Hacker News: hyp0thetical

New comment by hyp0thetical in "How We Made IPFS Content Publishing 10x Faster"

hyp0thetical — Thu, 02 Jul 2026 03:48:05 +0000

Did you ever try to keep files on a really good storage?

New comment by hyp0thetical in "How We Made IPFS Content Publishing 10x Faster"

hyp0thetical — Thu, 02 Jul 2026 03:47:34 +0000

Please optimise Pentium 2 for us!!

New comment by hyp0thetical in "[dead]"

hyp0thetical — Mon, 03 Nov 2025 18:58:47 +0000

We've released "TheWhisper," a project that modernizes Whisper models and makes them affordable for streaming real-time transcription. The project includes an Electron macOS app with a deployment example and optimized engines for NVIDIA GPUs. We will continue to improve it! Love to hear your feedback and features requests.

Serving accelerated FLUX models on Modal

hyp0thetical — Tue, 07 Oct 2025 22:40:51 +0000

Article URL: https://docs.thestage.ai/tutorials/source/modal_thestage.html

Comments URL: https://news.ycombinator.com/item?id=45509843

Points: 1

# Comments: 0

New comment by hyp0thetical in "Comprehensive Guide to Text-to-Image Quality Metrics"

hyp0thetical — Thu, 04 Sep 2025 11:19:56 +0000

Our research team put together a comprehensive guide to check perceptual quality, sharpness, color, prompt alignment, and more.

All the tricky image quality questions researchers usually ask are covered! You are welcome to ask questions and request additional metrics!

Comprehensive Guide to Text-to-Image Quality Metrics

hyp0thetical — Thu, 04 Sep 2025 11:19:56 +0000

Article URL: https://docs.thestage.ai/tutorials/source/text2image_evaluation_tutorial.html

Comments URL: https://news.ycombinator.com/item?id=45125983

Points: 5

# Comments: 1

New comment by hyp0thetical in "Can LLMs recognise ASCII images?"

hyp0thetical — Fri, 15 Aug 2025 19:36:04 +0000

Our tests with accelerated Elastic Models show they factorize tasks by identifying features line-by-line, then combine them using statistical patterns common in ASCII art.

If we think about it, attention scores will look like a heatmap of the original image. So, transformers have an internal representation of the image inside, and if they can recognize images by default, it's some sort of image classifier which provides token ID as an image class. The tests are, to be honest, trivial, but anyway funny :)

Try Deepseek-Qwen-14B in our tutorial - running at 120 tok/s on H100 and 40 tok/s on L40s, up to 3x faster than the original implementation! Fully free, get your API token and start!

Can LLMs recognise ASCII images?

hyp0thetical — Fri, 15 Aug 2025 19:36:04 +0000

Article URL: https://docs.thestage.ai/tutorials/source/elastic_transformers.html

Comments URL: https://news.ycombinator.com/item?id=44916552

Points: 5

# Comments: 1

New comment by hyp0thetical in "Learn basics of quantization with beta access to TheStage AI's QLIP framework"

hyp0thetical — Tue, 12 Aug 2025 20:31:54 +0000

We're happy to provide AI research and engineering teams access to our PyTorch framework for applying quantization algorithms that are tightly aligned with NVIDIA's compiler for efficient inference. The framework also enables convenient research of new algorithms!

We hold several patents and have published articles, including a CVPR oral presentation on DNNs compression. We've built tools we enjoy using ourselves and hope they can benefit others too!

The framework supports various quantization setups:

- Integer and Float quantization

- Symmetric and Asymmetric quantization

- Dynamic and Static quantization

- Multiple granularity options: per-tensor, per-channel, per-token, etc

- Pre-defined configuration schemas compatible with NVIDIA GPUs that are easy to set up and use

Learn basics of quantization with beta access to TheStage AI's QLIP framework

hyp0thetical — Tue, 12 Aug 2025 20:31:54 +0000

Article URL: https://docs.thestage.ai/tutorials/source/quantization_tutorial.html

Comments URL: https://news.ycombinator.com/item?id=44881471

Points: 3

# Comments: 1