from Hacker News

limoce

joined 4/3/21, 1:02 PM has 861 karma

• FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

  by limoce on 7/11/25, 10:36 AM, with 107 comments

• Polaris: A Post-training recipe for scaling RL on Advanced Reasoning models

  by limoce on 7/9/25, 6:58 AM, with 1 comments

• Overclocking LLM Reasoning: Monitoring and Controlling LLM Thinking Path Lengths

  by limoce on 7/6/25, 12:53 PM, with 0 comments

• Neutrino: Probing-Based eBPF-Like GPU Kernel Profiling

  by limoce on 7/1/25, 10:33 AM, with 0 comments

• Machine Learning Conferences Should Establish "Refutations and Critiques" Track

  by limoce on 6/26/25, 10:27 AM, with 0 comments

• SuperGPQA: Scaling LLM Evaluation Across 285 Graduate Disciplines

  by limoce on 3/4/25, 7:26 AM, with 0 comments

• SepLLM: Accelerate LLMs by Compressing One Segment into One Separator

  by limoce on 3/3/25, 1:27 PM, with 2 comments

• Step-Video-T2V: The Practice, Challenges, and Future of Video Foundation Model

  by limoce on 2/17/25, 9:54 AM, with 5 comments

• Logic R1: Reproduce DeepSeek R1 Zero on 2K Logic Puzzle Dataset

  by limoce on 2/5/25, 4:02 AM, with 0 comments

• Libnginx: Nginx as a Shared Library

  by limoce on 2/4/25, 7:56 AM, with 0 comments

• DeepSeek-VL2: Moe Vision-Language Models for Advanced Multimodal Understanding [pdf]

  by limoce on 12/13/24, 12:53 PM, with 0 comments

• Fast vectorizable algorithms of binary searching for floating point numbers

  by limoce on 11/15/24, 12:53 AM, with 0 comments

• New OpenAI Feature: Predicted Outputs

  by limoce on 11/5/24, 2:47 AM, with 7 comments

• Collaborative Filtering Is Wrong and Here Is Why

  by limoce on 10/24/24, 9:03 AM, with 0 comments

• REST: A Plug-and-Play Method for Accelerating LLM Without Additional Training

  by limoce on 10/20/24, 6:13 AM, with 0 comments

• Smoke 'em if you got 'em: Hacker gains root access using cigarette lighter

  by limoce on 10/12/24, 1:20 PM, with 0 comments

• O1 Replication Journey: A Strategic Progress Report

  by limoce on 10/9/24, 8:09 AM, with 0 comments

• Failures of Gradient-Based Deep Learning (2017) [pdf]

  by limoce on 8/15/24, 10:19 AM, with 0 comments

• Qwen2-VL

  by limoce on 8/14/24, 8:20 AM, with 0 comments

• Qwen2-Math

  by limoce on 8/8/24, 3:00 PM, with 38 comments

• FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention

  by limoce on 8/8/24, 7:24 AM, with 24 comments

• MiniCPM-v2.6: GPT-4V Level MLLM for Single/Multi Image and Video on Your Phone

  by limoce on 8/7/24, 2:00 AM, with 0 comments

• MindSearch: LLM-Based Web Search Engine Similar to Perplexity.ai and SearchGPT

  by limoce on 8/1/24, 8:53 AM, with 0 comments

• Turbo Sparse: Achieving LLM SOTA Performance with Minimal Activated Parameters

  by limoce on 6/12/24, 11:01 AM, with 0 comments

• PowerInfer-2: Fast Large Language Model Inference on a Smartphone

  by limoce on 6/11/24, 2:19 PM, with 0 comments

• Large-scale photonic chiplet Taichi empowers 160TOPS/W AI

  by limoce on 4/12/24, 7:49 AM, with 0 comments

• Asterinas: OS kernel written in Rust and providing Linux-compatible ABI

  by limoce on 3/5/24, 8:52 AM, with 0 comments

• Mq-deadline scalability improvements (with more than 100% improvement)

  by limoce on 1/20/24, 12:04 PM, with 0 comments

• Researchers Create First Functional Semiconductor Made from Graphene

  by limoce on 1/5/24, 4:45 AM, with 0 comments

• Wayland Enjoyed Many Successes in 2023

  by limoce on 1/2/24, 3:54 AM, with 0 comments

• Improving our safety with a physical quantities and units library

  by limoce on 12/23/23, 6:01 AM, with 68 comments

• Nesting chinstrap penguins sleep by seconds-long microsleeps

  by limoce on 12/22/23, 2:31 PM, with 0 comments

• PowerInfer: High-Speed Large Language Model Serving on Consumer-Grade GPUs

  by limoce on 12/19/23, 12:19 PM, with 1 comments

• Zpoline: System Call Hook for Linux

  by limoce on 7/20/23, 4:18 AM, with 0 comments

• Randomized Single-Source Shortest Path Algo. On Undirected Real-Weighted Graphs

  by limoce on 7/11/23, 7:26 AM, with 0 comments

• ChatGPT powered Rust proc macro that generates code at compile-time

  by limoce on 3/13/23, 2:31 AM, with 0 comments

• Memcpy is faster than memset on Intel i7 12700 with glibc 2.36

  by limoce on 1/1/23, 12:50 PM, with 2 comments

• Year-in-search-trends: Visualization of search interest over time

  by limoce on 12/30/22, 1:28 PM, with 0 comments

• Make call_rcu() lazy to save power

  by limoce on 12/30/22, 12:58 PM, with 0 comments

• IOMMUFD

  by limoce on 12/30/22, 12:57 PM, with 0 comments

• Chinese Chipmaker Loongson Readies 3A6000 to Tackle Zen 3 and Tiger Lake

  by limoce on 10/18/22, 2:26 PM, with 3 comments

• Skyplane: Optimizing Transfer Cost and Throughput Using Cloud-Aware Overlays

  by limoce on 10/18/22, 9:39 AM, with 1 comments

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

  by limoce on 9/21/22, 2:32 PM, with 0 comments

• eRPC: A fast remote procedure call library for datacenters

  by limoce on 8/30/22, 7:01 AM, with 0 comments

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

  by limoce on 8/16/22, 3:03 PM, with 55 comments

• How hard is it to open a file (2016)

  by limoce on 8/14/22, 1:19 AM, with 0 comments