github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #14887] [MERGED] mlxrunner: share KV cache across conversations with common prefixes #20167

New Issue
Closed
opened 2026-04-16 07:29:03 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-16 07:29:03 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/14887
Author: @jessegross
Created: 3/16/2026
Status: ✅ Merged
Merged: 3/18/2026
Merged by: @jessegross

Base: main ← Head: jessegross/mlx-trie

📝 Commits (2)

  • 18456ba mlxrunner: fix Slice(0, 0) returning full dimension instead of empty
  • 2d9b8c3 mlxrunner: share KV cache across conversations with common prefixes

📊 Changes

12 files changed (+2761 additions, -180 deletions)

View changed files

📝 x/mlxrunner/cache.go (+484 -101)
📝 x/mlxrunner/cache/cache.go (+243 -39)
➕ x/mlxrunner/cache/cache_test.go (+271 -0)
📝 x/mlxrunner/cache/recurrent.go (+57 -29)
➕ x/mlxrunner/cache/recurrent_test.go (+44 -0)
➕ x/mlxrunner/cache_test.go (+859 -0)
➕ x/mlxrunner/cache_trie.go (+296 -0)
➕ x/mlxrunner/cache_trie_test.go (+455 -0)
📝 x/mlxrunner/mlx/mlx.go (+4 -0)
📝 x/mlxrunner/mlx/slice.go (+24 -8)
📝 x/mlxrunner/pipeline.go (+23 -2)
📝 x/mlxrunner/sample/sample.go (+1 -1)

📄 Description

Enable multiple conversations to reuse cached computations when they share token prefixes (e.g. the same system prompt). A prefix trie tracks shared regions so switching between conversations only recomputes tokens that diverge. Inactive conversation state is paged from active GPU memory to other memory and restored on demand, with LRU eviction to keep memory usage bounded.

🔄 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/14887 **Author:** [@jessegross](https://github.com/jessegross) **Created:** 3/16/2026 **Status:** ✅ Merged **Merged:** 3/18/2026 **Merged by:** [@jessegross](https://github.com/jessegross) **Base:** `main` ← **Head:** `jessegross/mlx-trie` --- ### 📝 Commits (2) - [`18456ba`](https://github.com/ollama/ollama/commit/18456ba2011c896a647d3534aa200561c475f06d) mlxrunner: fix Slice(0, 0) returning full dimension instead of empty - [`2d9b8c3`](https://github.com/ollama/ollama/commit/2d9b8c333ae5209c922ad0bf3b934fbb49c88ab7) mlxrunner: share KV cache across conversations with common prefixes ### 📊 Changes **12 files changed** (+2761 additions, -180 deletions) <details> <summary>View changed files</summary> 📝 `x/mlxrunner/cache.go` (+484 -101) 📝 `x/mlxrunner/cache/cache.go` (+243 -39) ➕ `x/mlxrunner/cache/cache_test.go` (+271 -0) 📝 `x/mlxrunner/cache/recurrent.go` (+57 -29) ➕ `x/mlxrunner/cache/recurrent_test.go` (+44 -0) ➕ `x/mlxrunner/cache_test.go` (+859 -0) ➕ `x/mlxrunner/cache_trie.go` (+296 -0) ➕ `x/mlxrunner/cache_trie_test.go` (+455 -0) 📝 `x/mlxrunner/mlx/mlx.go` (+4 -0) 📝 `x/mlxrunner/mlx/slice.go` (+24 -8) 📝 `x/mlxrunner/pipeline.go` (+23 -2) 📝 `x/mlxrunner/sample/sample.go` (+1 -1) </details> ### 📄 Description Enable multiple conversations to reuse cached computations when they share token prefixes (e.g. the same system prompt). A prefix trie tracks shared regions so switching between conversations only recomputes tokens that diverge. Inactive conversation state is paged from active GPU memory to other memory and restored on demand, with LRU eviction to keep memory usage bounded. --- <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 07:29:03 -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#20167
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?