Module layers

Structs

• CausalMasker
• Conv3dConfig
• Conv3dNoBias
• DeepSeekV2RopeConfig
• DeepSeekV2RotaryEmbedding
• F32RmsNorm
• Llama3RopeConfig
• MatMul
  Matrix multiplication, configurable to be via f16 (to use the faster GEMM kernels) optionally.
• PhiRopeConfig
• PhiRotaryEmbedding
  RoPE supporting LongRope
• QLinear
• QRmsNorm
• Qwen2VLRotaryEmbedding
• RmsNorm
• RotaryEmbedding
• Sdpa

Enums

• Activation
• DeepSeekV2RopeScaling
• Llama3RopeType
• Llama3RotaryEmbedding
  RoPE for Llama3
• PhiRopeScalingConfig
• ScaledRopeType

Functions

• get_use_matmul_via_f16
• repeat_kv