From 11d89fba7bbf4f4ea2cada17470ccd214d471e85 Mon Sep 17 00:00:00 2001
From: gavinbee <29419542+gavinbee@users.noreply.github.com>
Date: Sat, 30 May 2026 00:14:14 -0400
Subject: [PATCH] Document how to downgrade Ollama on Windows

The troubleshooting doc told users to "downgrade to 0.12.x" but not how.
Add a section covering winget --version (with the catalog-pruning and
won't-downgrade-over-newer caveats), the GitHub-release OllamaSetup.exe
fallback, and a macOS/Linux note.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
---
 docs/troubleshooting.md | 46 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 46 insertions(+)

diff --git a/docs/troubleshooting.md b/docs/troubleshooting.md
index b398228..4902941 100644
--- a/docs/troubleshooting.md
+++ b/docs/troubleshooting.md
@@ -36,6 +36,52 @@ ollama --version
 
 If it reports 0.13.x or newer and you hit either symptom above on an NVIDIA card, the 0.12.x downgrade is the current remedy.
 
+## Downgrading Ollama to 0.12.x
+
+### Windows
+
+The install scripts use `winget`, and `winget` *can* pin a version:
+
+```
+winget install Ollama.Ollama --version 0.12.11
+```
+
+Two caveats:
+
+- **The version must still be in the `winget` catalog.** Ollama ships
+  frequently and old manifests get pruned, so a given 0.12.x may no longer
+  be offered. List what's actually available with:
+  ```
+  winget show Ollama.Ollama --versions
+  ```
+- **`winget` won't silently downgrade over a newer install.** If a newer
+  version is already present, add `--force`, or uninstall first
+  (`winget uninstall Ollama.Ollama`).
+
+If the version you want isn't in the catalog (the common case for a
+specific older release), use the **GitHub release installer** instead —
+it's version-pinned and always available:
+
+1. Open <https://github.com/ollama/ollama/releases>.
+2. Find the 0.12.x release you want (0.12.11 is known-good on an 8 GB
+   RTX-class card).
+3. Download `OllamaSetup.exe` from that release's assets and run it. It
+   installs over the existing version.
+
+After installing, confirm with `ollama --version` (expect `0.12.11`).
+On a good version the daemon log shows the model GPU-offloading —
+e.g. `offloaded 37/37 layers to GPU` — rather than a `100% CPU` fallback.
+
+### macOS / Linux
+
+Re-run the official installer pinned to a version, or download the
+matching release asset from <https://github.com/ollama/ollama/releases>.
+On Linux the install script honors a version:
+
+```
+curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION=0.12.11 sh
+```
+
 ## Ollama not found / daemon won't start
 
 See [installation.md](installation.md#troubleshooting-install-issues).