add some new ops, fix some operators and add batch operations to cert…

…ain operators. (#747) * 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 include/ggml/ggml.h Co-authored-by: Georgi Gerganov <[email protected]> * Update src/ggml-cuda.cu Co-authored-by: Georgi Gerganov <[email protected]> * Update src/ggml-metal.m Co-authored-by: Georgi Gerganov <[email protected]> * Update src/ggml-metal.m Co-authored-by: Georgi Gerganov <[email protected]> * Update src/ggml-metal.metal Co-authored-by: Georgi Gerganov <[email protected]> * 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 <[email protected]> Co-authored-by: slaren <[email protected]>
ggerganov · Mar 3, 2024 · 2746808 · 2746808
1 parent d9cd6b5
commit 2746808
Show file tree

Hide file tree

Showing 11 changed files with 867 additions and 56 deletions.
diff --git a/examples/simple/simple-backend.cpp b/examples/simple/simple-backend.cpp
@@ -19,6 +19,13 @@
 #include <string>
 #include <vector>
 
+static void ggml_log_callback_default(ggml_log_level level, const char * text, void * user_data) {
+ (void) level;
+ (void) user_data;
+ fputs(text, stderr);
+ fflush(stderr);
+}
+
 // This is a simple model with two tensors a and b
 struct simple_model {
  struct ggml_tensor * a;
@@ -47,7 +54,7 @@ void load_model(simple_model & model, float * a, float * b, int rows_A, int cols
 
 #ifdef GGML_USE_METAL
  fprintf(stderr, "%s: using Metal backend\n", __func__);
- ggml_metal_log_set_callback(ggml_log_callback_default, nullptr);
+ ggml_backend_metal_log_set_callback(ggml_log_callback_default, nullptr);
  model.backend = ggml_backend_metal_init();
  if (!model.backend) {
  fprintf(stderr, "%s: ggml_backend_metal_init() failed\n", __func__);
@@ -191,7 +198,7 @@ int main(void) {
  // 55.00 90.00 126.00 28.00
  // 50.00 54.00 42.00 64.00 ]
 
- printf("mul mat (%d x %d) (transposed result):\n[", result->ne[0], result->ne[1]);
+ printf("mul mat (%d x %d) (transposed result):\n[", (int) result->ne[0], (int) result->ne[1]);
  for (int j = 0; j < result->ne[1] /* rows */; j++) {
  if (j > 0) {
  printf("\n");

diff --git a/examples/simple/simple-ctx.cpp b/examples/simple/simple-ctx.cpp
@@ -108,7 +108,7 @@ int main(void) {
  // 55.00 90.00 126.00 28.00
  // 50.00 54.00 42.00 64.00 ]
 
- printf("mul mat (%d x %d) (transposed result):\n[", result->ne[0], result->ne[1]);
+ printf("mul mat (%d x %d) (transposed result):\n[", (int) result->ne[0], (int) result->ne[1]);
  for (int j = 0; j < result->ne[1] /* rows */; j++) {
  if (j > 0) {
  printf("\n");

diff --git a/include/ggml/ggml.h b/include/ggml/ggml.h
@@ -454,6 +454,8 @@ extern "C" {
  GGML_OP_POOL_2D,
  GGML_OP_UPSCALE, // nearest interpolate
  GGML_OP_PAD,
+ GGML_OP_ARANGE,
+ GGML_OP_TIMESTEP_EMBEDDING,
  GGML_OP_ARGSORT,
  GGML_OP_LEAKY_RELU,
 
@@ -1661,6 +1663,15 @@ extern "C" {
  int p2,
  int p3);
 
+ // Ref: https://github.com/CompVis/stable-diffusion/blob/main/ldm/modules/diffusionmodules/util.py#L151
+ // timesteps: [N,]
+ // return: [N, dim]
+ GGML_API struct ggml_tensor * ggml_timestep_embedding(
+ struct ggml_context * ctx,
+ struct ggml_tensor * timesteps,
+ int dim,
+ int max_period);
+
  // sort rows
  enum ggml_sort_order {
  GGML_SORT_ORDER_ASC,
@@ -1672,6 +1683,12 @@ extern "C" {
  struct ggml_tensor * a,
  enum ggml_sort_order order);
 
+ GGML_API struct ggml_tensor * ggml_arange(
+ struct ggml_context * ctx,
+ float start,
+ float stop,
+ float step);
+
  // top k elements per row
  GGML_API struct ggml_tensor * ggml_top_k(
  struct ggml_context * ctx,