matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 08:47:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update llama.cpp vendor code to commit d7cfe1ff (#9356)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-02-26 20:34:44 -08:00
committed by GitHub
parent 2db96c18e7
commit d7d7e99662
149 changed files with 18215 additions and 11009 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
llama/patches/0008-add-unpad-operator.patch
View File

@@ -15,10 +15,10 @@ Subject: [PATCH] add unpad operator
8 files changed, 220 insertions(+), 2 deletions(-)
diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index c714fc8c..1bc50fca 100644
index dd0c6a96..8d269a9c 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -499,6 +499,7 @@ extern "C" {
@@ -487,6 +487,7 @@ extern "C" {
GGML_OP_UPSCALE, // nearest interpolate
GGML_OP_PAD,
GGML_OP_PAD_REFLECT_1D,
@@ -26,7 +26,7 @@ index c714fc8c..1bc50fca 100644
GGML_OP_ARANGE,
GGML_OP_TIMESTEP_EMBEDDING,
GGML_OP_ARGSORT,
@@ -1735,6 +1736,15 @@ extern "C" {
@@ -1743,6 +1744,15 @@ extern "C" {
int p0,
int p1);
@@ -43,10 +43,10 @@ index c714fc8c..1bc50fca 100644
// timesteps: [N,]
// return: [N, dim]
diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
index b7fefb9d..b307d554 100644
index 72325349..2f606d82 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -10588,6 +10588,59 @@ static void ggml_compute_forward_pad_reflect_1d(
@@ -10844,6 +10844,59 @@ static void ggml_compute_forward_pad_reflect_1d(
}
}
@@ -106,7 +106,7 @@ index b7fefb9d..b307d554 100644
// ggml_compute_forward_arange
static void ggml_compute_forward_arange_f32(
@@ -12690,6 +12743,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
@@ -13137,6 +13190,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
{
ggml_compute_forward_pad_reflect_1d(params, tensor);
} break;
@@ -117,7 +117,7 @@ index b7fefb9d..b307d554 100644
case GGML_OP_ARANGE:
{
ggml_compute_forward_arange(params, tensor);
@@ -13033,6 +13090,7 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
@@ -13484,6 +13541,7 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
case GGML_OP_UPSCALE:
case GGML_OP_PAD:
case GGML_OP_PAD_REFLECT_1D:
@@ -126,10 +126,10 @@ index b7fefb9d..b307d554 100644
case GGML_OP_TIMESTEP_EMBEDDING:
case GGML_OP_ARGSORT:
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index aaa79ea4..9286f866 100644
index 36165840..1adf08fa 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -2082,6 +2082,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
@@ -2198,6 +2198,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_PAD:
ggml_cuda_op_pad(ctx, dst);
break;
@@ -139,8 +139,8 @@ index aaa79ea4..9286f866 100644
case GGML_OP_ARANGE:
ggml_cuda_op_arange(ctx, dst);
break;
@@ -3010,6 +3013,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
case GGML_OP_GROUP_NORM:
@@ -3197,6 +3200,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
return ggml_is_contiguous(op->src[0]);
case GGML_OP_UPSCALE:
case GGML_OP_PAD:
+ case GGML_OP_UNPAD:
@@ -148,7 +148,7 @@ index aaa79ea4..9286f866 100644
case GGML_OP_TIMESTEP_EMBEDDING:
case GGML_OP_LEAKY_RELU:
diff --git a/ggml/src/ggml-cuda/pad.cu b/ggml/src/ggml-cuda/pad.cu
index aba539e8..39fd4b16 100644
index aba539e8..b4b87409 100644
--- a/ggml/src/ggml-cuda/pad.cu
+++ b/ggml/src/ggml-cuda/pad.cu
@@ -47,3 +47,49 @@ void ggml_cuda_op_pad(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
@@ -201,6 +201,7 @@ index aba539e8..39fd4b16 100644
+ src0->ne[0], src0->ne[1], src0->ne[2], src0->ne[3],
+ dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3], stream);
+}
\ No newline at end of file
diff --git a/ggml/src/ggml-cuda/pad.cuh b/ggml/src/ggml-cuda/pad.cuh
index 8fd386b0..e2ededc3 100644
--- a/ggml/src/ggml-cuda/pad.cuh
@@ -211,10 +212,10 @@ index 8fd386b0..e2ededc3 100644
void ggml_cuda_op_pad(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+void ggml_cuda_op_unpad(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
diff --git a/ggml/src/ggml-metal/ggml-metal.m b/ggml/src/ggml-metal/ggml-metal.m
index cd8ef741..318addec 100644
index fd9a4e77..e4c093f9 100644
--- a/ggml/src/ggml-metal/ggml-metal.m
+++ b/ggml/src/ggml-metal/ggml-metal.m
@@ -311,6 +311,7 @@ enum ggml_metal_kernel_type {
@@ -331,6 +331,7 @@ static void ggml_backend_metal_device_rel(struct ggml_backend_metal_device_conte
GGML_METAL_KERNEL_TYPE_UPSCALE_F32,
GGML_METAL_KERNEL_TYPE_PAD_F32,
GGML_METAL_KERNEL_TYPE_PAD_REFLECT_1D_F32,
@@ -222,7 +223,7 @@ index cd8ef741..318addec 100644
GGML_METAL_KERNEL_TYPE_ARANGE_F32,
GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32,
GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC,
@@ -910,6 +911,7 @@ static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t de
@@ -946,6 +947,7 @@ @implementation GGMLMetalClass
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_UPSCALE_F32, upscale_f32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_PAD_F32, pad_f32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_PAD_REFLECT_1D_F32, pad_reflect_1d_f32, true);
@@ -230,7 +231,7 @@ index cd8ef741..318addec 100644
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32, timestep_embedding_f32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ARANGE_F32, arange_f32, true);
GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC, argsort_f32_i32_asc, true);
@@ -1145,6 +1147,7 @@ static bool ggml_metal_supports_op(const struct ggml_backend_metal_device_contex
@@ -1254,6 +1256,7 @@ static bool ggml_metal_supports_op(const struct ggml_backend_metal_device_contex
case GGML_OP_UPSCALE:
case GGML_OP_PAD:
case GGML_OP_PAD_REFLECT_1D:
@@ -238,7 +239,7 @@ index cd8ef741..318addec 100644
case GGML_OP_ARANGE:
case GGML_OP_TIMESTEP_EMBEDDING:
case GGML_OP_ARGSORT:
@@ -3348,6 +3351,36 @@ static void ggml_metal_encode_node(
@@ -3469,6 +3472,36 @@ static void ggml_metal_encode_node(
const int nth = MIN(1024, ne0);
@@ -276,10 +277,10 @@ index cd8ef741..318addec 100644
} break;
case GGML_OP_ARANGE:
diff --git a/ggml/src/ggml-metal/ggml-metal.metal b/ggml/src/ggml-metal/ggml-metal.metal
index 8ba43904..204c93e6 100644
index d092a169..f38909d0 100644
--- a/ggml/src/ggml-metal/ggml-metal.metal
+++ b/ggml/src/ggml-metal/ggml-metal.metal
@@ -2944,6 +2944,51 @@ kernel void kernel_pad_reflect_1d_f32(
@@ -2953,6 +2953,51 @@ kernel void kernel_pad_reflect_1d_f32(
}
}
@@ -332,10 +333,10 @@ index 8ba43904..204c93e6 100644
device char * dst,
constant int64_t & ne0,
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 2bbe5f48..7ffcd907 100644
index 7fc06724..635aa299 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -954,6 +954,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
@@ -962,6 +962,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
"UPSCALE",
"PAD",
"PAD_REFLECT_1D",
@@ -343,16 +344,16 @@ index 2bbe5f48..7ffcd907 100644
"ARANGE",
"TIMESTEP_EMBEDDING",
"ARGSORT",
@@ -987,7 +988,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
@@ -996,7 +997,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
"OPT_STEP_ADAMW",
};
-static_assert(GGML_OP_COUNT == 82, "GGML_OP_COUNT != 82");
+static_assert(GGML_OP_COUNT == 83, "GGML_OP_COUNT != 83");
-static_assert(GGML_OP_COUNT == 83, "GGML_OP_COUNT != 83");
+static_assert(GGML_OP_COUNT == 84, "GGML_OP_COUNT != 84");
static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
"none",
@@ -1050,6 +1051,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
@@ -1059,6 +1060,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
"upscale(x)",
"pad(x)",
"pad_reflect_1d(x)",
@@ -360,16 +361,16 @@ index 2bbe5f48..7ffcd907 100644
"arange(start, stop, step)",
"timestep_embedding(timesteps, dim, max_period)",
"argsort(x)",
@@ -1083,7 +1085,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
@@ -1093,7 +1095,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
"adamw(x)",
};
-static_assert(GGML_OP_COUNT == 82, "GGML_OP_COUNT != 82");
+static_assert(GGML_OP_COUNT == 83, "GGML_OP_COUNT != 83");
-static_assert(GGML_OP_COUNT == 83, "GGML_OP_COUNT != 83");
+static_assert(GGML_OP_COUNT == 84, "GGML_OP_COUNT != 84");
static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2");
@@ -4214,6 +4216,25 @@ struct ggml_tensor * ggml_pad_reflect_1d(
@@ -4225,6 +4227,25 @@ struct ggml_tensor * ggml_pad_reflect_1d(
return result;
}