Alexqzx commited on
Commit
4bd4eed
·
verified ·
1 Parent(s): 5178690

Replace architecture figure with an introductory overview

Browse files
Files changed (2) hide show
  1. README.md +1 -5
  2. README_CN.md +1 -5
README.md CHANGED
@@ -83,11 +83,7 @@ Documented training recipes combine long sequences, multimodal inputs, and super
83
 
84
  ## Model Architecture
85
 
86
- SenseNova-6.8-Loop-9B-Preview combines a 32-layer multimodal decoder with a repeated middle block. The figure illustrates the order of computation; the detailed operations inside each layer and the vision encoder are omitted.
87
-
88
- ![Input passes through early layers, then layers 4 through 27 twice with shared weights and two mHC streams, followed by final layers and text output.](assets/loop9b-architecture.svg)
89
-
90
- *Figure 1. Decoder execution order. Layer indices are zero-based; the two passes share block weights.*
91
 
92
  ### Looped decoder
93
 
 
83
 
84
  ## Model Architecture
85
 
86
+ SenseNova-6.8-Loop-9B-Preview builds on a 32-layer multimodal decoder that combines shared-weight looped computation, hybrid attention, and two-stream manifold-constrained hyper-connections (mHC). The following sections describe how these components extend the computation path and manage information flow through the model.
 
 
 
 
87
 
88
  ### Looped decoder
89
 
README_CN.md CHANGED
@@ -83,11 +83,7 @@ tags:
83
 
84
  ## 模型架构
85
 
86
- SenseNova-6.8-Loop-9B-Preview 采用 32 层多模态解码器,其中的中间层模块会重复执行。下图展示了计算顺序,省略了各层内部操作和视觉编码器的细节。
87
-
88
- ![输入先经过前部层,再通过共享权重与双流 mHC 的第 4–27 层两次,最后经过后部层并生成文本。](assets/loop9b-architecture.svg)
89
-
90
- *图 1:解码器的执行顺序。层编号从 0 开始,两次循环执行共享模块权重。*
91
 
92
  ### 循环解码器
93
 
 
83
 
84
  ## 模型架构
85
 
86
+ SenseNova-6.8-Loop-9B-Preview 以 32 层多模态解码器为基础,结合共享权重的循环计算、混合注意力与双流流形约束超连接(mHC)。下文将分别介绍这些组件如何拓展模型的计算路径,并组织模型内部的信息流动。
 
 
 
 
87
 
88
  ### 循环解码器
89