matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-14 01:37:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml: Free ggml_backend_buffer_t when releasing buffer

Browse Source 
When ggml_backend_buffer_free() is called, the device memory
is released but not all backends consistently release the actual
ggml_backend_buffer_t in system RAM, causing a memory leak.

Bug #10040
This commit is contained in:
Jesse Gross
2025-04-14 12:12:36 -07:00
committed by Jesse Gross
parent 0fe487e732
commit ccb7eb8135
6 changed files with 224 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
llama/patches/0008-add-unpad-operator.patch
View File

@@ -126,10 +126,10 @@ index 72325349..2f606d82 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 36165840..1adf08fa 100644
index 1b0d074b..c7a957c8 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -2198,6 +2198,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
@@ -2200,6 +2200,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
case GGML_OP_PAD:
ggml_cuda_op_pad(ctx, dst);
break;
@@ -139,7 +139,7 @@ index 36165840..1adf08fa 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
@@ -3199,6 +3202,7 @@ 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: