mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-22 14:53:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

GGML update to ec98e2002 (#13451)

Browse Source 
* Revert "add support for NVIDIA Nemotron 3 Nano"

This reverts commit e7d2ae9d69421012e9a8765c06a3fdf0e45b12f3.

* GGML update to 380b4c984

Remove MaskBatchPadding as GGML_KQ_MASK_PAD is no longer present (no
padding required)

* update to c45f89d55

* ec98e2002

solar pro needed more adjusting - needs verification

* review comments
This commit is contained in:
Daniel Hiltgen
2025-12-17 13:13:55 -08:00
committed by GitHub
parent 1c094038bc
commit 49a9c9ba6a
127 changed files with 8128 additions and 6710 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
llama/patches/0020-ggml-No-alloc-mode.patch
View File

@@ -10,13 +10,13 @@ must be recreated with no-alloc set to false before loading data.
---
ggml/include/ggml-backend.h | 1 +
ggml/src/ggml-backend-impl.h | 16 +++
ggml/src/ggml-backend.cpp | 72 +++++++++-
ggml/src/ggml-backend.cpp | 75 ++++++++++-
ggml/src/ggml-cuda/common.cuh | 62 ++++++++-
ggml/src/ggml-cuda/ggml-cuda.cu | 224 ++++++++++++++++++++++++++------
5 files changed, 331 insertions(+), 44 deletions(-)
5 files changed, 333 insertions(+), 45 deletions(-)
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
index 2763f2bd6..b3b5b356a 100644
index 93c95602d..dbbb61d9c 100644
--- a/ggml/include/ggml-backend.h
+++ b/ggml/include/ggml-backend.h
@@ -305,6 +305,7 @@ extern "C" {
@@ -75,13 +75,19 @@ index 0f5b03cef..7bdf9d81f 100644
struct ggml_backend {
diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 312ca873c..4092dfe8a 100644
index 498186a7c..7746e8b92 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -41,6 +41,19 @@ ggml_backend_buffer_t ggml_backend_buft_alloc_buffer(ggml_backend_buffer_type_t
@@ -36,11 +36,25 @@ const char * ggml_backend_buft_name(ggml_backend_buffer_type_t buft) {
}
ggml_backend_buffer_t ggml_backend_buft_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
- GGML_ASSERT(buft);
if (size == 0) {
// return a dummy buffer for zero-sized allocations
return ggml_backend_buffer_init(buft, {}, NULL, 0);
}
+
+ if (buft->no_alloc) {
+ ggml_backend_buffer_t buf;
+
@@ -95,10 +101,11 @@ index 312ca873c..4092dfe8a 100644
+ return buf;
+ }
+
GGML_ASSERT(buft);
+ GGML_ASSERT(buft);
return buft->iface.alloc_buffer(buft, size);
}
@@ -95,7 +108,8 @@ ggml_backend_buffer_t ggml_backend_buffer_init(
@@ -94,7 +108,8 @@ ggml_backend_buffer_t ggml_backend_buffer_init(
/* .buft = */ buft,
/* .context = */ context,
/* .size = */ size,
@@ -108,7 +115,7 @@ index 312ca873c..4092dfe8a 100644
};
return buffer;
@@ -127,6 +141,12 @@ void * ggml_backend_buffer_get_base(ggml_backend_buffer_t buffer) {
@@ -126,6 +141,12 @@ void * ggml_backend_buffer_get_base(ggml_backend_buffer_t buffer) {
return NULL;
}
@@ -118,10 +125,10 @@ index 312ca873c..4092dfe8a 100644
+ return (void *)ggml_backend_buffer_get_alignment(buffer);
+ }
+
void * base = buffer->iface.get_base(buffer);
GGML_ASSERT(base != NULL && "backend buffer base cannot be NULL");
@@ -731,6 +751,12 @@ struct ggml_backend_sched {
// FIXME JG: a multi_buffer has a non-zero size, according to the above comment get_base is not optional,
// I don't know whether the above comment is correct
if (!buffer->iface.get_base) {
@@ -736,6 +757,12 @@ struct ggml_backend_sched {
int debug_realloc;
int debug_graph_size;
int debug_prev_graph_size;
@@ -134,7 +141,7 @@ index 312ca873c..4092dfe8a 100644
};
#define hash_id(tensor) ggml_hash_find_or_insert(&sched->hash_set, tensor)
@@ -1630,6 +1656,17 @@ ggml_backend_sched_t ggml_backend_sched_new(
@@ -1635,6 +1662,17 @@ ggml_backend_sched_t ggml_backend_sched_new(
size_t graph_size,
bool parallel,
bool op_offload) {
@@ -152,7 +159,7 @@ index 312ca873c..4092dfe8a 100644
GGML_ASSERT(n_backends > 0);
GGML_ASSERT(n_backends <= GGML_SCHED_MAX_BACKENDS);
GGML_ASSERT(ggml_backend_dev_type(ggml_backend_get_device(backends[n_backends - 1])) == GGML_BACKEND_DEVICE_TYPE_CPU);
@@ -1682,11 +1719,14 @@ ggml_backend_sched_t ggml_backend_sched_new(
@@ -1687,11 +1725,14 @@ ggml_backend_sched_t ggml_backend_sched_new(
sched->events[b][c] = ggml_backend_event_new(backends[b]->device);
}
}
@@ -167,7 +174,7 @@ index 312ca873c..4092dfe8a 100644
ggml_backend_sched_reset(sched);
@@ -1701,6 +1741,10 @@ void ggml_backend_sched_free(ggml_backend_sched_t sched) {
@@ -1706,6 +1747,10 @@ void ggml_backend_sched_free(ggml_backend_sched_t sched) {
for (int c = 0; c < sched->n_copies; c++) {
ggml_backend_event_free(sched->events[b][c]);
}
@@ -178,7 +185,7 @@ index 312ca873c..4092dfe8a 100644
}
ggml_gallocr_free(sched->galloc);
ggml_free(sched->ctx);
@@ -1746,6 +1790,24 @@ bool ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph *
@@ -1765,6 +1810,24 @@ bool ggml_backend_sched_reserve(ggml_backend_sched_t sched, struct ggml_cgraph *
return false;
}
@@ -203,7 +210,7 @@ index 312ca873c..4092dfe8a 100644
ggml_backend_sched_reset(sched);
return true;
@@ -1851,7 +1913,13 @@ size_t ggml_backend_sched_get_attempted_buffer_size(ggml_backend_sched_t sched,
@@ -1870,7 +1933,13 @@ size_t ggml_backend_sched_get_attempted_buffer_size(ggml_backend_sched_t sched,
int backend_index = ggml_backend_sched_backend_id(sched, backend);
GGML_ASSERT(backend_index >= 0 && backend_index < sched->n_backends);
@@ -219,7 +226,7 @@ index 312ca873c..4092dfe8a 100644
void ggml_backend_sched_set_tensor_backend(ggml_backend_sched_t sched, struct ggml_tensor * node, ggml_backend_t backend) {
diff --git a/ggml/src/ggml-cuda/common.cuh b/ggml/src/ggml-cuda/common.cuh
index c4529f5d9..8b0fb5d42 100644
index 9fcb2f9fd..e800ee8f6 100644
--- a/ggml/src/ggml-cuda/common.cuh
+++ b/ggml/src/ggml-cuda/common.cuh
@@ -37,6 +37,41 @@
@@ -264,7 +271,7 @@ index c4529f5d9..8b0fb5d42 100644
#define STRINGIZE_IMPL(...) #__VA_ARGS__
#define STRINGIZE(...) STRINGIZE_IMPL(__VA_ARGS__)
@@ -938,6 +973,9 @@ struct ggml_cuda_pool {
@@ -941,6 +976,9 @@ struct ggml_cuda_pool {
virtual void * alloc(size_t size, size_t * actual_size) = 0;
virtual void free(void * ptr, size_t size) = 0;
@@ -274,7 +281,7 @@ index c4529f5d9..8b0fb5d42 100644
};
template<typename T>
@@ -1229,11 +1267,15 @@ struct ggml_backend_cuda_context {
@@ -1232,11 +1270,15 @@ struct ggml_backend_cuda_context {
// pool
std::unique_ptr<ggml_cuda_pool> pools[GGML_CUDA_MAX_DEVICES][GGML_CUDA_MAX_STREAMS];
@@ -292,7 +299,7 @@ index c4529f5d9..8b0fb5d42 100644
}
return *pools[device][curr_stream_no];
}
@@ -1241,6 +1283,22 @@ struct ggml_backend_cuda_context {
@@ -1244,6 +1286,22 @@ struct ggml_backend_cuda_context {
ggml_cuda_pool & pool() {
return pool(device);
}
@@ -316,7 +323,7 @@ index c4529f5d9..8b0fb5d42 100644
struct ggml_cuda_mm_fusion_args_host {
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index 17062697b..ede1d089a 100644
index 25548629d..eeaae3fe4 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -365,6 +365,8 @@ const ggml_cuda_device_info & ggml_cuda_info() {