Alexqzx commited on
Commit
5178690
·
verified ·
1 Parent(s): a3f9d4b

Remove training sequence length caveat

Browse files
Files changed (2) hide show
  1. README.md +0 -2
  2. README_CN.md +0 -2
README.md CHANGED
@@ -81,8 +81,6 @@ Documented training recipes combine long sequences, multimodal inputs, and super
81
  | Training context | Up to 262,144 tokens |
82
  | Existing export precision | BF16 backbone; FP32 mHC routing parameters |
83
 
84
- Training sequence length is a configuration, not a guarantee of inference support or performance at every length.
85
-
86
  ## Model Architecture
87
 
88
  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.
 
81
  | Training context | Up to 262,144 tokens |
82
  | Existing export precision | BF16 backbone; FP32 mHC routing parameters |
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.
README_CN.md CHANGED
@@ -81,8 +81,6 @@ tags:
81
  | 训练上下文长度 | 最长 262,144 个 token |
82
  | 现有导出精度 | 主干为 BF16;mHC 路由参数为 FP32 |
83
 
84
- 训练序列长度属于训练配置,并不代表推理实现支持所有对应长度,也不构成各长度下的性能保证。
85
-
86
  ## 模型架构
87
 
88
  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。
 
81
  | 训练上下文长度 | 最长 262,144 个 token |
82
  | 现有导出精度 | 主干为 BF16;mHC 路由参数为 FP32 |
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。