Knowledge Base

Atlassian Insights & Tutorials

Real Atlassian solutions to real problems — no fluff, no SEO spam.

Articles 2

goose compaction: my 27B re-read the file 95% of the time — quoting it in the summary didn't help

Our swarm workers kept re-reading files they had already read. Context compaction was summarising the tool output that held the file. Pasting the file into the summary does not fix it; returning the last turns verbatim does. Measured three ways on the same 27B.

Mihai PerdumAug 11, 202612 min read

ArticleAI CodingLocal AI

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

Multi-token prediction is merged in llama.cpp and still an open PR in mlx-lm. I measured both on an M3 Ultra with the same model. Every default MTP setting was slower than no MTP at all, and the runtime that deletes the MTP head outright is still the fastest thing on the box.

Mihai PerdumAug 10, 202615 min read

Atlassian Insights & Tutorials

Articles 2

goose compaction: my 27B re-read the file 95% of the time — quoting it in the summary didn't help

More Articles

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

goose compaction: my 27B re-read the file 95% of the time — quoting it in the summary didn't help

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

Atlassian Insights & Tutorials

Articles 2

goose compaction: my 27B re-read the file 95% of the time — quoting it in the summary didn't help

More Articles

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

goose compaction: my 27B re-read the file 95% of the time — quoting it in the summary didn't help

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x