CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-14 11:07:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-cpu: templateify ggml_compute_forward_rope_f32 and _f16 (#16805)

Browse Source 
* extract rotate_pairs logic from ggml_compute_forward_rope_f32

* templateify ggml_compute_forward_rope_f32 and _f16

* abort when rope type not supported, remove GLM from test-rope

* add imrope branch to switch

* add rope tests for perf

* Update ggml/src/ggml-cpu/ops.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

* Update ggml/src/ggml-cpu/ops.cpp

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
duduta
2025-11-11 13:33:24 +02:00
committed by GitHub
parent 8c583242ad
commit 73460f6278
3 changed files with 76 additions and 268 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/test-backend-ops.cpp
View File

@@ -7603,6 +7603,22 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_perf() {
test_cases.emplace_back(new test_add_id(GGML_TYPE_F32, GGML_TYPE_F32, 2880, 32, 4, n_token));
}
for (bool fw : {true, false}) { // fw == forward
for (ggml_type type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
for (bool ff : {false, true}) { // freq_factors
for (float v : { 0, 1 }) {
test_cases.emplace_back(new test_rope(type, {128, 32, 512, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // llama 7B
test_cases.emplace_back(new test_rope(type, {128, 64, 512, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // llama 65B
test_cases.emplace_back(new test_rope(type, { 80, 32, 512, 1}, 20, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // neox (stablelm)
test_cases.emplace_back(new test_rope(type, { 64, 8, 512, 1}, 64, GGML_ROPE_TYPE_NEOX, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, {128, 12, 512, 1}, 128, GGML_ROPE_TYPE_MROPE, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // rope_multi,m-rope (qwen2vl 2B)
test_cases.emplace_back(new test_rope(type, {128, 12, 2, 1}, 128, GGML_ROPE_TYPE_IMROPE, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // rope_multi,imrope (qwen3vl 2B)
test_cases.emplace_back(new test_rope(type, { 80, 16, 2, 1}, 80, GGML_ROPE_TYPE_VISION, 512, 1.0f, 0.0f, 1.0f, ff, v, fw)); // rope_multi,m-rope (qwen2vl ViT)
}
}
}
}
std::vector<std::array<int64_t, 4>> reduce_rows_cases = {
{ 8192, 1, 1, 1 },
{ 8192, 8192, 1, 1 },

11
tests/test-rope.cpp
View File

@@ -138,7 +138,7 @@ int main(int /*argc*/, const char ** /*argv*/) {
struct ggml_tensor * x;
// rope f32
for (int m = 0; m < 6; ++m) {
for (int m = 0; m < 5; ++m) {
const int ndims = 4;
const int64_t n_rot = 128;
@@ -153,7 +153,7 @@ int main(int /*argc*/, const char ** /*argv*/) {
x = get_random_tensor_f32(ctx0, ndims, ne, -1.0f, 1.0f);
int mode = -1;
if (m < 3) {
if (m < 2) {
struct ggml_tensor * p0 = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, ne[2]);
struct ggml_tensor * p1 = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, ne[2]);
struct ggml_tensor * p2 = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, ne[2]);
@@ -163,8 +163,8 @@ int main(int /*argc*/, const char ** /*argv*/) {
((int32_t *) p1->data)[i] = n_past_2 - n_past_0;
((int32_t *) p2->data)[i] = n_past_2 + i;
}
// test mode 0, 2, 4 (standard, GPT-NeoX, GLM)
mode = m == 0 ? 0 : m == 1 ? 2 : 4;
// test mode 0, 2 (standard, GPT-NeoX)
mode = m == 0 ? GGML_ROPE_TYPE_NORMAL : GGML_ROPE_TYPE_NEOX;
// 100, 101, 102, ..., 172
r0 = ggml_rope(ctx0, x, p0, n_rot, mode);
@@ -180,7 +180,8 @@ int main(int /*argc*/, const char ** /*argv*/) {
struct ggml_tensor * p2 = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, ne[2] * 4);
int sections[4] = {16, 24, 24, 0};
mode = (m == 3) ? GGML_ROPE_TYPE_MROPE : (m == 4) ? GGML_ROPE_TYPE_VISION : GGML_ROPE_TYPE_IMROPE;
mode = (m == 2) ? GGML_ROPE_TYPE_MROPE : (m == 3) ? GGML_ROPE_TYPE_VISION : GGML_ROPE_TYPE_IMROPE;
for (int i = 0; i < ne[2]; ++i) {
for (int j = 0; j < 4; ++j) {