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 #5704] [CLOSED] Add TensorSplit option to runners and API #11891

New Issue
Closed
opened 2026-04-12 23:41:57 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-12 23:41:57 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/5704
Author: @NormalFishDev
Created: 7/15/2024
Status: ❌ Closed

Base: main ← Head: rework-tensor-split

📝 Commits (5)

📊 Changes

2 files changed (+28 additions, -24 deletions)

View changed files

📝 api/types.go (+25 -23)
📝 llm/server.go (+3 -1)

📄 Description

This pull request adds non-breaking functionality to Ollama function NewLlamaServer and adds a TensorSplit field to the Runner struct in api/types.go.

  • Add option to pass a tensor_split in "options" object for generate api to manually define how tensors should be split with llama.cpp.
  • Add conditional to check for manual tensor_split value in the Runner options to set the tensor_split parameter, defaults to estimate.TensorSplit when no manual tensor_split is passed (works exactly the same for existing applications).
  • Add TensorSplit to Runner struct in api/types.go for accessing the tensor_split value.

I have added this functionality because the team I work with is using Ollama on a server with 8 GPUs. We have run into issues with the automatic tensor splitting causing the server to crash due to unbalanced splitting between the GPUs (the splitting does not account for buffer VRAM usage after the model is loaded). Being able to specify manually the tensor splits has made it easier to implement Ollama on the server.

🔄 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/5704 **Author:** [@NormalFishDev](https://github.com/NormalFishDev) **Created:** 7/15/2024 **Status:** ❌ Closed **Base:** `main` ← **Head:** `rework-tensor-split` --- ### 📝 Commits (5) - [`d841d88`](https://github.com/ollama/ollama/commit/d841d885db24f834ab6473faa39814be68ab3cb0) Add TensorSplit option - [`9f45b69`](https://github.com/ollama/ollama/commit/9f45b696761cbdc110f7fcfb48b6ada0d1958a16) Fix TensorSplit Option - [`4fb8cc1`](https://github.com/ollama/ollama/commit/4fb8cc11ce3874e4840dd18ecb99c015eb588a66) Format changed files - [`070ac1d`](https://github.com/ollama/ollama/commit/070ac1d600d59a3895330210d3617b58fab4a1cc) Rewrite tensor split logic - [`3b100a9`](https://github.com/ollama/ollama/commit/3b100a91f838ec404e10ee7c5f9948403ce211f5) Format changed files with gofmt ### 📊 Changes **2 files changed** (+28 additions, -24 deletions) <details> <summary>View changed files</summary> 📝 `api/types.go` (+25 -23) 📝 `llm/server.go` (+3 -1) </details> ### 📄 Description This pull request adds non-breaking functionality to Ollama function `NewLlamaServer` and adds a `TensorSplit` field to the `Runner` struct in `api/types.go`. - Add option to pass a `tensor_split` in "options" object for generate api to manually define how tensors should be split with llama.cpp. - Add conditional to check for manual `tensor_split` value in the Runner options to set the `tensor_split` parameter, defaults to `estimate.TensorSplit` when no manual `tensor_split` is passed (works exactly the same for existing applications). - Add `TensorSplit` to `Runner` struct in `api/types.go` for accessing the `tensor_split` value. I have added this functionality because the team I work with is using Ollama on a server with 8 GPUs. We have run into issues with the automatic tensor splitting causing the server to crash due to unbalanced splitting between the GPUs (the splitting does not account for buffer VRAM usage after the model is loaded). Being able to specify manually the tensor splits has made it easier to implement Ollama on the server. --- <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 2026-04-12 23:41:57 -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#11891
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?