nimmo/nixos-config

Watch

Fork 0

Pin ollama-cuda and disable Electra dgpu Ollama #102

Merged

nimmo merged 1 commit from issue-99-trilium-vega-mcp into main

2026-07-27 14:01:46 +01:00

nimmo commented

2026-07-27 12:46:13 +01:00

Owner

Refs #101
Removing the ollama module from the dGPU specialisation and pinning ollama-cuda to avoid broken builds.

Refs #101 Removing the ollama module from the dGPU specialisation and pinning ollama-cuda to avoid broken builds.

nimmo added 1 commit

2026-07-27 12:46:18 +01:00

Pin ollama-cuda and disable Electra dgpu Ollama

Build and Push Attic Cache / test (pull_request) Successful in 52s

Details

Build and Push Attic Cache / build (pull_request) Has been skipped

Details

AI pull-request review / review (pull_request) Successful in 3m39s

Details

5bf271b520

Refs #101

nimmo commented

2026-07-27 12:49:58 +01:00

Author

Owner

Automated code review

Reviewed commit: 5bf271b520565b8cffdd512a6b199b948aae2be6

Verdict: Requires further work

Resolve the blocking or important findings and investigate failed deterministic checks before merging.

Overall assessment

The PR pins a dedicated nixpkgs revision for ollama-cuda and consolidates Electra's Ollama deployment to a single AMD/Vulkan instance. A global overlay in the common base module applies this pinned package to all hosts, which is architecturally suboptimal but functionally correct.

Blocking findings

None.

Important findings

Global nixpkgs overlay in common module forces unnecessary evaluation (modules/common/base.nix:38; high confidence)
nixpkgs.overlays = [
(final: prev: {
ollama-cuda = pinnedOllamaCudaPkgs.ollama-cuda;
})
];
Impact: The overlay is applied to every host importing the common module, forcing evaluation of the pinned nixpkgs-ollama-cuda input even on hosts that do not use or require CUDA acceleration. This increases flake evaluation time and couples hardware-specific package versions to base system configuration.
Suggested fix: Scope the overlay to host-specific configurations (e.g., hosts/electra/default.nix) or guard it with a conditional check for CUDA/NVIDIA hardware support before importing/evaluating the pinned input.

Suggestions

None.

Tests and validation

No configured deterministic checks.

Questions

Is the global overlay in modules/common/base.nix intentional to ensure all hosts use the pinned version if they ever enable CUDA, or should it be scoped strictly to Electra where the NVIDIA instance is disabled?

Review limitations

The diff does not show the full contents of modules/profiles/ollama.nix, so I cannot verify if the nvidia.enable = mkForce false override fully suppresses residual dependencies or if host/port/models fields are required by the module schema.
Automated validation gates (nix flake check, CI) are external to this review and their results are not available here.

Diff coverage

README.md: reviewed — included in a context-limited batch
flake.lock: reviewed — included in a context-limited batch
flake.nix: reviewed — included in a context-limited batch
hosts/electra/default.nix: reviewed — included in a context-limited batch
lib/auto-update-policy.nix: reviewed — included in a context-limited batch
modules/common/base.nix: reviewed — included in a context-limited batch

Review metadata

Reviewer: forgejo-ai-review v1
Model: qwen3.6:35b-a3b-q8_0
Guidance loaded: .forgejo/ai-review.md, .forgejo/ai-review.yml, AGENTS.md, README.md
Commits considered: 1
Previous automated review: none found
Findings: 1

## Automated code review Reviewed commit: `5bf271b520565b8cffdd512a6b199b948aae2be6` ## Verdict: Requires further work Resolve the blocking or important findings and investigate failed deterministic checks before merging. ### Overall assessment The PR pins a dedicated nixpkgs revision for ollama-cuda and consolidates Electra's Ollama deployment to a single AMD/Vulkan instance. A global overlay in the common base module applies this pinned package to all hosts, which is architecturally suboptimal but functionally correct. ### Blocking findings None. ### Important findings - **Global nixpkgs overlay in common module forces unnecessary evaluation** (modules/common/base.nix:38; high confidence) nixpkgs.overlays = [ (final: prev: { ollama-cuda = pinnedOllamaCudaPkgs.ollama-cuda; }) ]; Impact: The overlay is applied to every host importing the common module, forcing evaluation of the pinned nixpkgs-ollama-cuda input even on hosts that do not use or require CUDA acceleration. This increases flake evaluation time and couples hardware-specific package versions to base system configuration. Suggested fix: Scope the overlay to host-specific configurations (e.g., hosts/electra/default.nix) or guard it with a conditional check for CUDA/NVIDIA hardware support before importing/evaluating the pinned input. ### Suggestions None. ### Tests and validation - No configured deterministic checks. ### Questions - Is the global overlay in modules/common/base.nix intentional to ensure all hosts use the pinned version if they ever enable CUDA, or should it be scoped strictly to Electra where the NVIDIA instance is disabled? ### Review limitations - The diff does not show the full contents of modules/profiles/ollama.nix, so I cannot verify if the nvidia.enable = mkForce false override fully suppresses residual dependencies or if host/port/models fields are required by the module schema. - Automated validation gates (nix flake check, CI) are external to this review and their results are not available here. ### Diff coverage - `README.md`: **reviewed** — included in a context-limited batch - `flake.lock`: **reviewed** — included in a context-limited batch - `flake.nix`: **reviewed** — included in a context-limited batch - `hosts/electra/default.nix`: **reviewed** — included in a context-limited batch - `lib/auto-update-policy.nix`: **reviewed** — included in a context-limited batch - `modules/common/base.nix`: **reviewed** — included in a context-limited batch <details><summary>Review metadata</summary> - Reviewer: forgejo-ai-review v1 - Model: qwen3.6:35b-a3b-q8_0 - Guidance loaded: .forgejo/ai-review.md, .forgejo/ai-review.yml, AGENTS.md, README.md - Commits considered: 1 - Previous automated review: none found - Findings: 1 </details>

nimmo merged commit 9a97590a58 into main

2026-07-27 14:01:46 +01:00

nimmo deleted branch issue-99-trilium-vega-mcp

2026-07-27 14:01:47 +01:00

nimmo referenced this pull request from a commit

2026-07-27 14:01:47 +01:00

Merge pull request 'Pin ollama-cuda and disable Electra dgpu Ollama' (#102) from issue-99-trilium-vega-mcp into main