Beidi Chen
Assistant Professor
Designs efficient algorithms and systems for training and serving large language models, including sparsity and long-context inference.
Beidi Chen's lab at CMU works on efficient training & inference and large language models. 54 papers since 2023, 51 at top venues, mostly NeurIPS and ICML. 5 PhD students. 13 alumni, most often to ByteDance and Microsoft. Applications close .
At a glance
- papers, 2023–2026
- 14/ year
- at T1 venues
- 94%
- PhD students
- 5
- MS students
- 0
- alumni listed
- 13
Papers by yearfrom the lab homepage, 2023–2026
- 7 papers2026 · T1 6 · collecting
- 18 papers2025 · T1 18
- 22 papers2024 · T1 20
- 7 papers2023 · T1 7
T1 venuesOtherVenues: NeurIPS, ICML, ICLR, ACL, ASPLOS
Research focuspaper titles that use each topic's words
- Large language models29 papers
- Efficient training & inference27 papers
- long-context1 paper
- serving large0 papers
- designs efficient0 papers
Paper list
From the lab homepage.
20267 papers
- NeurIPS (T1)AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs ↗
- EMNLP FindingsBeyond Confidence: Test-Time Scaling for Multi-Turn Search Agents via Retrieval Grounding ↗
- EMNLP (T1)CamoDocs: A Poisoning Attack Against Retrieval-Augmented Language Models Using Camouflaged Documents ↗
- NeurIPS (T1)MonarchRT: Efficient Attention for Real-Time Video Generation ↗
- ICLR (T1)STEM: Scaling Transformers with Embedding Modules ↗
- ICML (T1)ThunderAgent: A Simple, Fast and Program-Aware Agentic Inference System ↗
- NeurIPS (T1)Vortex: Efficient and Programmable Sparse Attention Serving for AI Agents ↗
202518 papers
- NeurIPS (T1)Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts ↗
- ICLR (T1)APE: Faster and Longer Context-Augmented Generation via Adaptive Parallel Encoding ↗
- ICLR (T1)FSA: An Alternative Efficient Implementation of Native Sparse Attention Kernel ↗
- ICML (T1)GSM-Infinite: How Do Your LLMs Behave over Infinitely Increasing Context Length and Reasoning Complexity? ↗
- ICLR (T1)Jackpot: Optimal Budgeted Rejection Sampling for Extreme Actor–Policy Mismatch RL ↗
- NeurIPS (T1)Kinetics: Rethinking Test-Time Scaling Laws ↗
- MICRO (T1)LLM.265: Video Codecs are Secretly Tensor Codecs ↗
- ICLR (T1)MagicDec: Breaking the Latency-Throughput Tradeoff for Long Context Generation with Speculative Decoding ↗
- ICLR (T1)MagicPIG: LSH Sampling for Efficient LLM Generation ↗
- ICLR (T1)Memory Mosaics ↗
- NeurIPS (T1)Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generation ↗
- ICLR (T1)Prosperity before Collapse: How Far Can Off-Policy RL Reach with Stale Data on LLMs? ↗
202422 papers
- ICML (T1)Compress, Then Prompt: Improving Accuracy-Efficiency Trade-Off of LLM Inference with Transferable Prompt ↗
- ICLR (T1)Efficient Streaming Language Models with Attention Sinks ↗
- NeurIPS (T1)Found in the Middle: How Language Models Use Long Contexts Better via Plug-and-Play Positional Encoding ↗
- ICML (T1)Galore: Memory-Efficient LLM Training by Gradient Low-Rank Projection ↗
- ICML (T1)Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference ↗
- ICML (T1)HexGen: Generative Inference of Large Language Model over Heterogeneous Environment ↗
- ICLR (T1)Joma: Demystifying Multilayer Transformers via Joint Dynamics of MLP and Attention ↗
- ICML (T1)KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache ↗
- ACL (T1)Layer Skip: Enabling Early Exit Inference and Self-Speculative Decoding ↗
- NeurIPS (T1)Learn To be Efficient: Build Structured Sparsity in Large Language Models ↗
- ICML (T1)LoCoCo: Dropping In Convolutions for Long Context Compression ↗
- NeurIPS (T1)Megalodon: Efficient LLM Pretraining and Inference with Unlimited Context Length ↗
20237 papers
- ICML (T1)CocktailSGD: Fine-tuning Foundation Models over 500Mbps Networks ↗
- ICML (T1)Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time ↗
- COLT (T1)Fast Algorithms for a New Relaxation of Optimal Transport ↗
- ICML (T1)FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU ↗
- NeurIPS (T1)H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models ↗
- NeurIPS (T1)Laughing Hyena Distillery: Extracting Compact Recurrences from Convolutions ↗
- NeurIPS (T1)Scan and Snap: Understanding Training Dynamics and Token Composition in 1-layer Transformer ↗
Where alumni went13 listed, 15% to industry
Industry 2Academia 11
Most often to
- ByteDance
- Microsoft
Also to
- Carnegie Mellon University
- Princeton University
- PhD at Rice University
Apply to CMUFall 2027
CMU SCS Computer Science PhD
- Round
- Single round
- Opens
- GRE
- See program page
- English
- TOEFL / IELTS / Duolingo (F-1/J-1 non-native speakers)
- Fee
- USD 100 (USD 80 by early deadline)
Confirmed on the program page.
Similar labsnearest by research summary
- USNYU
Panos Ipeirotis
- Time series, anomaly & spatio-temporal data
- CNZhejiang
Wangze Ni
- Efficient training & inference
- USUMD
Cody Buntain
- Computational social science
- CNTsinghua
Guoliang Li
- Time series, anomaly & spatio-temporal data
- USGeorgia Tech
Haesun Park
- Visualization & visual analytics
- USUC Berkeley
Serina Chang
- Human-AI collaboration & trust
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
CMUcloses