HackerTrans
TopNewTrendsCommentsPastAskShowJobs

OsamaJaber

229 karmajoined 9 месяцев назад

Submissions

Compiler for LLMs, world models, and AGI

arxiv.org
2 points·by OsamaJaber·4 дня назад·0 comments

Auto: The AGI Compiler

github.com
4 points·by OsamaJaber·5 дней назад·2 comments

Build You Own Model

runinfra.ai
2 points·by OsamaJaber·9 дней назад·0 comments

RunInfra: Optimize any open model down to the kernel, deploy in 5 min

runinfra.ai
2 points·by OsamaJaber·12 дней назад·0 comments

Compiles any HuggingFace model into a single persistent megakernel

twitter.com
2 points·by OsamaJaber·25 дней назад·0 comments

Mega Kernels, Written by Agents

arxiv.org
2 points·by OsamaJaber·в прошлом месяце·0 comments

AutoMegaKernel: Compiling a LLM into a single CUDA kernel

arxiv.org
3 points·by OsamaJaber·в прошлом месяце·0 comments

AutoMegaKernel: Compile an LLM into one provably-correct CUDA megakernel

github.com
4 points·by OsamaJaber·в прошлом месяце·0 comments

StreamIndex: Memory-bounded compressed sparse attention via streaming top-k

arxiv.org
4 points·by OsamaJaber·2 месяца назад·0 comments

Show HN: AutoKernel, Auto GPU Kernel Optimization

arxiv.org
2 points·by OsamaJaber·2 месяца назад·0 comments

DeepSeek V4's indexer dies at 65K. We got it to 1M on 6GB

arxiv.org
5 points·by OsamaJaber·2 месяца назад·0 comments

AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Search

arxiv.org
4 points·by OsamaJaber·2 месяца назад·0 comments

DeepSeek V4's indexer OOMs at 65K context. We got it to 1M in 6G

arxiv.org
8 points·by OsamaJaber·2 месяца назад·0 comments

Ouroboros: Dynamic Weight Generation for Recursive Transformers

arxiv.org
2 points·by OsamaJaber·3 месяца назад·0 comments

Tide: Token-Informed Depth Execution for Per-Token Early Exit in LLM Inference

arxiv.org
3 points·by OsamaJaber·3 месяца назад·1 comments

Own your AI. Optimized down to the kernel

runinfra.ai
1 points·by OsamaJaber·3 месяца назад·0 comments

[untitled]

1 points·by OsamaJaber·4 месяца назад·0 comments

[untitled]

1 points·by OsamaJaber·4 месяца назад·0 comments

Agents with "Hands"

openfang.sh
7 points·by OsamaJaber·5 месяцев назад·18 comments

Open-Source Agent Operating System

github.com
11 points·by OsamaJaber·5 месяцев назад·3 comments

comments

OsamaJaber
·5 дней назад·discuss
thread: https://x.com/Akashi203/status/2074495867449434389
OsamaJaber
·2 месяца назад·discuss
the bottleneck isn't generation, it's verification
OsamaJaber
·3 месяца назад·discuss
Good to see this exist. Inference providers quietly swap quant levels. Most users never check. A standard verifier from the model maker is the right move, would love to see other labs ship the same
OsamaJaber
·3 месяца назад·discuss
The modified MIT clause is sneakier than people think. Hit 100M users or $20M a month and you have to slap "Kimi K2.6" on your UI. That covers any consumer app worth building. Not really open, more like free until you matter. Llama pulled the same move
OsamaJaber
·3 месяца назад·discuss
That's why infra needs stricter internal walls than normal SaaS
OsamaJaber
·3 месяца назад·discuss
Small models in the browser are a different optimization problem than small models on a server. On server you chase throughput so you batch. In browser you're stuck at batch size 1, which means kernel launch overhead and memory bandwidth dominate, not FLOPs
OsamaJaber
·5 месяцев назад·discuss
Appreciate it :)
OsamaJaber
·5 месяцев назад·discuss
Keep talking like that and people will colonize space
OsamaJaber
·5 месяцев назад·discuss
MCP support is the real story here Means you're not locked into Claude or Codex Can plug in whatever agent you want
OsamaJaber
·5 месяцев назад·discuss
Nice. The config fragmentation across tools is a real annoyance Does it handle conflicts if tools expect different formats for the same setting?
OsamaJaber
·5 месяцев назад·discuss
The requirements gap point is underrated. AI guesses where a human would ask By the time you catch it in review, you've already wasted the time you saved -_-
OsamaJaber
·5 месяцев назад·discuss
Surprised it took this long for someone to write it up properly
OsamaJaber
·5 месяцев назад·discuss
250 C files were deleted. 2032 to go. Watching Zig slowly eat libc from the inside is one of the more satisfying long term projects to follow
OsamaJaber
·5 месяцев назад·discuss
30+ years maintaining one of the most critical pieces of infrastructure on nearly every Linux and Unix system, and he's currently looking for a sponsor to fund continued development. Every company running sudo in production owes this man. Someone should fix that
OsamaJaber
·5 месяцев назад·discuss
Great job! This is the kind of project that should exist for every complex system Systems like vLLM's codebase are massive and hard to follow Would love to see the same approach for other infra (a nano-Kubernetes, nano Postgres.....
OsamaJaber
·5 месяцев назад·discuss
Sooooo underrated Incredibly handy when you're away from your machine
OsamaJaber
·5 месяцев назад·discuss
The real issue was never AI in Windows It was AI with no clear user benefit. A Copilot button in Notepad doesn't solve a problem anyone has Good to see them pulling back, but the test will be whether the features they keep actually earn their place in the workflow instead of just being there because someone had a KPI to hit
OsamaJaber
·5 месяцев назад·discuss
We did something similar with TypeScript strict mode Turned it on per file with a ratchet count, and over a few months, the whole codebase was strict without ever blocking anyone
OsamaJaber
·5 месяцев назад·discuss
Finally Debugging slow queries without seeing what's happening inside the plan is just guessing
OsamaJaber
·5 месяцев назад·discuss
I didn't get it (Cat felling high and looking around meme)