Creates a graph node for multi-dimensional RoPE (MRoPE) used in vision transformers. Supports separate rotation for different positional dimensions (e.g., height, width, time).
ggml_rope_multi(
ctx,
a,
b,
c = NULL,
n_dims,
sections = c(0L, 0L, 0L, 0L),
mode = 0L,
n_ctx_orig = 0L,
freq_base = 10000,
freq_scale = 1,
ext_factor = 0,
attn_factor = 1,
beta_fast = 32,
beta_slow = 1
)Tensor with multi-dimensional RoPE applied
GGML context
Input tensor
Position tensor (int32)
Optional frequency factors tensor (NULL for default)
Number of dimensions to apply rotation to
Integer vector of length 4 specifying dimension sections for MRoPE
RoPE mode
Original context length the model was trained on
Base frequency for RoPE (default 10000 for most models)
Frequency scale factor (1.0 = no scaling)
YaRN extension factor (0.0 to disable)
Attention scale factor (typically 1.0)
YaRN parameter for fast dimensions
YaRN parameter for slow dimensions
Other rope:
ggml_rope_ext_inplace(),
ggml_rope_multi_inplace()