#Edge computing (5)

Fine-tune, Convert and Deploy Language Models on NVIDIA Jetson with Unsloth and llama.cpp

Official Jetson AI Lab tutorial: fine-tune models directly on Jetson with Unsloth and JetPack 7.2 using memory-efficient QLoRA, export to GGUF, and run locally with llama.cpp. Two hands-on examples: Qwen3.5-4B vision-language model on Jetson Orin Nano (LaTeX OCR fine-tuning), and NVIDIA Nemotron 3.5 Lightning 30B-A3B on Jetson AGX Thor (3 training steps in 44.8s, 66.1 tokens/sec with Q4_K_M quantization). No cloud required.

BLOG

NVIDIAJetsonEdge AIUnslothQLoRA

Jetson AI Lab Tutorial: Fine-tune and Quantize AI Models on Jetson

@NVIDIARobotics · 249 views · 2026-08-31

NVIDIAJetsonEdge AI

Rockchip RK3588 8-Core AIoT Application Processor (SoC)

The RK3588 is Rockchip's high-performance application processor for ARM PCs, edge computing and AIoT terminals, and is widely used in robot main controllers and edge-AI modules. Built on an 8nm process, it integrates 4x Cortex-A76 plus 4x Cortex-A55 cores with a shared 3MB L3 cache, an ARM Mali-G610 MP4 GPU, and a 6 TOPS triple-core NPU supporting INT4/INT8/INT16/FP16/BF16/TF32 mixed-precision operations. Its memory interface supports four-channel 64-bit LPDDR4/LPDDR4X/LPDDR5 with up to 32GB address space, plus eMMC 5.1 HS400 storage. Video capabilities include H.265/VP9 8K@60fps decoding and H.264/H.265 8K@30fps encoding, while the hardware ISP handles up to 48-megapixel input with HDR, 3A, LSC and 3DNR. Connectivity covers HDMI 2.1/eDP/DP/MIPI, PCIe 3.0/2.0, SATA 3.0, RGMII, USB 3.1/2.0 and Type-C, in an FCBGA1088L package (23x23 mm, 0.65 mm ball pitch).

COMPONENT

SoCRK3588ControllerEdge computingNPU

0:19

$8 ESP32-S3 Runs 29M-Parameter LLM

@ardchain · 95 views · 2026-07-24

Edge AIESP32LLM

0:14

NVIDIA Cosmos 3 Edge: 4B Edge World Model for Robots

@Xudong07452910 · 169 views · 2026-07-21

NVIDIACosmos 3 EdgeWorld Models