matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-14 09:47:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update to commit de4c07f93 (#10655)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-05-12 12:17:26 -07:00
committed by GitHub
parent ad035ad595
commit 0cefd46f23
113 changed files with 8097 additions and 4383 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
llama/patches/0007-add-unpad-operator.patch
View File

@@ -18,7 +18,7 @@ adds the unpad operator to GGML
10 files changed, 223 insertions(+), 2 deletions(-)
diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index 1b8603e7..53ef31b2 100644
index e91dedf1..8dc107ba 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -489,6 +489,7 @@ extern "C" {
@@ -29,7 +29,7 @@ index 1b8603e7..53ef31b2 100644
GGML_OP_ARANGE,
GGML_OP_TIMESTEP_EMBEDDING,
GGML_OP_ARGSORT,
@@ -1777,6 +1778,15 @@ extern "C" {
@@ -1781,6 +1782,15 @@ extern "C" {
int p0,
int p1);
@@ -46,10 +46,10 @@ index 1b8603e7..53ef31b2 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 64405449..34624cca 100644
index a30e67f2..835e6495 100644
--- a/ggml/src/ggml-cpu/ggml-cpu.c
+++ b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -1964,6 +1964,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
@@ -1951,6 +1951,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
{
ggml_compute_forward_pad_reflect_1d(params, tensor);
} break;
@@ -60,7 +60,7 @@ index 64405449..34624cca 100644
case GGML_OP_ARANGE:
{
ggml_compute_forward_arange(params, tensor);
@@ -2287,6 +2291,7 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
@@ -2274,6 +2278,7 @@ 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:
@@ -69,10 +69,10 @@ index 64405449..34624cca 100644
case GGML_OP_TIMESTEP_EMBEDDING:
case GGML_OP_ARGSORT:
diff --git a/ggml/src/ggml-cpu/ops.cpp b/ggml/src/ggml-cpu/ops.cpp
index 7413192b..becdae07 100644
index 955fec59..1868a10c 100644
--- a/ggml/src/ggml-cpu/ops.cpp
+++ b/ggml/src/ggml-cpu/ops.cpp
@@ -6703,6 +6703,61 @@ void ggml_compute_forward_pad_reflect_1d(
@@ -6690,6 +6690,61 @@ void ggml_compute_forward_pad_reflect_1d(
}
}
@@ -147,10 +147,10 @@ index dc081b9e..a7125555 100644
void ggml_compute_forward_timestep_embedding(const struct ggml_compute_params * params, struct ggml_tensor * dst);
void ggml_compute_forward_argsort(const struct ggml_compute_params * params, struct ggml_tensor * dst);
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 04ce764e..491acccb 100644
index cb0d8528..6fe86674 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -2223,6 +2223,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
@@ -2238,6 +2238,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_PAD:
ggml_cuda_op_pad(ctx, dst);
break;
@@ -160,7 +160,7 @@ index 04ce764e..491acccb 100644
case GGML_OP_ARANGE:
ggml_cuda_op_arange(ctx, dst);
break;
@@ -3197,6 +3200,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
@@ -3212,6 +3215,7 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
case GGML_OP_UPSCALE:
return op->src[0]->type == GGML_TYPE_F32 && op->op_params[0] == GGML_SCALE_MODE_NEAREST;
case GGML_OP_PAD:
@@ -233,10 +233,10 @@ 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 425524d0..112abef6 100644
index 1b56f858..7641247e 100644
--- a/ggml/src/ggml-metal/ggml-metal.m
+++ b/ggml/src/ggml-metal/ggml-metal.m
@@ -341,6 +341,7 @@ static void ggml_backend_metal_device_rel(struct ggml_backend_metal_device_conte
@@ -347,6 +347,7 @@ 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,
@@ -244,7 +244,7 @@ index 425524d0..112abef6 100644
GGML_METAL_KERNEL_TYPE_ARANGE_F32,
GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32,
GGML_METAL_KERNEL_TYPE_ARGSORT_F32_I32_ASC,
@@ -1277,6 +1278,7 @@ @implementation GGMLMetalClass
@@ -1294,6 +1295,7 @@ @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);
@@ -252,7 +252,7 @@ index 425524d0..112abef6 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);
@@ -1647,6 +1649,7 @@ static bool ggml_metal_supports_op(const struct ggml_backend_metal_device_contex
@@ -1655,6 +1657,7 @@ static bool ggml_metal_supports_op(const struct ggml_backend_metal_device_contex
case GGML_OP_POOL_2D:
case GGML_OP_PAD:
case GGML_OP_PAD_REFLECT_1D:
@@ -260,7 +260,7 @@ index 425524d0..112abef6 100644
case GGML_OP_TIMESTEP_EMBEDDING:
case GGML_OP_ARGSORT:
case GGML_OP_LEAKY_RELU:
@@ -4047,6 +4050,36 @@ static bool ggml_metal_encode_node(
@@ -4184,6 +4187,36 @@ static bool ggml_metal_encode_node(
const int nth = MIN(1024, ne0);
@@ -298,10 +298,10 @@ index 425524d0..112abef6 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 9f4147e9..6ceb3cef 100644
index 9cfddf45..080a943b 100644
--- a/ggml/src/ggml-metal/ggml-metal.metal
+++ b/ggml/src/ggml-metal/ggml-metal.metal
@@ -2975,6 +2975,51 @@ kernel void kernel_pad_reflect_1d_f32(
@@ -3121,6 +3121,51 @@ kernel void kernel_pad_reflect_1d_f32(
}
}
@@ -354,7 +354,7 @@ index 9f4147e9..6ceb3cef 100644
device char * dst,
constant ggml_metal_kargs_arange & args,
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 7654ae17..3c57aff8 100644
index 8a654624..6b034d35 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -923,6 +923,7 @@ static const char * GGML_OP_NAME[GGML_OP_COUNT] = {
@@ -391,7 +391,7 @@ index 7654ae17..3c57aff8 100644
static_assert(GGML_OP_POOL_COUNT == 2, "GGML_OP_POOL_COUNT != 2");
@@ -4270,6 +4272,25 @@ struct ggml_tensor * ggml_pad_reflect_1d(
@@ -4274,6 +4276,25 @@ struct ggml_tensor * ggml_pad_reflect_1d(
return result;
}