vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add some new ops, fix some operators and add batch operations to certain operators. (ggml/747)

Browse source 
* cuda: fix group_norm

* cuda: add batch inference support for ggml_pad/ggml_upscale

* add ggml_arrange

* add ggml_timestep_embedding

* update ggml_arange/ggml_timestep_embedding tests

* cuda: fix im2col

* add ggml_arange/ggml_timestep_embbeding support for metal backend

* fix some bugs

* fix some bugs

* Update ggml.h

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

* Update ggml-cuda.cu

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

* Update ggml-metal.m

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

* Update ggml-metal.m

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

* Update ggml-metal.metal

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

* modify according to the review comments

* ggml : fix compile warnings + code style

* ggml : normalize compute_forward calls + fix seg fault in debug

* minor

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
leejet 2024-03-03 20:23:52 +08:00 • committed by Georgi Gerganov
parent 82f3e668ad
commit 7d43c585dc
No known key found for this signature in database
GPG key ID: BF970631944C16B7
6 changed files with 550 additions and 52 deletions
Download patch file Download diff file Expand all files Collapse all files

46
tests/test-backend-ops.cpp
View file

@ -1412,6 +1412,50 @@ struct test_pad : public test_case {
}
};
// GGML_OP_ARANGE
struct test_arange : public test_case {
const ggml_type type;
const float start;
const float stop;
const float step;
std::string vars() override {
return VARS_TO_STR4(type, start, stop, step);
}
test_arange(ggml_type type = GGML_TYPE_F32,
float start = 0.f, float stop = 10.f, float step = 1.f)
: type(type), start(start), stop(stop), step(step) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * out = ggml_arange(ctx, start, stop, step);
return out;
}
};
// GGML_OP_TIMESTEP_EMBEDDING
struct test_timestep_embedding : public test_case {
const ggml_type type;
const std::array<int64_t, 4> ne_a;
const int dim;
const int max_period;
std::string vars() override {
return VARS_TO_STR4(type, ne_a, dim, max_period);
}
test_timestep_embedding(ggml_type type = GGML_TYPE_F32,
std::array<int64_t, 4> ne_a = {2, 1, 1, 1},
int dim = 320, int max_period=10000)
: type(type), ne_a(ne_a), dim(dim), max_period(max_period) {}
ggml_tensor * build_graph(ggml_context * ctx) override {
ggml_tensor * a = ggml_new_tensor(ctx, type, 4, ne_a.data());
ggml_tensor * out = ggml_timestep_embedding(ctx, a, dim, max_period);
return out;
}
};
// GGML_OP_LEAKY_RELU
struct test_leaky_relu : public test_case {
const ggml_type type;
@ -2126,6 +2170,8 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
test_cases.emplace_back(new test_group_norm());
test_cases.emplace_back(new test_acc());
test_cases.emplace_back(new test_pad());
test_cases.emplace_back(new test_arange());
test_cases.emplace_back(new test_timestep_embedding());
test_cases.emplace_back(new test_leaky_relu());
// these tests are disabled to save execution time, but they can be handy for debugging