EdgeRock Open-Sources Constant-Memory Multimodal Inference for Edge Devices

EdgeRock, an Apache-2.0 open-source project, achieves strong multimodal reasoning with constant per-request memory of about 97.76 MiB regardless of context length, scoring 82.9 on AIME with minimal degradation, making it well-suited for memory-constrained edge devices like robot dogs and autonomous vehicles.

2026-10-23 ~ 2026-10-23 · 4 related posts

• EdgeRock-4B locks推理 memory at 97.76 MiB per request while scoring 82.9% on AIME — xiaohu · 2026-10-23
• Why EdgeRock-4B fits robots and edge devices: fixed memory that never balloons — xiaohu · 2026-10-23
• EdgeRock: Open-Source Multimodal Reasoning at Constant Memory With O(1) History — xiaohu · 2026-10-23

1 near-duplicate retellings: xiaohu