CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml webgpu: support for rope,div,sub,glu,scale,cont operators (#16187)

Browse Source 
* Work on rope

* Simplify inplace operation generation and combine mul/add generation

* Work on rope variants

* implement neox rope

* rope complete

* Add sub,div,glu operators

* implement scale op

* Update cpy shader to handle cont/more types

* formatting

* Update test vars printing for rope,rms_norm

* Avoid ROPE hardcoded constants

* Add TODO to change ROPE constants to enum

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

* fix TODO comment

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Reese Levine
2025-09-30 09:57:51 -07:00
committed by GitHub
parent d1c84a662d
commit 8d78cd2613
17 changed files with 1534 additions and 397 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
tests/test-backend-ops.cpp
View File

@@ -2733,23 +2733,30 @@ struct test_scale : public test_case {
const std::array<int64_t, 4> ne;
float scale;
float bias;
bool inplace;
std::string vars() override {
return VARS_TO_STR4(type, ne, scale, bias);
return VARS_TO_STR5(type, ne, scale, bias, inplace);
}
test_scale(ggml_type type = GGML_TYPE_F32,
std::array<int64_t, 4> ne = {10, 10, 10, 10},
float scale = 2.0f,
float bias = 0.0f)
: type(type), ne(ne), scale(scale), bias(bias) {}
float bias = 0.0f,
bool inplace = false)
: type(type), ne(ne), scale(scale), bias(bias), inplace(inplace) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne.data());
ggml_set_param(a);
ggml_set_name(a, "a");
ggml_tensor * out = ggml_scale_bias(ctx, a, scale, bias);
ggml_tensor * out;
if (inplace) {
out = ggml_scale_bias_inplace(ctx, a, scale, bias);
} else {
out = ggml_scale_bias(ctx, a, scale, bias);
}
ggml_set_name(out, "out");
return out;
@@ -2906,16 +2913,18 @@ struct test_rms_norm : public test_case {
const std::array<int64_t, 4> ne;
const bool v; // whether a is a non-contiguous view
const float eps;
const bool inplace; // whether to do the operation inplace
std::string vars() override {
return VARS_TO_STR4(type, ne, v, eps);
return VARS_TO_STR5(type, ne, v, eps, inplace);
}
test_rms_norm(ggml_type type = GGML_TYPE_F32,
std::array<int64_t, 4> ne = {64, 5, 4, 3},
bool v = false,
float eps = 1e-6f)
: type(type), ne(ne), v(v), eps(eps) {}
float eps = 1e-6f,
bool inplace = false)
: type(type), ne(ne), v(v), eps(eps), inplace(inplace) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne.data());
@@ -2927,7 +2936,12 @@ struct test_rms_norm : public test_case {
ggml_set_name(a, "view of a");
}
ggml_tensor * out = ggml_rms_norm(ctx, a, eps);
ggml_tensor * out;
if (inplace) {
out = ggml_rms_norm_inplace(ctx, a, eps);
} else {
out = ggml_rms_norm(ctx, a, eps);
}
ggml_set_name(out, "out");
return out;
@@ -3832,17 +3846,18 @@ struct test_rope : public test_case {
bool ff;
int v; // view (1 : non-contiguous a)
bool forward;
bool inplace;
std::string vars() override {
// forward can be inferred from the op, does not need to be printed
return VARS_TO_STR10(type, ne_a, n_dims, mode, n_ctx, fs, ef, af, ff, v);
return VARS_TO_STR11(type, ne_a, n_dims, mode, n_ctx, fs, ef, af, ff, v, inplace);
}
test_rope(ggml_type type = GGML_TYPE_F32,
std::array<int64_t, 4> ne_a = {10, 5, 3, 1},
int n_dims = 10, int mode = 0, int n_ctx = 512, float fs = 1.0f,
float ef = 0.0f, float af = 0.0f, bool ff = false, int v = 0, bool forward = true)
: type(type), ne_a(ne_a), n_dims(n_dims), mode(mode), n_ctx(n_ctx), fs(fs), ef(ef), af(af), ff(ff), v(v), forward(forward) {}
int n_dims = 10, int mode = GGML_ROPE_TYPE_NORMAL, int n_ctx = 512, float fs = 1.0f,
float ef = 0.0f, float af = 0.0f, bool ff = false, int v = 0, bool forward = true, bool inplace = false)
: type(type), ne_a(ne_a), n_dims(n_dims), mode(mode), n_ctx(n_ctx), fs(fs), ef(ef), af(af), ff(ff), v(v), forward(forward), inplace(inplace) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * a;
@@ -3887,7 +3902,11 @@ struct test_rope : public test_case {
GGML_ASSERT(n_dims/4 > 0);
int rope_sections[4] = {n_dims/4, n_dims/4, 0, 0}; // Vision-RoPE only use first two dimension for image (x, y) coordinate
if (forward) {
out = ggml_rope_multi (ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
if (inplace) {
out = ggml_rope_multi_inplace(ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
} else {
out = ggml_rope_multi(ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
} else {
out = ggml_rope_multi_back(ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
@@ -3895,14 +3914,22 @@ struct test_rope : public test_case {
GGML_ASSERT(n_dims/3 > 0);
int rope_sections[4] = {n_dims/3, n_dims/3, n_dims/3, 0};
if (forward) {
out = ggml_rope_multi (ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
if (inplace) {
out = ggml_rope_multi_inplace(ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
} else {
out = ggml_rope_multi(ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
} else {
out = ggml_rope_multi_back(ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
}
} else {
if (forward) {
out = ggml_rope_ext (ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
if (inplace) {
out = ggml_rope_ext_inplace(ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
} else {
out = ggml_rope_ext(ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
} else {
out = ggml_rope_ext_back(ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
}
@@ -6183,9 +6210,11 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
//add_test_bin_bcast(type, {3, 3, 2560, 1280}, {2, 1, 1, 1});
}
// single in-place tests, especially important for WebGPU backend since kernels for in-place vs. not are different
// single inplace tests, especially important for WebGPU backend since kernels for inplace vs. not are different
test_cases.emplace_back(new test_bin_bcast(ggml_add_inplace, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
test_cases.emplace_back(new test_bin_bcast(ggml_mul_inplace, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
test_cases.emplace_back(new test_bin_bcast(ggml_sub_inplace, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
test_cases.emplace_back(new test_bin_bcast(ggml_div_inplace, GGML_TYPE_F32, {16, 5, 4, 3}, {1, 1, 1, 1}, 16));
// fusion
test_cases.emplace_back(new test_bin_bcast(ggml_add, GGML_TYPE_F32, {10, 5, 4, 3}, {2, 1, 1, 1}, 2));
@@ -6200,6 +6229,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_add1());
test_cases.emplace_back(new test_scale());
test_cases.emplace_back(new test_scale(GGML_TYPE_F32, {10, 10, 10, 10}, 2.0f, 1.0f));
test_cases.emplace_back(new test_scale(GGML_TYPE_F32, {10, 10, 10, 10}, 2.0f, 1.0f, true)); // inplace test
test_cases.emplace_back(new test_scale(GGML_TYPE_F32, {100, 10, 10, 10}, 2.0f, 1.0f));
test_cases.emplace_back(new test_softcap(GGML_TYPE_F32, {10, 10, 10, 10}, 50.0f));
test_cases.emplace_back(new test_silu_back());
@@ -6212,6 +6242,10 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_rms_norm_back(GGML_TYPE_F32, {64, 5, 4, 3}, eps));
test_cases.emplace_back(new test_l2_norm (GGML_TYPE_F32, {64, 5, 4, 3}, eps));
}
// in-place tests
test_cases.emplace_back(new test_rms_norm(GGML_TYPE_F32, {64, 5, 4, 3}, false, 1e-6f, true));
for (float eps : {0.0f, 1e-6f, 1e-4f, 1e-1f, 1.0f}) {
test_cases.emplace_back(new test_rms_norm_mul_add(GGML_TYPE_F32, {64, 5, 4, 3}, eps, false));
test_cases.emplace_back(new test_rms_norm_mul_add(GGML_TYPE_F32, {64, 5, 4, 3}, eps, true));
@@ -6559,26 +6593,26 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
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, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 7B
test_cases.emplace_back(new test_rope(type, {128, 32, 2, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw)); // llama 7B
if (all) {
test_cases.emplace_back(new test_rope(type, {128, 40, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 13B
test_cases.emplace_back(new test_rope(type, {128, 52, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 30B
test_cases.emplace_back(new test_rope(type, {128, 64, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 65B
test_cases.emplace_back(new test_rope(type, {128, 40, 2, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw)); // llama 13B
test_cases.emplace_back(new test_rope(type, {128, 52, 2, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw)); // llama 30B
test_cases.emplace_back(new test_rope(type, {128, 64, 2, 1}, 128, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw)); // llama 65B
}
if (all) {
test_cases.emplace_back(new test_rope(type, { 64, 1, 2, 1}, 64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
test_cases.emplace_back(new test_rope(type, { 64, 71, 2, 1}, 64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
test_cases.emplace_back(new test_rope(type, { 64, 8, 2, 1}, 64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, { 64, 1, 2, 1}, 64, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
test_cases.emplace_back(new test_rope(type, { 64, 71, 2, 1}, 64, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
test_cases.emplace_back(new test_rope(type, { 64, 8, 2, 1}, 64, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 20, 0, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 32, 0, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 4, 1}, 32, 0, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 20, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 32, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 4, 1}, 32, GGML_ROPE_TYPE_NORMAL, 512, fs, ef, af, ff, v, fw));
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 20, 2, 512, fs, ef, af, ff, v, fw)); // neox (stablelm)
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 32, 2, 512, fs, ef, af, ff, v, fw)); // neox (phi-2)
test_cases.emplace_back(new test_rope(type, { 80, 32, 4, 1}, 32, 2, 512, fs, ef, af, ff, v, fw)); // neox (phi-2)
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 20, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (stablelm)
test_cases.emplace_back(new test_rope(type, { 80, 32, 2, 1}, 32, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (phi-2)
test_cases.emplace_back(new test_rope(type, { 80, 32, 4, 1}, 32, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (phi-2)
}
if (all) {
@@ -6589,7 +6623,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_rope(type, { 80, 16, 2, 1}, 80, GGML_ROPE_TYPE_VISION, 512, fs, ef, af, ff, v, fw)); // rope_multi,m-rope (qwen2vl ViT)
}
test_cases.emplace_back(new test_rope(type, { 64, 128, 2, 1}, 64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, { 64, 128, 2, 1}, 64, GGML_ROPE_TYPE_NEOX, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
}
}
@@ -6600,6 +6634,15 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
}
}
// single inplace test per type/mode/ff
for (ggml_type type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
for (int mode : {GGML_ROPE_TYPE_NORMAL, GGML_ROPE_TYPE_NEOX, GGML_ROPE_TYPE_MROPE, GGML_ROPE_TYPE_VISION}) {
for (bool ff : {false, true}) {
test_cases.emplace_back(new test_rope(type, {128, 32, 2, 1}, 128, mode, 512, 1.4245f, 0.7465f, 1.4245f, ff, 0, true, true));
}
}
}
for (int v : { 0, 1, 2, 3 }) {
for (int dim : { 0, 1, 2, 3, }) {
test_cases.emplace_back(new test_concat(GGML_TYPE_F32, {11, 12, 13, 14}, 7, dim, v));