mirror of
https://github.com/ollama/ollama.git
synced 2025-11-11 10:57:36 +01:00
Llama cpp bump (df1b612): granite docling / mamba2 optimizations / multimodal encoding fixes (#12552)
* feat: Bump llama.cpp to df1b612 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(mtmd): Correctly encode text chunks during mtmd tokenization There can be text chunks that appear interspersed with the image embeddings that contain template delimiter tokens for some models. These need to be correctly translated to text tokens. Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * tests: Use MtmdChunk in image_test Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * style: Fix unnecessary conversion linting Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix(ggml): Revert changes to ggml_hip.cpp These changes were done largely by our code assistant and are likely wrong Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Revert changes in mem_nvml.cpp Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update sync point to 1deee0 This brings in several more optimization commits and model support for EmbeddingGemma Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Update patches for 1deee0 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: sync for bump to 1deee0 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: Bad patch updates with errant `+` Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * feat: Bump llama.cpp/ggml to 7049736 Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> * fix: format-patches after latest bump Branch: LlamaCPPBump-GraniteDocling Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> --------- Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
This commit is contained in:
Gabe Goodhart
GitHub
@@ -16,10 +16,10 @@ must be recreated with no-alloc set to false before loading data.
|
||||
5 files changed, 310 insertions(+), 44 deletions(-)
|
||||
|
||||
diff --git a/ggml/include/ggml-backend.h b/ggml/include/ggml-backend.h
|
||||
index 48777212..d4352663 100644
|
||||
index 229bf387..1ff53ed0 100644
|
||||
--- a/ggml/include/ggml-backend.h
|
||||
+++ b/ggml/include/ggml-backend.h
|
||||
@@ -303,6 +303,7 @@ extern "C" {
|
||||
@@ -305,6 +305,7 @@ extern "C" {
|
||||
|
||||
// Initialize a backend scheduler, backends with low index are given priority over backends with high index
|
||||
GGML_API ggml_backend_sched_t ggml_backend_sched_new(ggml_backend_t * backends, ggml_backend_buffer_type_t * bufts, int n_backends, size_t graph_size, bool parallel, bool op_offload);
|
||||
@@ -28,7 +28,7 @@ index 48777212..d4352663 100644
|
||||
|
||||
// Initialize backend buffers from a measure graph
|
||||
diff --git a/ggml/src/ggml-backend-impl.h b/ggml/src/ggml-backend-impl.h
|
||||
index 07784d6f..869dc07d 100644
|
||||
index 6792ba98..3c3f22fc 100644
|
||||
--- a/ggml/src/ggml-backend-impl.h
|
||||
+++ b/ggml/src/ggml-backend-impl.h
|
||||
@@ -26,12 +26,17 @@ extern "C" {
|
||||
@@ -218,7 +218,7 @@ index cb2b9956..6ef5eeaf 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 c4246b65..448badf0 100644
|
||||
index e0abde54..28d6bcd7 100644
|
||||
--- a/ggml/src/ggml-cuda/common.cuh
|
||||
+++ b/ggml/src/ggml-cuda/common.cuh
|
||||
@@ -35,6 +35,31 @@
|
||||
@@ -253,7 +253,7 @@ index c4246b65..448badf0 100644
|
||||
#define STRINGIZE_IMPL(...) #__VA_ARGS__
|
||||
#define STRINGIZE(...) STRINGIZE_IMPL(__VA_ARGS__)
|
||||
|
||||
@@ -880,6 +905,9 @@ struct ggml_cuda_pool {
|
||||
@@ -856,6 +881,9 @@ struct ggml_cuda_pool {
|
||||
|
||||
virtual void * alloc(size_t size, size_t * actual_size) = 0;
|
||||
virtual void free(void * ptr, size_t size) = 0;
|
||||
@@ -263,7 +263,7 @@ index c4246b65..448badf0 100644
|
||||
};
|
||||
|
||||
template<typename T>
|
||||
@@ -1023,11 +1051,11 @@ struct ggml_backend_cuda_context {
|
||||
@@ -999,11 +1027,11 @@ struct ggml_backend_cuda_context {
|
||||
// pool
|
||||
std::unique_ptr<ggml_cuda_pool> pools[GGML_CUDA_MAX_DEVICES];
|
||||
|
||||
@@ -277,7 +277,7 @@ index c4246b65..448badf0 100644
|
||||
}
|
||||
return *pools[device];
|
||||
}
|
||||
@@ -1035,4 +1063,20 @@ struct ggml_backend_cuda_context {
|
||||
@@ -1011,4 +1039,20 @@ struct ggml_backend_cuda_context {
|
||||
ggml_cuda_pool & pool() {
|
||||
return pool(device);
|
||||
}
|
||||
@@ -299,7 +299,7 @@ index c4246b65..448badf0 100644
|
||||
+ }
|
||||
};
|
||||
diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
|
||||
index e51c5035..d324bc68 100644
|
||||
index 827e3205..811462c7 100644
|
||||
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
|
||||
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
|
||||
@@ -350,6 +350,8 @@ const ggml_cuda_device_info & ggml_cuda_info() {
|
||||
@@ -540,7 +540,7 @@ index e51c5035..d324bc68 100644
|
||||
};
|
||||
|
||||
ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device) {
|
||||
@@ -3008,6 +3070,7 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph, int node_idx,
|
||||
@@ -3011,6 +3073,7 @@ static bool ggml_cuda_can_fuse(const struct ggml_cgraph * cgraph, int node_idx,
|
||||
|
||||
static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx, ggml_cgraph * cgraph,
|
||||
bool & graph_evaluated_or_captured, bool & use_cuda_graph, bool & cuda_graph_update_required) {
|
||||
@@ -548,7 +548,7 @@ index e51c5035..d324bc68 100644
|
||||
// flag used to determine whether it is an integrated_gpu
|
||||
const bool integrated = ggml_cuda_info().devices[cuda_ctx->device].integrated;
|
||||
|
||||
@@ -3023,6 +3086,11 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
|
||||
@@ -3026,6 +3089,11 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
|
||||
continue;
|
||||
}
|
||||
|
||||
@@ -560,7 +560,7 @@ index e51c5035..d324bc68 100644
|
||||
static bool disable_fusion = (getenv("GGML_CUDA_DISABLE_FUSION") != nullptr);
|
||||
if (!disable_fusion) {
|
||||
|
||||
@@ -3149,6 +3217,7 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
|
||||
@@ -3152,6 +3220,7 @@ static void evaluate_and_capture_cuda_graph(ggml_backend_cuda_context * cuda_ctx
|
||||
|
||||
static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, ggml_cgraph * cgraph) {
|
||||
ggml_backend_cuda_context * cuda_ctx = (ggml_backend_cuda_context *)backend->context;
|
||||
@@ -568,7 +568,7 @@ index e51c5035..d324bc68 100644
|
||||
|
||||
ggml_cuda_set_device(cuda_ctx->device);
|
||||
|
||||
@@ -3228,6 +3297,71 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
|
||||
@@ -3231,6 +3300,71 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend,
|
||||
return GGML_STATUS_SUCCESS;
|
||||
}
|
||||
|
||||
@@ -640,7 +640,7 @@ index e51c5035..d324bc68 100644
|
||||
static void ggml_backend_cuda_event_record(ggml_backend_t backend, ggml_backend_event_t event) {
|
||||
ggml_backend_cuda_context * cuda_ctx = (ggml_backend_cuda_context *)backend->context;
|
||||
|
||||
@@ -3268,6 +3402,9 @@ static const ggml_backend_i ggml_backend_cuda_interface = {
|
||||
@@ -3271,6 +3405,9 @@ static const ggml_backend_i ggml_backend_cuda_interface = {
|
||||
/* .event_record = */ ggml_backend_cuda_event_record,
|
||||
/* .event_wait = */ ggml_backend_cuda_event_wait,
|
||||
/* .graph_optimize = */ NULL,
|
||||
|
||||
Reference in New Issue
Block a user