[PR #12529] [MERGED] Bring back escape valve for llm libraries and fix Jetpack6 crash #19132

Closed
opened 2026-04-16 06:58:01 -05:00 by GiteaMirror · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/ollama/ollama/pull/12529
Author: @dhiltgen
Created: 10/7/2025
Status: Merged
Merged: 10/7/2025
Merged by: @dhiltgen

Base: mainHead: lib_override


📝 Commits (2)

  • bea5d17 Bring back escape valve for llm libraries
  • 0811750 Bypass extra discovery on jetpack systems

📊 Changes

3 files changed (+46 additions, -8 deletions)

View changed files

📝 discover/gpu.go (+34 -0)
📝 discover/runner.go (+10 -0)
📝 docs/troubleshooting.md (+2 -8)

📄 Description

If the new discovery logic picks the wrong library, this gives users the ability to force a specific one using the same pattern as before. This can also potentially speed up bootstrap discovery if any of the libraries take a long time to load and ultimately expose no devices. For example unsupported AMD iGPUS can sometimes take a while to discover and rule out.

On at least Jetpack6, cuda_v12 appears to expose the iGPU, but crashes later on in cublasInit so if we detect a Jetpack, short-circuit and use that variant.

Fixes #12521


🔄 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/12529 **Author:** [@dhiltgen](https://github.com/dhiltgen) **Created:** 10/7/2025 **Status:** ✅ Merged **Merged:** 10/7/2025 **Merged by:** [@dhiltgen](https://github.com/dhiltgen) **Base:** `main` ← **Head:** `lib_override` --- ### 📝 Commits (2) - [`bea5d17`](https://github.com/ollama/ollama/commit/bea5d1794aca101dcd9de9bb7df0c195da947807) Bring back escape valve for llm libraries - [`0811750`](https://github.com/ollama/ollama/commit/0811750300ed83f3f3b44e619ade1d476087f103) Bypass extra discovery on jetpack systems ### 📊 Changes **3 files changed** (+46 additions, -8 deletions) <details> <summary>View changed files</summary> 📝 `discover/gpu.go` (+34 -0) 📝 `discover/runner.go` (+10 -0) 📝 `docs/troubleshooting.md` (+2 -8) </details> ### 📄 Description If the new discovery logic picks the wrong library, this gives users the ability to force a specific one using the same pattern as before. This can also potentially speed up bootstrap discovery if any of the libraries take a long time to load and ultimately expose no devices. For example unsupported AMD iGPUS can sometimes take a while to discover and rule out. On at least Jetpack6, cuda_v12 appears to expose the iGPU, but crashes later on in cublasInit so if we detect a Jetpack, short-circuit and use that variant. Fixes #12521 --- <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 06:58:01 -05:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#19132