Unverified50% confidenceFactExact time

Agnes-3.0-Flash's positional encoding uses tri-axis rotation (text, height, width), with interleaved mrope segment ratios of 11:11:10, a base of 1e7, applied only to the first 25% (64 dimensions) of each head dim

1

Sources

50%

Confidence

Long-term

Relevance

9/12/2026

First Seen

Sources

Agnes-3.0-Flash Deep Dive: A 33B Hybrid-Attention Multimodal Model

Related Entities

Agnes-3.0-Flashmodel mropeconcept

Related Claims

UnverifiedAgnes-3.0-Flash is equipped with a standalone vision encoder: 27 layers, hidden dimension 1152, patch size 16, projected to 5120 dimensions after 2×2 spatial merging71% similar UnverifiedOpenCV's cornerSubPix function can improve corner localization accuracy from the integer-pixel level to the 0.01-pixel level; skipping this step results in corner localization errors of 0.5–1 pixel.60% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/916675

API

curl https://kongchang.com/api/v1/knowledge/claims/916675

MCP

get_claim(id=916675)