Expired50% confidenceReleaseExact time
The Gemma 3 series released in 2025 uses a SigLIP vision encoder jointly trained with a language model, supports image understanding, and extends the context window to 128K tokens.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/6/2026
First Seen
Valid until: 10/4/2026(expired)
Sources
AIVenture: Master Agent Development and Vibe Coding Through a Dungeon Game
bilibili多巴胺矩阵7/3/2026
Related Claims
UnverifiedThe Gemma series of models uses the SigLIP vision encoder to process image inputs.74% similarUnverifiedGemma 4支持完全离线状态下的图像文字提取（OCR），采用视觉-语言多模态架构63% similarUnverified混元系列采用晚融合（Late Fusion）的改良版本，结合视觉指令微调（Visual Instruction Tuning）技术，通过视觉编码器将图像信息转化为与文本token同维度的向量表示62% similarVerifiedThe LayoutLM model introduced by Microsoft Research in 2020 fuses three types of information during training: text content, font position coordinates, and image features.59% similarUnverifiedLayoutLMv3 is the third-generation document understanding pre-trained model released by Microsoft Research Asia in 2022, which unifies three modalities—text content, two-dimensional coordinate positions, and document images—into a Transformer architecture.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/114954API
curl https://kongchang.com/api/v1/knowledge/claims/114954MCP
get_claim(id=114954)