Hacker News: galeos

New comment by galeos in "AMD acquires Taalas to boost inference performance by etching models in silicon"

galeos — Fri, 07 Aug 2026 13:04:30 +0000

Is there scope to implement ternary models using this approach to minimise die area of the model parameters?

1.58bit LLM Optimised Tensor Core

galeos — Tue, 01 Jul 2025 18:24:48 +0000

Article URL: https://github.com/microsoft/T-MAC

Comments URL: https://news.ycombinator.com/item?id=44436729

Points: 2

# Comments: 0

BitNet 1.58bit GPU Inference Kernel

galeos — Tue, 20 May 2025 16:55:50 +0000

Article URL: https://github.com/microsoft/BitNet/blob/main/gpu/README.md

Comments URL: https://news.ycombinator.com/item?id=44043600

Points: 2

# Comments: 0

New comment by galeos in "Everything we announced at our first LlamaCon"

galeos — Tue, 29 Apr 2025 19:34:04 +0000

How did you find ModernBERT performance Vs prior BERT models?

Microsoft beat H200 Deepseek inference with MI300

galeos — Thu, 24 Apr 2025 20:06:01 +0000

Article URL: https://techcommunity.microsoft.com/blog/MachineLearningBlog/accelerating-deepseek-inference-with-amd-mi300-a-collaborative-breakthrough/4407673

Comments URL: https://news.ycombinator.com/item?id=43786945

Points: 3

# Comments: 1

Modular's CUDA alternative is ready

galeos — Wed, 23 Apr 2025 19:26:58 +0000

Article URL: https://www.eetimes.com/after-three-years-modulars-cuda-alternative-is-ready/

Comments URL: https://news.ycombinator.com/item?id=43775779

Points: 2

# Comments: 0

New comment by galeos in "BitNet b1.58 2B4T Technical Report"

galeos — Thu, 17 Apr 2025 11:10:03 +0000

You can try out the model in a demo they have setup: https://bitnet-demo.azurewebsites.net/

BitNet b1.58 2B4T Technical Report

galeos — Thu, 17 Apr 2025 07:27:11 +0000

Article URL: https://arxiv.org/abs/2504.12285

Comments URL: https://news.ycombinator.com/item?id=43714004

Points: 111

# Comments: 30

Microsoft BitNet 1.58bit LLM 2B4T released

galeos — Wed, 16 Apr 2025 11:54:45 +0000

Article URL: https://huggingface.co/microsoft/bitnet-b1.58-2B-4T

Comments URL: https://news.ycombinator.com/item?id=43704219

Points: 2

# Comments: 0

Mi300 Huggingface

galeos — Wed, 16 Apr 2025 11:24:16 +0000

Article URL: https://huggingface.co/blog/huggingface-amd-mi300

Comments URL: https://news.ycombinator.com/item?id=43703985

Points: 2

# Comments: 0

Bitnet.cpp: Efficient Inference for 1.58bit LLMs

galeos — Thu, 27 Feb 2025 16:55:32 +0000

Article URL: https://arxiv.org/abs/2502.11880

Comments URL: https://news.ycombinator.com/item?id=43196126

Points: 1

# Comments: 0

Matryoshka Quantization

galeos — Sun, 16 Feb 2025 16:04:06 +0000

Article URL: https://arxiv.org/abs/2502.06786

Comments URL: https://news.ycombinator.com/item?id=43069048

Points: 2

# Comments: 0

1-Bit AI Infrastructure

galeos — Fri, 15 Nov 2024 14:28:23 +0000

Article URL: https://arxiv.org/abs/2410.16144

Comments URL: https://news.ycombinator.com/item?id=42147252

Points: 157

# Comments: 30

New comment by galeos in "BERTs Are Generative In-Context Learners"

galeos — Thu, 14 Nov 2024 12:35:09 +0000

My understanding is that BERT can still outperform LLMs for sentiment classification?

Microsoft BitNet: inference framework for 1-bit LLMs

galeos — Fri, 18 Oct 2024 09:10:36 +0000

Article URL: https://github.com/microsoft/BitNet

Comments URL: https://news.ycombinator.com/item?id=41877609

Points: 173

# Comments: 33

Apollo to offer Intel multibillion-dollar investment

galeos — Sun, 22 Sep 2024 21:14:42 +0000

Article URL: https://www.bloomberg.com/news/articles/2024-09-22/apollo-is-said-to-offer-multibillion-dollar-investment-in-intel

Comments URL: https://news.ycombinator.com/item?id=41620137

Points: 5

# Comments: 3

Fine-Tuning LLMs to 1.58bit

galeos — Wed, 18 Sep 2024 15:33:23 +0000

Article URL: https://huggingface.co/blog/1_58_llm_extreme_quantization

Comments URL: https://news.ycombinator.com/item?id=41581068

Points: 52

# Comments: 3

Evidence of dark oxygen production at the abyssal seafloor

galeos — Tue, 23 Jul 2024 23:04:26 +0000

Article URL: https://www.nature.com/articles/s41561-024-01480-8

Comments URL: https://news.ycombinator.com/item?id=41051941

Points: 9

# Comments: 0

Lamini Memory Tuning: 10x Fewer Hallucinations

galeos — Thu, 13 Jun 2024 22:29:40 +0000

Article URL: https://www.lamini.ai/blog/lamini-memory-tuning

Comments URL: https://news.ycombinator.com/item?id=40675577

Points: 128

# Comments: 57

New comment by galeos in "Nvidia Conquers Latest AI Tests"

galeos — Thu, 13 Jun 2024 15:48:03 +0000

These are MLPerf training results. I think current ternary quantization research is focused more on speeding up inference?