github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #12811] [MERGED] Enable op_offload to improve partial offload performance #19236

New Issue
Closed
opened 2026-04-16 07:01:40 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-16 07:01:40 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/12811
Author: @jessegross
Created: 10/29/2025
Status: ✅ Merged
Merged: 10/30/2025
Merged by: @jessegross

Base: main ← Head: jessegross/op_offload

📝 Commits (2)

  • dbb461c ollamarunner: Worst case batch for token generation
  • 921ba0b ggml: Enable op_offload to improve partial offload performance

📊 Changes

15 files changed (+422 additions, -135 deletions)

View changed files

➖ llama/patches/0019-Enable-CUDA-Graphs-for-gemma3n.patch (+0 -58)
➕ llama/patches/0019-ggml-Add-batch-size-hint.patch (+300 -0)
📝 llama/patches/0022-ggml-No-alloc-mode.patch (+20 -19)
📝 ml/backend.go (+5 -0)
📝 ml/backend/ggml/ggml.go (+17 -1)
📝 ml/backend/ggml/ggml/include/ggml-backend.h (+4 -1)
📝 ml/backend/ggml/ggml/src/ggml-backend-impl.h (+2 -2)
📝 ml/backend/ggml/ggml/src/ggml-backend.cpp (+13 -6)
📝 ml/backend/ggml/ggml/src/ggml-blas/ggml-blas.cpp (+2 -1)
📝 ml/backend/ggml/ggml/src/ggml-cpu/ggml-cpu.cpp (+3 -1)
📝 ml/backend/ggml/ggml/src/ggml-cuda/ggml-cuda.cu (+29 -37)
📝 ml/backend/ggml/ggml/src/ggml-metal/ggml-metal.cpp (+3 -1)
📝 ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp (+2 -1)
📝 runner/ollamarunner/multimodal.go (+3 -0)
📝 runner/ollamarunner/runner.go (+19 -7)

📄 Description

When a model is partially offloaded to system RAM, we can either do the calculations on the CPU or we can temporarily transfer the data to the GPU to do the calculations there. Small batches tend to be better on the CPU, large batches on the GPU.

The llamarunner used the GPU in most cases and the ollamarunner used the CPU. Although the ollamarunner saw an improvement in token generation performance, there was a large performance hit in prompt processing (3-10x).

There is an existing heuristic to dynamically switch between these two modes but in practice it doesn't have enough information to accurately make that decision. This adds authoritative data to make the check work to get the best of both worlds.

🔄 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/12811 **Author:** [@jessegross](https://github.com/jessegross) **Created:** 10/29/2025 **Status:** ✅ Merged **Merged:** 10/30/2025 **Merged by:** [@jessegross](https://github.com/jessegross) **Base:** `main` ← **Head:** `jessegross/op_offload` --- ### 📝 Commits (2) - [`dbb461c`](https://github.com/ollama/ollama/commit/dbb461cc7360e664fd21422e28174c6b3802ea85) ollamarunner: Worst case batch for token generation - [`921ba0b`](https://github.com/ollama/ollama/commit/921ba0bc848dc8d6499f164e78c42bfceba6048d) ggml: Enable op_offload to improve partial offload performance ### 📊 Changes **15 files changed** (+422 additions, -135 deletions) <details> <summary>View changed files</summary> ➖ `llama/patches/0019-Enable-CUDA-Graphs-for-gemma3n.patch` (+0 -58) ➕ `llama/patches/0019-ggml-Add-batch-size-hint.patch` (+300 -0) 📝 `llama/patches/0022-ggml-No-alloc-mode.patch` (+20 -19) 📝 `ml/backend.go` (+5 -0) 📝 `ml/backend/ggml/ggml.go` (+17 -1) 📝 `ml/backend/ggml/ggml/include/ggml-backend.h` (+4 -1) 📝 `ml/backend/ggml/ggml/src/ggml-backend-impl.h` (+2 -2) 📝 `ml/backend/ggml/ggml/src/ggml-backend.cpp` (+13 -6) 📝 `ml/backend/ggml/ggml/src/ggml-blas/ggml-blas.cpp` (+2 -1) 📝 `ml/backend/ggml/ggml/src/ggml-cpu/ggml-cpu.cpp` (+3 -1) 📝 `ml/backend/ggml/ggml/src/ggml-cuda/ggml-cuda.cu` (+29 -37) 📝 `ml/backend/ggml/ggml/src/ggml-metal/ggml-metal.cpp` (+3 -1) 📝 `ml/backend/ggml/ggml/src/ggml-vulkan/ggml-vulkan.cpp` (+2 -1) 📝 `runner/ollamarunner/multimodal.go` (+3 -0) 📝 `runner/ollamarunner/runner.go` (+19 -7) </details> ### 📄 Description When a model is partially offloaded to system RAM, we can either do the calculations on the CPU or we can temporarily transfer the data to the GPU to do the calculations there. Small batches tend to be better on the CPU, large batches on the GPU. The llamarunner used the GPU in most cases and the ollamarunner used the CPU. Although the ollamarunner saw an improvement in token generation performance, there was a large performance hit in prompt processing (3-10x). There is an existing heuristic to dynamically switch between these two modes but in practice it doesn't have enough information to accurately make that decision. This adds authoritative data to make the check work to get the best of both worlds. --- <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 07:01:40 -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#19236
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?