github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #9987] [MERGED] Improve memory estimates for sliding window attention #18384

New Issue
Closed
opened 2026-04-16 06:33:32 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-16 06:33:32 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/9987
Author: @jessegross
Created: 3/25/2025
Status: ✅ Merged
Merged: 3/26/2025
Merged by: @jessegross

Base: main ← Head: jessegross/swa_estimation

📝 Commits (3)

  • 965e61e kvcache: Sliding window cache only needs a single batch total
  • 76628c3 llm: Fix debug logging for memory estimates
  • 8216285 ggml: Support heterogeneous KV cache layer sizes in memory estimation

📊 Changes

6 files changed (+52 additions, -31 deletions)

View changed files

📝 fs/ggml/ggml.go (+27 -12)
📝 kvcache/causal.go (+2 -2)
📝 llm/memory.go (+16 -10)
📝 llm/memory_test.go (+2 -2)
📝 llm/server.go (+1 -1)
📝 server/sched.go (+4 -4)

📄 Description

Recent optimizations for sliding window attention significantly reduced memory usage for models that use it. However, our memory estimates didn't reflect that, causing us to spill over to system memory unnecessarily in cases where GPU memory is constrained.

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/ollama/ollama/pull/9987 **Author:** [@jessegross](https://github.com/jessegross) **Created:** 3/25/2025 **Status:** ✅ Merged **Merged:** 3/26/2025 **Merged by:** [@jessegross](https://github.com/jessegross) **Base:** `main` ← **Head:** `jessegross/swa_estimation` --- ### 📝 Commits (3) - [`965e61e`](https://github.com/ollama/ollama/commit/965e61e28669d4f8071766ca803997419a323448) kvcache: Sliding window cache only needs a single batch total - [`76628c3`](https://github.com/ollama/ollama/commit/76628c3a21a1ba89454a1f59e5b1e84cba2c1a94) llm: Fix debug logging for memory estimates - [`8216285`](https://github.com/ollama/ollama/commit/82162856cddf1780f7d35d3653ab26dc3d74a3bd) ggml: Support heterogeneous KV cache layer sizes in memory estimation ### 📊 Changes **6 files changed** (+52 additions, -31 deletions) <details> <summary>View changed files</summary> 📝 `fs/ggml/ggml.go` (+27 -12) 📝 `kvcache/causal.go` (+2 -2) 📝 `llm/memory.go` (+16 -10) 📝 `llm/memory_test.go` (+2 -2) 📝 `llm/server.go` (+1 -1) 📝 `server/sched.go` (+4 -4) </details> ### 📄 Description Recent optimizations for sliding window attention significantly reduced memory usage for models that use it. However, our memory estimates didn't reflect that, causing us to spill over to system memory unnecessarily in cases where GPU memory is constrained. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
GiteaMirror added the pull-request label 2026-04-16 06:33:32 -05:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label pull-request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#18384
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?