github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-22 21:51:37 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[PR #11090] [MERGED] New Memory Management #23977

New Issue
Closed
opened 2026-04-19 17:18:58 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-19 17:18:58 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/11090
Author: @jessegross
Created: 6/16/2025
Status: ✅ Merged
Merged: 8/14/2025
Merged by: @jessegross

Base: main ← Head: jessegross/memory

📝 Commits (1)

  • 76ed257 llm: New memory management

📊 Changes

26 files changed (+1855 additions, -895 deletions)

View changed files

📝 discover/amd_linux.go (+34 -31)
📝 envconfig/config.go (+3 -0)
📝 fs/ggml/ggml.go (+2 -0)
📝 llama/llama.go (+16 -0)
📝 llama/patches/0016-add-C-API-for-mtmd_input_text.patch (+0 -0)
➖ llama/patches/0016-temporary-prevent-rocm-cuda-mixed-loading.patch (+0 -32)
📝 llama/patches/0017-no-power-throttling-win32-with-gnuc.patch (+0 -0)
📝 llama/patches/0018-BF16-macos-version-guard.patch (+0 -0)
📝 llama/patches/0019-Enable-CUDA-Graphs-for-gemma3n.patch (+0 -0)
📝 llama/patches/0020-Disable-ggml-blas-on-macos-v13-and-older.patch (+0 -0)
📝 llama/patches/0021-fix-mtmd-audio.cpp-build-on-windows.patch (+0 -0)
📝 llama/patches/0022-ggml-No-alloc-mode.patch (+0 -0)
📝 llm/memory.go (+74 -20)
📝 llm/memory_test.go (+5 -5)
📝 llm/server.go (+868 -179)
📝 llm/server_test.go (+169 -0)
📝 ml/backend.go (+154 -8)
📝 ml/backend/ggml/ggml.go (+121 -103)
📝 ml/backend/ggml/ggml/src/ggml-backend-reg.cpp (+2 -10)
📝 runner/llamarunner/runner.go (+86 -60)

...and 6 more files

📄 Description

This changes the memory allocation strategy from upfront estimation to tracking actual allocations done by the engine and reacting to that. The goal is avoid issues caused by both under-estimation (crashing) and over-estimation (low performance due to under-utilized GPUs).

It is currently opt-in and can be enabled for models running on the Ollama engine by setting OLLAMA_NEW_ESTIMATES=1. Behavior in other cases is unchanged and will continue to use the existing estimates.

🔄 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/11090 **Author:** [@jessegross](https://github.com/jessegross) **Created:** 6/16/2025 **Status:** ✅ Merged **Merged:** 8/14/2025 **Merged by:** [@jessegross](https://github.com/jessegross) **Base:** `main` ← **Head:** `jessegross/memory` --- ### 📝 Commits (1) - [`76ed257`](https://github.com/ollama/ollama/commit/76ed257c0c4c3d2a622b50b4db020c4ff0ba4640) llm: New memory management ### 📊 Changes **26 files changed** (+1855 additions, -895 deletions) <details> <summary>View changed files</summary> 📝 `discover/amd_linux.go` (+34 -31) 📝 `envconfig/config.go` (+3 -0) 📝 `fs/ggml/ggml.go` (+2 -0) 📝 `llama/llama.go` (+16 -0) 📝 `llama/patches/0016-add-C-API-for-mtmd_input_text.patch` (+0 -0) ➖ `llama/patches/0016-temporary-prevent-rocm-cuda-mixed-loading.patch` (+0 -32) 📝 `llama/patches/0017-no-power-throttling-win32-with-gnuc.patch` (+0 -0) 📝 `llama/patches/0018-BF16-macos-version-guard.patch` (+0 -0) 📝 `llama/patches/0019-Enable-CUDA-Graphs-for-gemma3n.patch` (+0 -0) 📝 `llama/patches/0020-Disable-ggml-blas-on-macos-v13-and-older.patch` (+0 -0) 📝 `llama/patches/0021-fix-mtmd-audio.cpp-build-on-windows.patch` (+0 -0) 📝 `llama/patches/0022-ggml-No-alloc-mode.patch` (+0 -0) 📝 `llm/memory.go` (+74 -20) 📝 `llm/memory_test.go` (+5 -5) 📝 `llm/server.go` (+868 -179) 📝 `llm/server_test.go` (+169 -0) 📝 `ml/backend.go` (+154 -8) 📝 `ml/backend/ggml/ggml.go` (+121 -103) 📝 `ml/backend/ggml/ggml/src/ggml-backend-reg.cpp` (+2 -10) 📝 `runner/llamarunner/runner.go` (+86 -60) _...and 6 more files_ </details> ### 📄 Description This changes the memory allocation strategy from upfront estimation to tracking actual allocations done by the engine and reacting to that. The goal is avoid issues caused by both under-estimation (crashing) and over-estimation (low performance due to under-utilized GPUs). It is currently opt-in and can be enabled for models running on the Ollama engine by setting OLLAMA_NEW_ESTIMATES=1. Behavior in other cases is unchanged and will continue to use the existing estimates. --- <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 2026-04-19 17:18:58 -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#23977
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?