fix: add Pascal CUDA architectures

[nanookclaw]

Summary

Closes #929.

This sets CMAKE_CUDA_ARCHITECTURES explicitly for the native llama.cpp CUDA build so the image includes Pascal targets sm_61 and sm_62. The list follows the issue's suggested architecture set and keeps the existing newer CUDA targets covered:

61;62;70;75;80;86;89;120;121

Without an explicit value, recent CUDA/CMake defaults can omit Pascal, which leaves GTX 10-series and similar cards without a compatible CUDA kernel and causes Docker Model Runner to fall back to CPU inference.

Validation

• git diff --check
• Confirmed the Dockerfile contains the architecture flag exactly once
• Confirmed the semicolon-separated CMake list remains a single argument when passed through the existing cmake-flags command-substitution pattern

I did not run a full CUDA image build locally because that requires pulling the NVIDIA CUDA builder image and compiling llama.cpp; there does not appear to be a focused lightweight Dockerfile flag test in this repo.

[gemini-code-assist]

Code Review

This pull request configures CUDA architectures in the Dockerfile for llamacpp. However, the specified architectures '120' and '121' are invalid and will cause compilation to fail, while the important Hopper architecture ('90') is missing. The reviewer recommends updating the list to use valid compute capabilities.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The CUDA architectures 120 and 121 are invalid. CUDA compute capabilities (architectures) do not go up to 120 or 121; these numbers likely confuse CUDA software versions (12.0, 12.1) with GPU architectures. Passing these invalid values will cause nvcc to fail during compilation with an error like nvcc fatal : Unsupported gpu architecture 'compute_120'.

Additionally, the list is missing Hopper (90), which is a highly important enterprise GPU architecture (e.g., NVIDIA H100).

Please update the list to use valid compute capabilities, such as 61;62;70;75;80;86;89;90.

    -DCMAKE_CUDA_ARCHITECTURES=61;62;70;75;80;86;89;90 \

References

1. Focus on correctness, safety, and maintainability — in that order. Critical issues include correctness bugs. ^(link)

[sourcery-ai]

Hey - I've left some high level feedback:

• Hard-coding CMAKE_CUDA_ARCHITECTURES in the Dockerfile makes it harder to support future GPU generations; consider wiring this through a build ARG or environment variable so the list can be overridden without editing the image definition.
• Since the semicolon-separated architectures are passed through shell and command substitution, it may be safer to either quote the -DCMAKE_CUDA_ARCHITECTURES=... argument or document this behavior near the cmake-flags usage to avoid accidental word-splitting changes in future edits.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- Hard-coding `CMAKE_CUDA_ARCHITECTURES` in the Dockerfile makes it harder to support future GPU generations; consider wiring this through a build ARG or environment variable so the list can be overridden without editing the image definition.
- Since the semicolon-separated architectures are passed through shell and command substitution, it may be safer to either quote the `-DCMAKE_CUDA_ARCHITECTURES=...` argument or document this behavior near the `cmake-flags` usage to avoid accidental word-splitting changes in future edits.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[ericcurtin]

Eventually we should use to using llama.cpp artefacts from llama.cpp directly... We will support whatever they deem suitable to support...

@ilopezluna PTAL