HackerTrans
TopNewTrendsCommentsPastAskShowJobs

somnial

7 karmajoined قبل 9 أشهر

Submissions

The Anatomy of an Instruction Pipeline Hazard

hiraditya.github.io
5 points·by somnial·قبل 3 ساعات·0 comments

Width vs. Depth: Speculating on the Margin

blog.doubleword.ai
17 points·by somnial·قبل 12 يومًا·1 comments

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

fergusfinn.com
2 points·by somnial·قبل شهرين·0 comments

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

fergusfinn.com
2 points·by somnial·قبل شهرين·0 comments

[untitled]

1 points·by somnial·قبل شهرين·0 comments

70x faster cold(ish) starts for SGLang

fergusfinn.com
1 points·by somnial·قبل 3 أشهر·0 comments

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

fergusfinn.com
1 points·by somnial·قبل 6 أشهر·0 comments

Parallel Primitives for Multi-Agent Workflows

fergusfinn.com
1 points·by somnial·قبل 6 أشهر·0 comments

Scheduling in LLM Inference

fergusfinn.com
1 points·by somnial·قبل 8 أشهر·0 comments

How fast can an LLM go?

fergusfinn.com
2 points·by somnial·قبل 9 أشهر·0 comments

comments

somnial
·قبل 17 يومًا·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
·الشهر الماضي·discuss
https://fergusfinn.com/blog/economics-of-speculative-decodin...

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