Deepseek4: concat inputs to stay under limit - #26496
Closed
am17an wants to merge 1 commit into
Closed
Conversation
ggerganov
reviewed
Aug 3, 2026
Comment on lines
+935
to
+940
| inp.state_idxs = dsv4_build_input_1d(ctx, GGML_TYPE_I32, dsv4_comp_n_state_idxs(plan), prefix + "state_idxs"); | ||
| // Keep state_idxs as a direct node source so the scheduler can rotate input copies. | ||
| ggml_tensor * state_idxs = inp.state_idxs ? ggml_dup(ctx, inp.state_idxs) : nullptr; | ||
| if (state_idxs) { | ||
| ggml_set_name(state_idxs, (prefix + "state_idxs_cpy").c_str()); | ||
| } |
Member
There was a problem hiding this comment.
I don't understand the idea of this ggml_dup
Contributor
Author
There was a problem hiding this comment.
This is for pipeline paralleism to not override the views as they are not set with the INPUT flag
Member
|
Let's go with #22789 for now. I want to take a deeper look in the pipeline parallel problem that you mentioned, but atm I can't repro. It could be something that needs a fix in the scheduler and the |
Contributor
Author
|
Yeah I think it boils down the views don't have the flag set. So taking a view from an input might lead to a race using pipeline parallel. |
Contributor
Author
|
Fixed by #22789 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Overview
Currently the number of inputs to the graph are more than
GGML_SCHED_MAX_INPUTS, concat the inputs so that they stay under limit.Additional information
Requirements