Skip to content

Retire SmolLM-360M and rename tabby-pico → tabby-nano-2#355

Merged
FuJacob merged 1 commit into
mainfrom
chore/retire-smollm-360m-and-rename-pico-to-nano-2
May 28, 2026
Merged

Retire SmolLM-360M and rename tabby-pico → tabby-nano-2#355
FuJacob merged 1 commit into
mainfrom
chore/retire-smollm-360m-and-rename-pico-to-nano-2

Conversation

@FuJacob
Copy link
Copy Markdown
Owner

@FuJacob FuJacob commented May 28, 2026

Summary

Drop SmolLM-360M-Instruct.Q8_0.gguf from the curated catalog and rename the remaining small SmolLM2 entry from tabby-pico-1 to tabby-nano-2. The 360M is no longer offered as a downloadable model, no longer in the bootstrap priority list, and no longer mapped to a product label; the 135M (SmolLM2-135M-Instruct-q8_0.gguf) becomes the smallest curated option.

Validation

swiftlint lint --quiet Cotabby/Models/LlamaRuntimeModels.swift
# exit 0

xcodebuild -project Cotabby.xcodeproj -scheme Cotabby -destination 'platform=macOS' build
# ** BUILD SUCCEEDED **

Local full-test execution still hits the documented Team ID code-signing mismatch on this machine. CI tests run cleanly against the merge ref.

Linked issues

None.

Risk / rollout notes

  • Existing users on the 135M keep working without action: settings persist the GGUF filename (actualModelName: String { filename }), not the display name, so the rename is purely a label change in the menu.
  • Existing users on the 360M also keep working: the locator picks any GGUF that exists on disk, and RuntimeModelCatalog.displayName(for:) falls back to the raw filename for entries it does not recognize, so a previously-downloaded 360M shows up as SmolLM-360M-Instruct.Q8_0.gguf and remains selectable.
  • New users no longer see the 360M offered for download. The bootstrap priority list moves the 135M up one slot when none of the larger models are present.
  • No structural project.yml edit; only file contents changed, so the XcodeGen check should remain green.

Greptile Summary

Removes SmolLM-360M-Instruct.Q8_0.gguf from the curated catalog and renames the 135M SmolLM2 entry's display alias from tabby-pico-1 to tabby-nano-2. All three catalog touch-points (displayName, downloadableModels, preferredModelNames) are updated consistently, the README count is corrected from five to four, and no stale references remain anywhere in the codebase.

  • LlamaRuntimeModels.swift: removes the 360M DownloadableRuntimeModel, drops it from the bootstrap priority list, and updates the displayName switch — the 135M entry is now the smallest curated option under the tabby-nano-2 label.
  • README.md: removes the 360M table row and reflects the tabby-nano-2 rename, keeping documentation in sync with the catalog.

Confidence Score: 5/5

Safe to merge — catalog changes are minimal, consistent across all three touch-points, and backward-compatible for existing users who have either model on disk.

All three locations that reference the retired 360M model are cleaned up together. The 135M rename is a display-label-only change since settings persist the raw filename. No stale string references remain in tests, source, or docs.

No files require special attention. The only follow-up worth considering is adding a test assertion for the new tabby-nano-2 mapping in ModelAndPresentationValueTests.swift.

Important Files Changed

Filename Overview
Cotabby/Models/LlamaRuntimeModels.swift Removes the SmolLM-360M catalog entry from displayName, downloadableModels, and preferredModelNames; renames the 135M alias from tabby-pico-1 to tabby-nano-2. All three touch-points are consistent and no stale references remain.
README.md Updates model count from five to four, removes the SmolLM-360M row, and renames tabby-pico-1 to tabby-nano-2 — matches the Swift catalog changes exactly.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User selects model] --> B{RuntimeModelCatalog.displayName}
    B -->|SmolLM2-135M-Instruct-q8_0.gguf| C[tabby-nano-2 ✅]
    B -->|Qwen3-0.6B-Q4_K_M.gguf| D[tabby-fast-1]
    B -->|gemma-4-E2B-it-Q4_K_M.gguf| E[tabby-balanced-1]
    B -->|gemma-4-E4B-it-Q4_K_M.gguf| F[tabby-max-1]
    B -->|SmolLM-360M-Instruct.Q8_0.gguf removed| G[raw filename fallback]
    B -->|unknown GGUF| G

    H[Bootstrap preferredModelNames] --> I[gemma-4-E4B]
    I --> J[gemma-4-E2B]
    J --> K[Qwen3-0.6B]
    K --> L[SmolLM2-135M ✅ smallest curated]
Loading

Reviews (1): Last reviewed commit: "Retire SmolLM-360M from the catalog and ..." | Re-trigger Greptile

Drop SmolLM-360M-Instruct.Q8_0.gguf from the curated catalog: it is no
longer offered as a downloadable model, no longer in the bootstrap
priority list, and no longer mapped to a display name. The 135M
SmolLM2 (previously surfaced as `tabby-pico-1`) takes its place as
the smallest curated option and is renamed `tabby-nano-2`. User
settings store the GGUF filename rather than the display name, so
existing selections of the 135M continue to load — they just show up
under the new label.

Users who had downloaded the 360M will continue to be able to select
it: the locator picks any existing GGUF on disk, and
`RuntimeModelCatalog.displayName(for:)` falls back to the raw
filename for entries it does not know about.

README's model table is updated to match.
@FuJacob FuJacob merged commit b86f64a into main May 28, 2026
4 checks passed
@FuJacob FuJacob deleted the chore/retire-smollm-360m-and-rename-pico-to-nano-2 branch May 28, 2026 10:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant