CANN: add support for partial RoPE and Vision mode (#17543)

* cann: add support for partial RoPE and Vision mode

Add support for two important RoPE variants: partial rotation (rope_dims < ne0)
and Vision mode rotation.

1. Support for partial RoPE (rope_dims < ne0):
   - Split tensor into head (first rope_dims dimensions) and tail portions
   - Apply rotation only to head portion using RotaryPositionEmbedding operator
   - Copy unrotated tail portion directly from source to destination
   - Handle both contiguous and non-contiguous tensor layouts

2. Support for Vision mode (GGML_ROPE_TYPE_VISION):
   - Set rope_dims = ne0 for Vision mode to rotate entire tensor
   - Vision mode pairs dimension i with dimension i+n_dims (where n_dims = ne0/2)
   - No tail handling needed since entire tensor is rotated

Implementation details:
   - Use has_tail flag to determine execution path: head/tail splitting when
     rope_dims < ne0, or full tensor rotation when rope_dims == ne0
   - Support both F32 and F16 data types with intermediate F32 conversion
   - Copy non-contiguous tensors to contiguous buffers before calling
     RotaryPositionEmbedding operator for compatibility
   - Improve cache invalidation logic to include rope_dims and indep_sects
     parameters

These enhancements enable CANN backend to handle various RoPE configurations
used in modern vision-language models and models with partial rotation.

* cann: fix review comment
This commit is contained in:
Chenguang Li
2025-12-09 17:53:23 +08:00
committed by GitHub
parent 0cdce38a97
commit ca709e427b
3 changed files with 162 additions and 72 deletions
+2 -2
View File
@@ -315,7 +315,7 @@ struct ggml_cann_rope_cache {
if (theta_scale_exp_host) {
free(theta_scale_exp_host);
}
if(position_select_index_host) {
if (position_select_index_host) {
free(position_select_index_host);
}
}
@@ -340,7 +340,7 @@ struct ggml_cann_rope_cache {
void set(int64_t theta_scale_length,
int64_t position_length,
float ext_factor,
float ext_factor,
float theta_scale,
float freq_scale,
float attn_factor,