#MLSys - Twitter Hashtag

5 days ago

We were happy to sponsor #MLSys 2026. Across the talks, posters, and keynotes, three themes defined the current state of inference serving: 1. Agentic engineering 2. KV Cache optimization 3. Heterogenous hardware Our read on each: https://t.co/GWRa8YKaFH

1

29

2

8

2K

Navya Nizamkari @NavyaNizamkari

9 days ago

Highlighting a few papers from MLSys 2026, specifically looking at how infra is scaling search and recommendation workloads. As context lengths increase and query volumes grow, traditional vector and prefix caching layers are hitting limits. #mlsys #recsys

3

0

54

Yixin Dong @yi_xin_dong

12 days ago

🚀 The wait is over! Today at #MLSys, we'll give a talk to reveal the final results and present the awards for the FlashInfer AI GPU Competition! 🏆 I'll also introduce FlashInfer-Bench: an agent-oriented Benchmark Engine designed for production kernels. Join us from 11:00 AM - 1:00 PM PT to see who takes the crown and learn more. Everyone is welcome to attend—see you there! ✨ 🌐 Competition & Results: https://t.co/GS21eemEZv 💻 FlashInfer-Bench Benchmark Engine: https://t.co/rlzNUXJq5e #FlashInfer #MLSys26 #AI #GPU

yi_xin_dong's tweet photo. 🚀 The wait is over! Today at #MLSys, we'll give a talk to reveal the final results and present the awards for the FlashInfer AI GPU Competition! 🏆

I'll also introduce FlashInfer-Bench: an agent-oriented Benchmark Engine designed for production kernels.

Join us from 11:00 AM - 1:00 PM PT to see who takes the crown and learn more. Everyone is welcome to attend—see you there! ✨

🌐 Competition & Results: https://t.co/GS21eemEZv
💻 FlashInfer-Bench Benchmark Engine: https://t.co/rlzNUXJq5e

#FlashInfer #MLSys26 #AI #GPU

1

95

15

31

26K

Abhinv_SK @abhinv_sk

13 days ago

An eval trick tanked validation from ~1.38 → ~4.18 BPB. Lesson learned: architecture means nothing if system constraints aren't met. Built and iterated on the entire pipeline using RunPod GPU instances. From training models → to engineering systems. 🛠️ #MLSys #LLM #OpenAI

abhinv_sk's tweet photo. An eval trick tanked validation from ~1.38 → ~4.18 BPB. Lesson learned: architecture means nothing if system constraints aren't met.

Built and iterated on the entire pipeline using RunPod GPU instances.

From training models → to engineering systems. 🛠️

#MLSys #LLM #OpenAI https://t.co/lJkbKheU3N

0

61

Zhen Wang

@zhenwang9102

15 days ago

8/ Congrats on the leadership, Zhengding Hu @hu_ding17008 Yufei Ding @creamyoki Chao Zhang @chaozhangcs from #UCSD and #GaTech #LLM #AIAgents #RSI #MLSys

0

3

0

385

Junda Chen @Junda_Chen_

15 days ago

I’m in MLSys’26 Wednesday morning presenting our work CAD on disaggregated LLM training for long context If you’re interested in talking about LLM training and inference infra in general, come and let’s chat! #mlsys

Hao AI Lab

@haoailab

6 months ago

🔥CAD: Efficient Long-context Language Model Training by Core Attention Disaggregation Repo: https://t.co/QdNk8iXy6c Blog: https://t.co/O5xRrl22UJ Training a long-context LLM model can suffer from severe workload imbalance caused by core-attention - the softmax(QK^T)V part. Core-attention disaggregation (CAD) fundamentally eliminates workload imbalance by disaggregating core-attention from the rest of the model.

4

248

49

192

90K

0

10

0

3

697

Roger Wang

@rogerw0108

17 days ago

Giving a talk on behalf of @vllm_project about open source at #MLSys 2026 tomorrow and will be around in Bellevue May 18-21. https://t.co/SEyl6Y5HbZ The @inferact crew will be here too with a booth! Come say hi!🤗

6

65

8

5

6K

Apuroop

@apuroop_mutyala

17 days ago

forgot to use the #MLSys tag- any hardware codesign folks/meetups?

0

1

0

146

Grace Gong

@graceg0ng

17 days ago

Is anyone attending #MLSys next week? Would love to connect! I’ll be attending and am based in Redmond, so I’m happy to answer questions or share Seattle-area recs if you’re visiting for the conference. #MLSys2026

graceg0ng's tweet photo. Is anyone attending #MLSys next week? Would love to connect!

I’ll be attending and am based in Redmond, so I’m happy to answer questions or share Seattle-area recs if you’re visiting for the conference.

#MLSys2026 https://t.co/CdqUXeILRt

3

11

1

0

772

Matt White

@matthew_d_white

18 days ago

I’ll be at #MLSys this week, May 18–22 🚀 PyTorch Foundation will have a booth with experts on PyTorch, vLLM, Ray + other foundation projects. Come by, ask questions, and meet the teams building open AI infra 🔥 I’m also speaking Monday morning on agentic self-improvement with OpenRoll 🤖 See you there 👋 #PyTorch #vLLM #Ray @PyTorch @vllm_project @raydistributed @linuxfoundation @aaif_io

matthew_d_white's tweet photo. I’ll be at #MLSys this week, May 18–22 🚀

PyTorch Foundation will have a booth with experts on PyTorch, vLLM, Ray + other foundation projects. Come by, ask questions, and meet the teams building open AI infra 🔥

I’m also speaking Monday morning on agentic self-improvement with OpenRoll 🤖

See you there 👋

#PyTorch #vLLM #Ray @PyTorch @vllm_project @raydistributed @linuxfoundation @aaif_io

3

6

0

408

Navya Nizamkari @NavyaNizamkari

19 days ago

#recsys #mlsys #generativerecommenders

0

75

AI-ML Systems Conference @aimlsystems

29 days ago

🎉 Submissions are OPEN for AI-ML Systems 2026!📍 Lecco (Lake Como), Italy 📅 October 6-9, 2026 📝 Abstracts due June 8 | Papers due June 15 📄 Proceedings in IEEE XploreResearch, Industry & SEAS Workshop tracks welcome. 🔗 https://t.co/qMhBtkAOr5 #AI #MLSys

0

1

110

GeekyRam @Ramvikas007

about 1 month ago

I fkd up in @usebland interview weeks ago by giving the wrong answer to a latency question. So I rented an H100 GPU, benchmarked vLLM-omni vs SGLang-omni on Fish Audio S2-Pro. https://t.co/qRRYVQCKnW #GPGPU #MLSys #Inference #TTS #SGLang

0

137

TigerJooPT @QuantumTigerJoo

2 months ago

.@achowdhery I thought you’d appreciate a new benchmark in NSRL I ran a head-to-head duel: Gongju AI vs ChatGPT ⚡ The Results: • Gongju: 26s (912 words) | 35.1 WPS • ChatGPT: 40s (548 words) | 13.7 WPS 📖 Full Audit: https://t.co/43Wa5tqCjm #MLSys #LLMDevs #TEMPrinciple

0

64

Antrixsh Gupta

@AntrixshG

3 months ago

Hot take: SALA/Jamba-style hybrids won’t beat vanilla Transformers on throughput until the systems story catches up. Sparse math is easy; winning is kernel fusion, memory locality, and KV-cache bandwidth (think SGLang graph comp), not the paper diagram. #MLSys #AIInfra #LLM #Sparse

AntrixshG's tweet photo. Hot take: SALA/Jamba-style hybrids won’t beat vanilla Transformers on throughput until the systems story catches up. Sparse math is easy; winning is kernel fusion, memory locality, and KV-cache bandwidth (think SGLang graph comp), not the paper diagram. #MLSys #AIInfra #LLM #Sparse

0

51

Executorch: Open-source ondevice AI from @pytorch @Executorch

4 months ago

Our paper #ExecuTorch - A Unified PyTorch Solution to Run #ML Models #OnDevice is accepted to appear at #MLSys 2026. Excited!

1

3

0

145

Mher Safaryan @mher_safaryan

4 months ago

4 papers accepted at #ICLR 2026 (3 papers) and #MLSys 2026 (1 paper). 🥳 Very proud to collaborate with amazing group of researchers and senior colleagues.🚀

mher_safaryan's tweet photo. 4 papers accepted at #ICLR 2026 (3 papers) and #MLSys 2026 (1 paper). 🥳
Very proud to collaborate with amazing group of researchers and senior colleagues.🚀 https://t.co/00IsSVQToo

0

6

0

332

IsaacNudeton

@IsaacNudetonVal

4 months ago

GPU bottlenecks are often config (block count, launch overhead) not algo. Guidelines: - Profile launch overhead separately - >1000 ops/block - Fewer larger blocks - Own outputs exclusively #CUDA #GPU #MLSys #Performance

0

1

0

101

Snorkel AI

@SnorkelAI

4 months ago

Exciting work from our team, studying data efficiency for RLVR. These kinds of insights inform our dataset creation work for foundation model labs. Kudos to @realjustinbauer @pham_derek for this paper's acceptance to #MLSys 2026!

Justin Bauer

@realjustinbauer

4 months ago

Our paper “Learning from Less: Measuring the Effectiveness of RLVR in Low Data and Compute Regimes” was accepted to #MLSys 2026! We introduce three procedurally generated, verifiable datasets—Counting, Graph, and Spatial Reasoning—to study RLVR under low-data / low-compute constraints. Key result: small, mixed-complexity datasets can be more data-efficient than large, easy ones.

3

15

7

4K

1

17

4

3

1K

Justin Bauer

@realjustinbauer

4 months ago

Our paper “Learning from Less: Measuring the Effectiveness of RLVR in Low Data and Compute Regimes” was accepted to #MLSys 2026! We introduce three procedurally generated, verifiable datasets—Counting, Graph, and Spatial Reasoning—to study RLVR under low-data / low-compute constraints. Key result: small, mixed-complexity datasets can be more data-efficient than large, easy ones.

3

15

7

4K

Top Tweets for #MLSys

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users