Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs(baseten.co)

215 points by philipkiely 1 year ago | 142 comments

Web LLM runs the vicuna-7B LLM in the browser and it’s impressive(simonwillison.net)

30 points by simonw 3 years ago | 1 comments

Raspberry Pi's New AI Hat Adds 8GB of RAM for Local LLMs(jeffgeerling.com)

228 points by ingve 9 months ago | 180 comments

OpenAI's new open-source model is basically Phi-5(seangoedecke.com)

372 points by emschwartz 1 year ago | 200 comments

Sean Goedecke's blog post arguing that OpenAI's open-weight gpt-oss models resemble Microsoft's Phi series, strong on benchmarks but weak on out-of-domain knowledge.

SmolLM2(simonwillison.net)

86 points by edward 2 years ago | 20 comments

Show HN: A rudimentary game engine to build four dimensional VR evironments(brainpaingames.com)

30 points by teemur 1 year ago | 1 comments

GPUs Go Brrr(hazyresearch.stanford.edu)

1056 points by nmstoker 2 years ago | 243 comments

Stanford Hazy Research blog post on what GPU hardware wants for efficient AI kernels, introducing the ThunderKittens CUDA kernel library.

AWS Bedrock to require sharing data with Anthropic for Mythos and future models

369 points by TomAnthony 4 months ago | 216 comments

Hacker News discussion of Anthropic requiring 30-day data retention for Mythos-class Claude models on AWS Bedrock, moving data outside AWS's security boundary.

Building the heap: racking 30 petabytes of hard drives for pretraining(si.inc)

392 points by nee1r 1 year ago | 267 comments

Standard Intelligence blog post describing how they built a 30-petabyte hard drive storage cluster in a San Francisco colocation center for under $500k to store video pretraining data.

GPT-3.5 Turbo fine-tuning and API updates(openai.com)

363 points by davidbarker 3 years ago | 205 comments

OpenAI blog post announcing fine-tuning support for GPT-3.5 Turbo, letting developers customize the model with their own data, plus related API updates.

My finetuned models beat OpenAI's GPT-4(mlops.systems)

376 points by majc2 2 years ago | 86 comments

Blog post evaluating finetuned Mistral, Llama3 and Solar LLMs against OpenAI's GPT-4 on accuracy for structured data extraction from press releases.

I don't like NumPy(dynomight.net)

399 points by MinimalAction 1 year ago | 164 comments

A Dynomight blog essay critiquing NumPy's design, arguing its array broadcasting and indexing make complex computations confusing.