TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

IEEE FP8 Formats for Machine Learning (Draft) [pdf]

13 pointsby avianes12 months ago

1 comment

hi-v-rocknroll12 months ago
Formats, plural.<p>sign:explicit leading mantissa bit:mantissa bits:exponent bits<p>1:0:(P-1):(8-P) where P ∈ [1,7]<p>P ∈ [3,5] appears to be more useful<p>binary8p3 -&gt; 1:0:2:5<p>binary8p4 -&gt; 1:0:3:4<p>binary8p5 -&gt; 1:0:4:3<p>Edge cases:<p>P = 8 would disallow all exponent bits, leaving a sign bit and 7 explicit mantissa bits of precision<p>P = 0 would be unsigned and 8 exponent bits with only an implicit precision bit but no explicit mantissa bits
评论 #40539685 未加载