tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-12 03:21:10 -04:00
Code Issues Packages Projects Releases Wiki Activity

ggml : add ggml_set_rows (#14274)

Browse Source 
* ggml : add ggml_set_rows

Add ggml_set_rows(a, b, c) which copies rows from 'b' into 'a' using
indices from 'c'.

ref: #8366

* use I64 for indices

* ggml : add repeat impl for i64

* ggml : add ggml_is_contiguous_rows

* ggml : ggml_set_rows support broadcast

* ggml : ggml_set_rows support quantized dst

ggml-ci

* ggml : support GGML_TYPE_F32 ".from_float" trait

* ggml : ggml_set_rows update comment + better index name

* tests : add ggml_set_rows

* metal : add ggml_set_rows implementation

ggml-ci

* ggml : simplify forward_dup_f32

* ggml : fix supports_op

* tests : add comment to set_rows

* ggml : leave the repeat_i64 for a separate PR

ggml-ci

* ggml : set_rows use std::min instead of MIN

* ggml : better error message for set_rows unsupported type

* metal : perform op->type check only once

* tests : more consistent implementation + more tests

ggml-ci

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Radoslav Gerganov
2025-06-27 16:41:40 +03:00
committed by GitHub
parent f667f1e624
commit 8d94219a4a
12 changed files with 653 additions and 204 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
ggml/include/ggml.h
View File

@@ -470,6 +470,7 @@ extern "C" {
GGML_OP_TRANSPOSE,
GGML_OP_GET_ROWS,
GGML_OP_GET_ROWS_BACK,
GGML_OP_SET_ROWS,
GGML_OP_DIAG,
GGML_OP_DIAG_MASK_INF,
GGML_OP_DIAG_MASK_ZERO,
@@ -687,6 +688,9 @@ extern "C" {
// true for tensor that is stored in memory as CxWxHxN and has been permuted to WxHxCxN
GGML_API bool ggml_is_contiguous_channels(const struct ggml_tensor * tensor);
// true if the elements in dimension 0 are contiguous, or there is just 1 block of elements
GGML_API bool ggml_is_contiguous_rows(const struct ggml_tensor * tensor);
GGML_API bool ggml_are_same_shape (const struct ggml_tensor * t0, const struct ggml_tensor * t1);
GGML_API bool ggml_are_same_stride(const struct ggml_tensor * t0, const struct ggml_tensor * t1);
@@ -1375,6 +1379,23 @@ extern "C" {
struct ggml_tensor * b, // row indices
struct ggml_tensor * c); // data for ggml_get_rows, only used for its shape
// a TD [n_embd, ne1, ne2, ne3]
// b TS [n_embd, n_rows, ne02, ne03] | ne02 == ne2, ne03 == ne3
// c I64 [n_rows, ne11, ne12, 1] | c[i] in [0, ne1)
//
// undefined behavior if destination rows overlap
//
// broadcast:
// ne2 % ne11 == 0
// ne3 % ne12 == 0
//
// return view(a)
GGML_API struct ggml_tensor * ggml_set_rows(
struct ggml_context * ctx,
struct ggml_tensor * a, // destination
struct ggml_tensor * b, // source
struct ggml_tensor * c); // row indices
GGML_API struct ggml_tensor * ggml_diag(
struct ggml_context * ctx,
struct ggml_tensor * a);