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 #15081] feat: autotune - automatic hardware-based performance optimization #61704

New Issue
Open
opened 2026-04-29 16:44:37 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-29 16:44:37 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/15081
Author: @GuilhermeP96
Created: 3/26/2026
Status: 🔄 Open

Base: main ← Head: main

📝 Commits (2)

  • 9d65fdf feat: add autotune package for automatic hardware-based performance optimization
  • f9ac5d5 docs: add autotune documentation to CLI reference and FAQ

📊 Changes

12 files changed (+1498 additions, -24 deletions)

View changed files

➕ autotune/README.md (+87 -0)
➕ autotune/api.go (+152 -0)
➕ autotune/apply.go (+82 -0)
➕ autotune/autotune_test.go (+332 -0)
➕ autotune/hardware.go (+120 -0)
➕ autotune/profile.go (+145 -0)
➕ autotune/tuner.go (+328 -0)
📝 cmd/cmd.go (+132 -1)
📝 docs/cli.mdx (+26 -0)
📝 docs/faq.mdx (+51 -0)
📝 envconfig/config.go (+27 -23)
📝 server/routes.go (+16 -0)

📄 Description

Summary

Adds an autotune package that automatically detects hardware (GPU VRAM, CPU cores, RAM) and applies optimal Ollama environment variables for best inference performance.

Changes

New package: autotune/

  • hardware.go — Detects GPU VRAM, CPU cores, total RAM
  • profile.go — Defines 5 performance profiles: balanced, speed, memory, multiuser, max
  • tuner.go — Selects the best profile based on detected hardware
  • apply.go — Applies profile settings as Ollama environment variables
  • api.go — HTTP handlers for the autotune API
  • autotune_test.go — Unit tests

API Endpoints

  • GET /api/autotune — Returns current autotune status and hardware info
  • GET /api/autotune/profiles — Lists all available profiles
  • POST /api/autotune — Applies a specific profile (body: {"profile": "speed"})

Integration Points

  • envconfig/ — Registered autotune environment variables
  • server/routes.go — Registered API route handlers
  • cmd/cmd.go — Added ollama autotune CLI commands (status, profiles, apply)

Documentation

  • docs/cli.mdx — Documented ollama autotune commands
  • docs/faq.mdx — Added autotune FAQ entry

Profiles

Profile Use Case Key Settings
balanced Default, general use Auto-scaled to hardware
speed Single-user, max throughput KEEP_ALIVE=30m, optimized threading
memory Low-VRAM GPUs (<6GB) Reduced context, conservative memory
multiuser Shared servers NUM_PARALLEL=4, shorter keep-alive
max Maximum performance, high-end GPUs Max context, aggressive settings

Testing

  • Unit tests included (go test ./autotune/)
  • Tested on: Intel Xeon E5-2676 v3 (24C/48T) + GTX 1660 SUPER 6GB + 56GB RAM
  • Built and validated on Windows with go build (v0.18.2-autotune rebased onto v0.18.4-rc0)

Motivation

Many users don't know the optimal environment variables for their hardware. Autotune eliminates guesswork by detecting the system capabilities and applying the best settings automatically, improving inference speed without manual configuration.

🔄 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/15081 **Author:** [@GuilhermeP96](https://github.com/GuilhermeP96) **Created:** 3/26/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (2) - [`9d65fdf`](https://github.com/ollama/ollama/commit/9d65fdf8f844b3518ec43a57fc9764d842e7db26) feat: add autotune package for automatic hardware-based performance optimization - [`f9ac5d5`](https://github.com/ollama/ollama/commit/f9ac5d5a073e0543c4c711a0ebbca856605e10e5) docs: add autotune documentation to CLI reference and FAQ ### 📊 Changes **12 files changed** (+1498 additions, -24 deletions) <details> <summary>View changed files</summary> ➕ `autotune/README.md` (+87 -0) ➕ `autotune/api.go` (+152 -0) ➕ `autotune/apply.go` (+82 -0) ➕ `autotune/autotune_test.go` (+332 -0) ➕ `autotune/hardware.go` (+120 -0) ➕ `autotune/profile.go` (+145 -0) ➕ `autotune/tuner.go` (+328 -0) 📝 `cmd/cmd.go` (+132 -1) 📝 `docs/cli.mdx` (+26 -0) 📝 `docs/faq.mdx` (+51 -0) 📝 `envconfig/config.go` (+27 -23) 📝 `server/routes.go` (+16 -0) </details> ### 📄 Description ## Summary Adds an **autotune** package that automatically detects hardware (GPU VRAM, CPU cores, RAM) and applies optimal Ollama environment variables for best inference performance. ## Changes ### New package: `autotune/` - **hardware.go** — Detects GPU VRAM, CPU cores, total RAM - **profile.go** — Defines 5 performance profiles: balanced, speed, memory, multiuser, max - **tuner.go** — Selects the best profile based on detected hardware - **apply.go** — Applies profile settings as Ollama environment variables - **api.go** — HTTP handlers for the autotune API - **autotune_test.go** — Unit tests ### API Endpoints - `GET /api/autotune` — Returns current autotune status and hardware info - `GET /api/autotune/profiles` — Lists all available profiles - `POST /api/autotune` — Applies a specific profile (body: `{"profile": "speed"}`) ### Integration Points - `envconfig/` — Registered autotune environment variables - `server/routes.go` — Registered API route handlers - `cmd/cmd.go` — Added `ollama autotune` CLI commands (status, profiles, apply) ### Documentation - `docs/cli.mdx` — Documented `ollama autotune` commands - `docs/faq.mdx` — Added autotune FAQ entry ## Profiles | Profile | Use Case | Key Settings | |---------|----------|-------------| | balanced | Default, general use | Auto-scaled to hardware | | speed | Single-user, max throughput | KEEP_ALIVE=30m, optimized threading | | memory | Low-VRAM GPUs (<6GB) | Reduced context, conservative memory | | multiuser | Shared servers | NUM_PARALLEL=4, shorter keep-alive | | max | Maximum performance, high-end GPUs | Max context, aggressive settings | ## Testing - Unit tests included (`go test ./autotune/`) - Tested on: Intel Xeon E5-2676 v3 (24C/48T) + GTX 1660 SUPER 6GB + 56GB RAM - Built and validated on Windows with `go build` (v0.18.2-autotune rebased onto v0.18.4-rc0) ## Motivation Many users don't know the optimal environment variables for their hardware. Autotune eliminates guesswork by detecting the system capabilities and applying the best settings automatically, improving inference speed without manual configuration. --- <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 2026-04-29 16:44:37 -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#61704
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?