李枫_Explore+the+lightweight+deep+learning+framework—TinyGrad__FengLi-20260328p(1)_20260329_111921.pdf

1、Explore Explore the lightweight the lightweight deep deep learning frameworklearning frameworkTinyGradTinyGrad Feng Li(李枫李枫) Mar 28,2026 An indie developer from China：The main translator of the book Gray Hat Hacking The Ethical Hackers Handbook,Fourth Edition (ISBN：9787302428671)&Linux Hardening in

2、Hostile Networks,First Edition (ISBN:9787115544384)Pure software development for 15 years (11 years on Mobile dev)Actively participating Open Source Communities:https:/ https:/ Recently,focus on infrastructure of Cloud/Edge Computing,AI,IoT,OS,Programming Languages&Runtimes,Network,Virtualization,RI

3、SC-V,EDA,5G/6G Who Am I Agenda I.Background An overview of TinyGrad Testbeds II.REMU Overview Practices III.Speeding up TinyGrad PustPython Monty Mojo IV.Reviving the support for TinyGrad in Exo Whats the problem?Our exploration V.Wrap-up 1)An Overview of TinyGrad 1.1 What is it I.Background https:/

4、tinygrad.org/https:/ https:/olivares.cl/blog/2025/06/10/introduction-to-tinygrad/Comparison.I.Background Source:https:/ Accelerators/Runtimes https:/docs.tinygrad.org/runtime/I.Background Tinybox https:/tinygrad.org/I.Background 1.2 Src I.Background https:/ .I.Background Install from source.On Testb

5、ed1:I.Background 1.3 Architecture and design.I.Background Source:https:/docs.tinygrad.org/Source:https:/ I.Background Source:https:/ 1.4 How it works I.Background.Source:https:/www.signalstack.news/blog/tinybox-unleashing-affordable-deep-learning-power-with-tinygrad/Source:https:/ I.Background Sourc

6、e:https:/ I.Background Source:https:/ I.Background Source:https:/ 2)Testbeds 2.1 Testbed1(GMK K8 PLUS)I.Background https:/ Our configs 96GB DDR5-5600 with Fedora 43(Kernel 6.19.9)RDNA https:/en.wikipedia.org/wiki/RDNA_(microarchitecture)I.Background 1)Overview https:/ RDNA3 emulator https:/ II.REMU