Skip to content

log more info when metal fails#6987

Merged
ggerganov merged 1 commit intoggml-org:masterfrom
bakkot:messages-for-metal-failures
Apr 30, 2024
Merged

log more info when metal fails#6987
ggerganov merged 1 commit intoggml-org:masterfrom
bakkot:messages-for-metal-failures

Conversation

@bakkot
Copy link
Copy Markdown
Contributor

@bakkot bakkot commented Apr 29, 2024

Messages codes are documented here.

Now if you try to use too much memory, instead of just

ggml_metal_graph_compute: command buffer 6 failed with status 5

you get

ggml_metal_graph_compute: command buffer 6 failed with status 5
out of memory

which is much nicer!

The "status 5" thing is kind of misleading, since that status is just "error" - the other statuses shouldn't be reachable after the earlier waitUntilCompleted, if I'm reading the docs right. Happy to tweak that part too while I'm here if you'd ike.

@ggerganov ggerganov merged commit a68a1e7 into ggml-org:master Apr 30, 2024
@slaren
Copy link
Copy Markdown
Member

slaren commented Apr 30, 2024

It looks like this is causing issues: #7008 #7009. Can this be fixed or should this be reverted for now?

@bakkot
Copy link
Copy Markdown
Contributor Author

bakkot commented Apr 30, 2024

@slaren #7010

@bakkot bakkot deleted the messages-for-metal-failures branch April 30, 2024 14:34
Seunghhon pushed a commit to Seunghhon/llama.cpp that referenced this pull request Apr 26, 2026
phuongncn pushed a commit to phuongncn/llama.cpp-gx10-dgx-sparks-deepseekv4 that referenced this pull request Apr 28, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants