HackerTrans
TopNewTrendsCommentsPastAskShowJobs

somnial

7 karmajoined 9 miesięcy temu

Submissions

The Anatomy of an Instruction Pipeline Hazard

hiraditya.github.io
5 points·by somnial·3 godziny temu·0 comments

Width vs. Depth: Speculating on the Margin

blog.doubleword.ai
17 points·by somnial·12 dni temu·1 comments

Pushing memory bound CUDA kernels past the speed of light with data compression

fergusfinn.com
2 points·by somnial·2 miesiące temu·0 comments

Speculative KV coding: ~4× losslessly compressed KV cache using a small model

fergusfinn.com
2 points·by somnial·2 miesiące temu·0 comments

[untitled]

1 points·by somnial·2 miesiące temu·0 comments

70x faster cold(ish) starts for SGLang

fergusfinn.com
1 points·by somnial·3 miesiące temu·0 comments

LLM powered data structures: A lock-free binary search tree

fergusfinn.com
1 points·by somnial·6 miesięcy temu·0 comments

Parallel Primitives for Multi-Agent Workflows

fergusfinn.com
1 points·by somnial·6 miesięcy temu·0 comments

Scheduling in LLM Inference

fergusfinn.com
1 points·by somnial·8 miesięcy temu·0 comments

How fast can an LLM go?

fergusfinn.com
2 points·by somnial·9 miesięcy temu·0 comments

comments

somnial
·17 dni temu·discuss
this is a blog post from a company that hosts open weights LLMs (https://www.doubleword.ai/). I think its possible it might have been tongue in cheek
somnial
·w zeszłym miesiącu·discuss
https://fergusfinn.com/blog/economics-of-speculative-decodin...

good point tho - plus for Deepseek the shared expert increases the overlap slightly
somnial
·w zeszłym miesiącu·discuss
true, but no reason the predictor model couldn't use linear attention (i.e. mamba, GDN etc) to predict KV caches