TensorFlow¶

	TensorFlow	TensorFlow Lite	TensorFlow Lite Micro	TensorFlow.js	TensorFlow Serving
Usage	Model development	Deployment to Microprocessors (laptops, mobile phones, RaspberryPi)	Deployment to Microcontrollers	Deployment to Web	Cloud, On-Prem
Optimized for	X86 TPU GPU	X86 ARM Cortex A	ARM Cortex M DSP MCU	Browser & Node
User	ML researcher	Application developer
Supports training	✅	❌	❌	❌	❌
Supports inference	✅ (inefficient)	✅	✅	✅	✅
No of ops supported	~1400	~130	~50
Native quantization support	❌	✅	✅
OS-independent (No OS required)	❌	❌	✅
Memory mapping of models	❌	✅	✅
Delegation to accelerators	✅	✅	❌
Distributed compute	Needed	Not needed	Not needed
Binary size	> 3MB	100KB	~10KB
Base memory footprint	~ 5MB	300KB	20KB
Weights	Variable	Fixed
Topology Neuron connections	Variable	Fixed

References¶

Last Updated: 2024-12-26 ; Contributors: AhmedThahir, web-flow