github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[PR #15122] runner: Remove CGO engines, use llama-server exclusively for GGML models #77333

New Issue
Open
opened 2026-05-05 10:00:26 -05:00 by GiteaMirror · 0 comments
GiteaMirror commented 2026-05-05 10:00:26 -05:00
Owner
Copy Link

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/15122
Author: @dhiltgen
Created: 3/28/2026
Status: 🔄 Open

Base: main ← Head: llama-runner

📝 Commits (5)

  • 4d6d6dd broad lint fixes to sidestep CI scope glitch
  • 869e724 runner: Remove CGO engines, use llama-server exclusively for GGML models
  • 417e731 llama/compat: load Ollama-format GGUFs in llama-server
  • dfd8d88 server: remove embeddings compat redirect
  • 7f166a8 runner: iteration...

📊 Changes

973 files changed (+12081 additions, -426187 deletions)

View changed files

📝 .github/workflows/release.yaml (+66 -16)
📝 .github/workflows/test.yaml (+25 -23)
📝 CMakeLists.txt (+10 -148)
📝 CMakePresets.json (+2 -140)
📝 Dockerfile (+131 -81)
➕ LLAMA_CPP_VERSION (+1 -0)
➖ Makefile.sync (+0 -76)
📝 app/ollama.iss (+0 -48)
📝 cmd/cmd.go (+38 -6)
📝 cmd/tui/signin_test.go (+0 -1)
📝 convert/convert.go (+51 -2)
📝 convert/convert_deepseekocr.go (+112 -13)
📝 convert/convert_gemma3.go (+225 -38)
📝 convert/convert_gemma3n.go (+72 -4)
📝 convert/convert_gemma4.go (+175 -58)
📝 convert/convert_glm4moelite.go (+39 -30)
📝 convert/convert_glmocr.go (+165 -136)
📝 convert/convert_gptoss.go (+26 -19)
📝 convert/convert_qwen25vl.go (+140 -16)
📝 convert/convert_qwen3vl.go (+154 -17)

...and 80 more files

📄 Description

Marking draft as this is still WIP.

Remove the vendored GGML and llama.cpp backend, CGO runner, GGML-based Go model implementations, and sample. llama-server (built from upstream llama.cpp via FetchContent) is now the sole inference engine for GGUF-based models. (Safetensor based models continue to run on the new MLX engine.) This allows us to more rapidly pick up new capabilities from llama.cpp as they come out.

On windows this now requires recent AMD driver versions to support ROCm v7 as llama.cpp currently does not support building against v6.

The conversion code for models is still WIP. For models where the existing library version is incompatible with llama.cpp, new models are pushed to dhiltgen/xxx, with temporary routing logic in this branch to translate library/xxx -> dhiltgen/xxx. e.g. ollama run gpt-oss:20b will temporarily map to dhiltgen/gpt-oss:20b which is llama.cpp compatible. Before merging the namespace routing logic will be removed and compatible models will be pushed to the library with selection logic to ensure the correct versions are used.

🔄 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/15122 **Author:** [@dhiltgen](https://github.com/dhiltgen) **Created:** 3/28/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `llama-runner` --- ### 📝 Commits (5) - [`4d6d6dd`](https://github.com/ollama/ollama/commit/4d6d6dd7cb0f1bba10da84a8e1f9d3d68c890b77) broad lint fixes to sidestep CI scope glitch - [`869e724`](https://github.com/ollama/ollama/commit/869e724922a8631e09a0bfb0801dc14036985c56) runner: Remove CGO engines, use llama-server exclusively for GGML models - [`417e731`](https://github.com/ollama/ollama/commit/417e731726538f0376d2b3b731ca766cbbda7eb0) llama/compat: load Ollama-format GGUFs in llama-server - [`dfd8d88`](https://github.com/ollama/ollama/commit/dfd8d88fccac5728ede0f8ed6710054f34151011) server: remove embeddings compat redirect - [`7f166a8`](https://github.com/ollama/ollama/commit/7f166a8b8b05aeb3cc7d73b285a40d01a9d82179) runner: iteration... ### 📊 Changes **973 files changed** (+12081 additions, -426187 deletions) <details> <summary>View changed files</summary> 📝 `.github/workflows/release.yaml` (+66 -16) 📝 `.github/workflows/test.yaml` (+25 -23) 📝 `CMakeLists.txt` (+10 -148) 📝 `CMakePresets.json` (+2 -140) 📝 `Dockerfile` (+131 -81) ➕ `LLAMA_CPP_VERSION` (+1 -0) ➖ `Makefile.sync` (+0 -76) 📝 `app/ollama.iss` (+0 -48) 📝 `cmd/cmd.go` (+38 -6) 📝 `cmd/tui/signin_test.go` (+0 -1) 📝 `convert/convert.go` (+51 -2) 📝 `convert/convert_deepseekocr.go` (+112 -13) 📝 `convert/convert_gemma3.go` (+225 -38) 📝 `convert/convert_gemma3n.go` (+72 -4) 📝 `convert/convert_gemma4.go` (+175 -58) 📝 `convert/convert_glm4moelite.go` (+39 -30) 📝 `convert/convert_glmocr.go` (+165 -136) 📝 `convert/convert_gptoss.go` (+26 -19) 📝 `convert/convert_qwen25vl.go` (+140 -16) 📝 `convert/convert_qwen3vl.go` (+154 -17) _...and 80 more files_ </details> ### 📄 Description Marking draft as this is still WIP. Remove the vendored GGML and llama.cpp backend, CGO runner, GGML-based Go model implementations, and sample. llama-server (built from upstream llama.cpp via FetchContent) is now the sole inference engine for GGUF-based models. (Safetensor based models continue to run on the new MLX engine.) This allows us to more rapidly pick up new capabilities from llama.cpp as they come out. On windows this now requires recent AMD driver versions to support ROCm v7 as llama.cpp currently does not support building against v6. The conversion code for models is still WIP. For models where the existing library version is incompatible with llama.cpp, new models are pushed to dhiltgen/xxx, with temporary routing logic in this branch to translate library/xxx -> dhiltgen/xxx. e.g. `ollama run gpt-oss:20b` will temporarily map to `dhiltgen/gpt-oss:20b` which is llama.cpp compatible. Before merging the namespace routing logic will be removed and compatible models will be pushed to the library with selection logic to ensure the correct versions are used. --- <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 2026-05-05 10:00:26 -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#77333
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?