Unverified50% confidenceFactExact time
LLaVA proposed a minimalist three-stage paradigm of visual encoder + projection layer + LLM in 2023, which was inherited by subsequent works such as InternVL and Qwen-VL.
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
DeepSeek神秘撤稿论文：用视觉基元突破AI多模态推理瓶颈
bilibili枫叶边城7/11/2026
Related Claims
VerifiedRoPE (Rotary Position Embedding) was proposed by Su Jianlin in 2021 and has become standard in mainstream models such as LLaMA, Qwen, and Mistral.66% similarUnverifiedLLaMA 3 is Meta's third-generation open-source large language model series, released in April 2024.64% similarUnverifiedvLLM was open-sourced by UC Berkeley in 2023 and increased LLM inference throughput by 2–4x through its PagedAttention technology.64% similarUnverifiedLlama 1于2023年2月发布，参数规模为7B至65B，采用纯Transformer解码器架构，引入RoPE位置编码和SwiGLU激活函数62% similarUnverifiedThe first-generation Qwen was released in August 2023, based on the standard Transformer architecture, employing RoPE, pre-layer RMS Norm, and SwiGLU activation functions, with bias terms removed from most linear layers.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/609003API
curl https://kongchang.com/api/v1/knowledge/claims/609003MCP
get_claim(id=609003)