Hacker News headlines

103

FP8 is ~100 tflops faster when the kernel name has "cutlass" in it (twitter.com/cis_female)

18 hours ago | limoce | twitter.com | best

5

Polaris: A Post-training recipe for scaling RL on Advanced Reasoning models (hkunlp.github.io)

2 days ago | limoce | github.io | newest

14

Overclocking LLM Reasoning: Monitoring and Controlling LLM Thinking Path Lengths (royeisen.github.io)

5 days ago | limoce | github.io | frontpage

1

Neutrino: Probing-Based eBPF-Like GPU Kernel Profiling (github.com/open-neutrino)

a week ago | limoce | github.com | newest

3

Machine Learning Conferences Should Establish "Refutations and Critiques" Track (arxiv.org)

2 weeks ago | limoce | arxiv.org | newest

1

SuperGPQA: Scaling LLM Evaluation Across 285 Graduate Disciplines (supergpqa.github.io)

4 months ago | limoce | github.io | newest

4

SepLLM: Accelerate LLMs by Compressing One Segment into One Separator (sepllm.github.io)

4 months ago | limoce | github.io | frontpage

10

Step-Video-T2V: The Practice, Challenges, and Future of Video Foundation Model (arxiv.org)

5 months ago | limoce | arxiv.org | frontpage

1

Logic R1: Reproduce DeepSeek R1 Zero on 2K Logic Puzzle Dataset (github.com/unakar)

5 months ago | limoce | github.com | newest

1

Libnginx: Nginx as a Shared Library (github.com/wkgcass)

5 months ago | limoce | github.com | newest

1

DeepSeek-VL2: Moe Vision-Language Models for Advanced Multimodal Understanding [pdf] (github.com/deepseek-ai)

7 months ago | limoce | github.com | newest

1

Fast vectorizable algorithms of binary searching for floating point numbers (github.com/fabiocannizzo)

8 months ago | limoce | github.com | newest

26

New OpenAI Feature: Predicted Outputs (simonwillison.net)

8 months ago | limoce | simonwillison.net | frontpage

2

Collaborative Filtering Is Wrong and Here Is Why (springer.com)

9 months ago | limoce | springer.com | newest

1

REST: A Plug-and-Play Method for Accelerating LLM Without Additional Training (sites.google.com)

9 months ago | limoce | google.com | newest

1

Smoke 'em if you got 'em: Hacker gains root access using cigarette lighter (tomshardware.com)

9 months ago | limoce | tomshardware.com | newest

1

O1 Replication Journey: A Strategic Progress Report (github.com/gair-nlp)

9 months ago | limoce | github.com | newest

1

Failures of Gradient-Based Deep Learning (2017) [pdf] (mlr.press)

11 months ago | limoce | mlr.press | newest

1

Qwen2-VL (huggingface.co)

11 months ago | limoce | huggingface.co | newest

64

Qwen2-Math (qwenlm.github.io)

11 months ago | limoce | github.io | best

56

FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention (pytorch.org)

11 months ago | limoce | pytorch.org | best

1

MiniCPM-v2.6: GPT-4V Level MLLM for Single/Multi Image and Video on Your Phone (github.com/openbmb)

11 months ago | limoce | github.com | newest

1

MindSearch: LLM-Based Web Search Engine Similar to Perplexity.ai and SearchGPT (github.com/internlm)

11 months ago | limoce | github.com | newest

1

Header-only and powerful C++20 static reflection library in 99 lines (github.com/ubpa)

a year ago | limoce | github.com | newest

2

Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters (arxiv.org)

a year ago | limoce | arxiv.org | frontpage

1

PowerInfer-2: Fast Large Language Model Inference on a Smartphone (arxiv.org)

a year ago | limoce | arxiv.org | newest

2

Large-scale photonic chiplet Taichi empowers 160TOPS/W AI (science.org)

a year ago | limoce | science.org | newest

1

Asterinas: OS kernel written in Rust and providing Linux-compatible ABI (github.com/asterinas)

a year ago | limoce | github.com | newest

1

Mq-deadline scalability improvements (with more than 100% improvement) (kernel.org)

a year ago | limoce | kernel.org | newest

2

Researchers Create First Functional Semiconductor Made from Graphene (technologynetworks.com)

a year ago | limoce | technologynetworks.com | newest

1

Wayland Enjoyed Many Successes in 2023 (phoronix.com)

a year ago | limoce | phoronix.com | newest

15

Improving our safety with a physical quantities and units library (open-std.org)

a year ago | limoce | open-std.org | best

1

Nesting chinstrap penguins sleep by seconds-long microsleeps (science.org)

a year ago | limoce | science.org | newest

2

PowerInfer: High-Speed Large Language Model Serving on Consumer-Grade GPUs (github.com/sjtu-ipads)

a year ago | limoce | github.com | newest

1

Zpoline: System Call Hook for Linux (github.com/yasukata)

a year ago | limoce | github.com | newest

1

Randomized Single-Source Shortest Path Algo. On Undirected Real-Weighted Graphs (arxiv.org)

2 years ago | limoce | arxiv.org | newest

1

ChatGPT powered Rust proc macro that generates code at compile-time (github.com/retrage)

2 years ago | limoce | github.com | newest

3

Memcpy is faster than memset on Intel i7 12700 with glibc 2.36 (gist.github.com)

2 years ago | limoce | github.com | frontpage

1

Year-in-search-trends: Visualization of search interest over time (github.com/joweich)

2 years ago | limoce | github.com | newest

1

Make call_rcu() lazy to save power (kernel.org)

2 years ago | limoce | kernel.org | newest

2

IOMMUFD (kernel.org)

2 years ago | limoce | kernel.org | newest

4

Chinese Chipmaker Loongson Readies 3A6000 to Tackle Zen 3 and Tiger Lake (wccftech.com)

2 years ago | limoce | wccftech.com | frontpage

3

Skyplane: Optimizing Transfer Cost and Throughput Using Cloud-Aware Overlays (arxiv.org)

2 years ago | limoce | arxiv.org | frontpage

1

Slirp Is Dead, Long Live Slirp: a New Approach to User-Mode Networking

2 years ago | limoce | sched.com | newest

1

eRPC: A fast remote procedure call library for datacenters

2 years ago | limoce | erpc.io | newest

79

Poly-time algorithm for deciding Hilbert Nullstellensatz. A proof of P=NP

2 years ago | limoce | arxiv.org | frontpage

2

How hard is it to open a file (2016)

2 years ago | limoce | criu.org | newest

3

Kernel TLS 1.3 Rx improvements in Linux 5.20

2 years ago | limoce | kernel.org | newest

2

Resin: Holistic Service for Dealing with Memory Leaks in Production Cloud Infra

2 years ago | limoce | microsoft.com | newest

2

YMTC launched 232-layer flash memory chip: performance improved by 50%

2 years ago | limoce | inf.news | frontpage