github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-22 13:42:25 -05:00
Code Issues 16.9k Packages Projects Releases 122 Wiki Activity

[PR #3241] [MERGED] update memory estimations for gpu offloading #57798

New Issue
Closed
opened 2026-04-29 12:31:39 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-04-29 12:31:39 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/3241
Author: @mxyng
Created: 3/19/2024
Status: ✅ Merged
Merged: 4/1/2024
Merged by: @mxyng

Base: main ← Head: mxyng/mem

📝 Commits (1)

  • 91b3e4d update memory calcualtions

📊 Changes

7 files changed (+121 additions, -85 deletions)

View changed files

📝 format/bytes.go (+16 -1)
📝 gpu/gpu.go (+11 -14)
📝 gpu/types.go (+3 -0)
📝 llm/dyn_ext_server.go (+2 -2)
📝 llm/ggml.go (+11 -0)
📝 llm/llm.go (+73 -63)
📝 server/routes.go (+5 -5)

📄 Description

take into account memory footprint of each layer

  1. replace percentage overhead with static overhead of 377 MiB for cuda and rocm
  2. add projector memory footprint to estimation
  3. add layer footprint to estimation on a per-layer basis, including output layers
  4. replace static kv memory footprint with pro-rated footprint based on how many layers to offload
  5. set minimum context length for multimodal models to 2048
  6. report memory requirements as structured logs

🔄 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/3241 **Author:** [@mxyng](https://github.com/mxyng) **Created:** 3/19/2024 **Status:** ✅ Merged **Merged:** 4/1/2024 **Merged by:** [@mxyng](https://github.com/mxyng) **Base:** `main` ← **Head:** `mxyng/mem` --- ### 📝 Commits (1) - [`91b3e4d`](https://github.com/ollama/ollama/commit/91b3e4d282fb8bcda65127b3bedb6f39dd4b7e41) update memory calcualtions ### 📊 Changes **7 files changed** (+121 additions, -85 deletions) <details> <summary>View changed files</summary> 📝 `format/bytes.go` (+16 -1) 📝 `gpu/gpu.go` (+11 -14) 📝 `gpu/types.go` (+3 -0) 📝 `llm/dyn_ext_server.go` (+2 -2) 📝 `llm/ggml.go` (+11 -0) 📝 `llm/llm.go` (+73 -63) 📝 `server/routes.go` (+5 -5) </details> ### 📄 Description take into account memory footprint of each layer 1. replace percentage overhead with static overhead of 377 MiB for cuda and rocm 2. add projector memory footprint to estimation 3. add layer footprint to estimation on a per-layer basis, including output layers 4. replace static kv memory footprint with pro-rated footprint based on how many layers to offload 5. set minimum context length for multimodal models to 2048 6. report memory requirements as structured logs --- <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 12:31:39 -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#57798
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?