vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'rd/concedo' into concedo_experimental

Browse source
This commit is contained in:
Concedo 2023-09-12 20:17:02 +08:00
commit d9f5793d22
1 changed files with 16 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

15
ggml-metal.m
View file

@ -1259,7 +1259,22 @@ void ggml_metal_graph_compute(
MTLCommandBufferStatus status = (MTLCommandBufferStatus) [ctx->command_buffers[i] status];
if (status != MTLCommandBufferStatusCompleted) {
if (status == MTLCommandBufferStatusError) {
// Check Metal error code
NSError *error = (MTLCommandBufferError) [ctx->command_buffers[i] error];
int mtl_error_code = [error code];
if (([error domain] == MTLCommandBufferErrorDomain) && ([error code] == MTLCommandBufferErrorOutOfMemory)) {
metal_printf("%s: command buffer %d failed with status MTLCommandBufferStatus.error (5) and error code \
MTLCommandBufferError.outOfMemory (8)\n");
printf("Metal ran out of memory. Maybe try a smaller context size, or a smaller (more coarsely quantized) model, \
preferably one under the recommended max working set size, or else fall back to running on CPU only.\n");
} else {
metal_printf("%s: command buffer %d failed with status MTLCommandBufferStatus.error (5) and error code %d\n",
__func__, i, mtl_error_code);
}
} else {
metal_printf("%s: command buffer %d failed with status %lu\n", __func__, i, status);
}
GGML_ASSERT(false);
}
}