Unverified50% confidenceReleaseExact time
iFlytek open-sourced two small models of Spark X 2.5, the 4B and 1.7B, targeting edge-side agent scenarios, employing a hybrid attention architecture with native support for up to 1 million token context.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/12/2026
First Seen
Valid until: 12/11/2026
Sources
Related Entities
Related Claims
UnverifiedHymba是NVIDIA于2024年提出的混合头注意力架构，在同一层内同时部署SSM头和注意力头，通过可学习门控机制融合输出，在1.5B参数规模下超越同规模纯Transformer和纯SSM模型67% similarUnverifiedThe model architecture is a cross-attention DiT (896 width × 16 layers) using 2D RoPE, QK-norm, SwiGLU, and adaLN-single.66% similarUnverifiedGLM-5.3-Flash adopts a hybrid architecture of sparse attention and linear attention, and runs entirely on a domestically produced AI chip cluster.65% similarUnverified许多HuggingFace模型仓库已经提供了ONNX版本，特别是针对CPU和移动端优化的量化模型62% similarUnverifiedLLaMA-2 70B采用80层、64个注意力头的架构，并使用GQA（8个KV头共享64个Q头）62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/912723API
curl https://kongchang.com/api/v1/knowledge/claims/912723MCP
get_claim(id=912723)