Skip to content

UPSTREAM PR #18795: Clean up t5 input builders - #900

Open
loci-dev wants to merge 2 commits into
mainfrom
upstream-PR18795-branch_gabe-l-hart-CleanUpT5InputBuilders
Open

UPSTREAM PR #18795: Clean up t5 input builders#900
loci-dev wants to merge 2 commits into
mainfrom
upstream-PR18795-branch_gabe-l-hart-CleanUpT5InputBuilders

Conversation

@loci-dev

@loci-dev loci-dev commented Jan 12, 2026

Copy link
Copy Markdown

Mirrored from ggml-org/llama.cpp#18795

Description

This PR removes some dead code in the graph input setters for T5 models and embedding models. It should be a functional No-Op.

(nice find @hansolosan)

@gabe-l-hart
fix: Remove unnecessary h loops where h was only ever 0
6674920 
Branch: CleanUpT5InputBuilders

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
@gabe-l-hart
fix: Remove unnecessary padding loop that is never hit anymore
ce97a44 
The upper bound used to use GGML_PAD(n_tokens, GGML_KQ_MASK_PAD), but was
removed in ggml-org/llama.cpp#17910 leaving the
loop dead.

Branch: CleanUpT5InputBuilders

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
@loci-review

loci-review Bot commented Jan 13, 2026

Copy link
Copy Markdown

Explore the complete analysis inside the Version Insights

I've successfully generated the summary report for your project. The report shows performance analysis for Pull Request #900 in the llama.cpp repository (owned by auroralabs-loci).

Key Highlights:

  1. Major Regression: The _S_max_size function shows a significant 55.79% increase in response time, indicating potential issues with memory allocation operations.

  2. Mixed Results: The operator= function shows a 9.80% increase in response time but a 97.86% increase in throughput.

  3. Minor Regressions: Several functions show 2-3% performance degradation in vocabulary processing, quantization, and vector assignment operations.

The report recommends prioritizing investigation of the memory allocation regression and reviewing the unique pointer assignment changes. Would you like me to provide more details on any specific aspect of this analysis?

@loci-dev
loci-dev force-pushed the main branch 26 times, most recently from d388dca to d36a5a9 Compare January 16, 2026 12:15
@loci-dev
loci-dev force-pushed the main branch 30 times, most recently from 0e2fcc8 to 5668a6a Compare January 24, 2026 07:09
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@loci-dev @gabe-l-hart