is frequent, diverse & subtle. See our paper and website for data, models, and more!
Work with
@_julianmichael_
@HannaHajishirzi
@LukeZettlemoyer
1
2
26
1
3
17
Excited to share our two
#acl2019nlp
papers! "Multi-hop Reading Comprehension through Question Decomposition and Rescoring" w/
@hllo_wrld
@LukeZettlemoyer
&
@HannaHajishirzi
Paper Check out our demo which answers to multi-hop Qs! 1/
1
13
65
1
1
15
Check out efficientQA results tomorrow :)
The results of the
#NeurIPS2020
competition on Efficient Open-Domain QA will be revealed tomorrow! Find out how participants achieved new SoTA and trained powerful-but-tiny systems.
Dec 11th, 4pm GMT/11am ET/8am PT
Info:
Livestream:
1
12
80
0
1
15
@akbc_conf
will start tomorrow, June 22nd 8am PST. Our virtual website at . Tomorrow we will have talks by
@professorwcohen
@jure
, Jamie Taylor, and Luna Dong, 5 spotlight papers followed by live discussions and hallway conversations.
@sameer_
@dipanjand
1
4
14
We are so excited to have you join us at AI2 😎🤩
Excited! I've started as a Research Scientist at
@allen_ai
(on
@ai2_allennlp
) working on all things "RLHF research" - it encompasses a lot.
Let's show the world that openness can foster broadly beneficial AI. I'm excited to work with everyone who wants to make that happen.
53
31
692
0
1
15
1/1 paper accepted at CVPR; ESPNetv2: A Light-weight, Power Efficient, and General Purpose Convolutional Neural Network; stay tune for the arxiv drafts.
0
3
11
0
1
11
Congratulations to the amazing
@suinleelab
🥰🎉🎉
Truly honored to give the Distinguished Keynote at
#ISMB2024
! At the end of the keynote, I shared "My Reflections on the ISCB Innovator Award: Inspiring Future Innovators." Check it out here!
6
6
70
0
2
10
Check out Akari's new work on building a *single* question answering model that can answer questions in any language! The model trains a cross-lingual retrieval model that can retrieve passages related to a question across many different languages.
#NLProc
Can we build a *single* open-domain QA model that works in *many* languages?
We’re excited to present 𝗖𝗢𝗥𝗔 using a single retriever and generator, showing SOTA results in 26 diverse languages on XOR QA & MKQA, including the unseen languages. [1/5]
3
60
263
0
1
10
OLMo performs either better or at par with other models of similar size. We love to invite the open-source community to continue advance the science of LMs.
0
1
9
Second day of
@akbc_conf
with five amazing speakers and new papers.
The second day of
#AKBC2020
is about to start in forty minutes, with 5 exciting talks by
@LukeZettlemoyer
,
@nnakashole
,
@strubell
, Andrew Su (Scripps Research), and
@Diyi_Yang
! In addition, there will be 12 paper presentations via video stream and live poster sessions.
0
8
24
1
1
9
Congrats all!
@davidjwadden
@timalthoff
And the winner of the ICWSM-21 Outstanding Study Design Paper is “The Effect of Moderation on Online Mental Health Conversations”. Congratulations to David Wadden, Tal August, Qisheng Li, and Tim Althoff! 🌠🎆
#ICWSM2021
0
3
25
0
0
9
@burkov
Compiling this table was really helpful to us throughout the training process; helped us chase down loss spikes; we thought it is great to share it with other researchers. We ablated several choices, not all of them though! we would love it if others carry out more ablations!
0
0
8
New
#acl2020nlp
paper "Logic-Guided Data Augmentation and Regularization for Consistent Question Answerings"!
We show SOTA QA models produce inconsistent predictions and introduce logic-guided data augmentation & consistency-based regularization.
1/
2
47
265
0
1
8
Looking forward to working with you
I'm... graduating!
Last week (the pic): goodbye party from the amazing
@biunlp
lab and the one and only
@yoavgo
Now: In the airport, on my way to Seattle for my postdoc at AI2 in the
@ai2_allennlp
team and UW, where I'll be working with
@nlpnoah
@HannaHajishirzi
and
@sameer_
25
4
221
0
0
8
Check our these two great papers by Sewon today.
#acl2019nlp
Check out our two presentations on QA today! (1) 11:30 Hall 4 "Compositional Questions Do Not Necessitate Multi-hop Reasoning" (2) 4-5:40 Poster
#B
"Multi-hop Reading Comprehension through Question Decomposition and Rescoring" Come and say hi 😀
0
5
37
0
0
8
@gabriel_ilharco
@xai
@gabriel_ilharco
we already miss you at uw :) great work; I guess you’ve had a good start, convincing
@xai
releasing weights
0
0
8
Had a great time hosting
@FidlerSanja
today at
@uwnlp
. Learned a lot about her lab’s recent exciting work.
0
0
6
Due to the COVID-19 pandemic, AKBC 2020 is moving to a fully virtual conference including invited and contributed talks, poster sessions, workshops and hallway discussions. 1/2
1
14
36
0
0
7
@mark_riedl
@rajammanabrolu
@anmarasovic
@janetlauyeung
@tongshuangwu
Should we all focus our AGIs toward building an NSF bio? It seems every academics need to feel the AGI here 😳
1
0
5
@GaryMarcus
We have a few datasets relevant to Math word problem solving. Please check out: MathQA , MAWPS:
0
1
5
You can still submit your full paper to
@akbc_conf
on Frb 13th even if you missed the abstract deadline.
@ML_NLP
#NLProc
#nlproc
#knowledge
#ACL2020
#ICML2020
AKBC organizers are pleased to announce that we’re opening up the abstract deadline for paper submissions to February 13! 1/n
1
4
2
0
0
5
OLMo will provide an opportunity to do open, collaborative research on llms
1
0
5
0
0
4
@jefrankle
@allen_ai
@mechanicaldirk
@i_beltagy
@soldni
Thanks
@jefrankle
for all your help and support ❤️❤️
0
0
4
1
0
4
Congrats
@DanielKhashabi
on his new position :) cheers
Life update: Thrilled to announce that I will join Johns Hopkins University
@jhuclsp
@jhucompsci
@JohnsHopkins
as an assistant professor of computer science in the fall! This is an honor of a lifetime and I will do my best to rise to the occasion.
82
24
555
0
0
4
@sleepinyourhat
Our project had a case study that medical drs actually preferred using our system for search rather than their go-to PubMed search.
@amini_aida
@davidjwadden
Tom Hope
0
0
4
@danqi_chen
@JonathanBerant
Thanks Danqi for all your help throughout the conference! Glad that it worked out well :)
0
0
3
0
0
3
@rajammanabrolu
@UCSanDiego
@ucsd_cse
@MosaicML
It was a pleasure to work with you. Congratulations!
1
0
2
0
0
2
Hanna Hajishirzi Retweeted
The secret to aligning LMs to human preferences is reinforcement learning. But Why&How is it used? Announcing
💻RL4LMs: library to train any
@huggingface
LM w/ RL
👾GRUE: benchmark of 6 NLP tasks+rewards
📈NLPO: new RL alg 4 LMs
🌐
6
115
418
0
0
1
@lucyluwang
@dsweld
@etzioni
@_DougDowney
@SemanticScholar
@allen_ai
You are amazing Lucy. Best of luck at your next endeavor.
0
0
1
1
0
1
0
0
1