Summary

Collapses the Open Source model catalog to a single, coherent four-tier ladder so users stop
seeing five names spread across two families (tabby-2-mini/base/pro plus a parallel
tabby-2-gemma-mini/-pro sub-brand). The new lineup is nano / mini / base / pro:

Display names are derived from the GGUF filename (the persisted state is the filename), so the
relabel needs no re-download and changes no one's selected model. All four download URLs were
verified to resolve (HTTP 200) before committing.

Onboarding now reads nano (Quick) / base (Everyday, Recommended) / pro (Powerful), and
tabby-2-base (gemma-4-E2B) is the default load.

Validation

xcodebuild ... test ... CODE_SIGNING_ALLOWED=NO CODE_SIGNING_REQUIRED=NO \
  -only-testing:CotabbyTests/RuntimeAndInputModelValueTests \
  -only-testing:CotabbyTests/OnboardingTemplateRecommenderTests
# ** TEST SUCCEEDED **  Executed 17 tests, 0 failures
#   RuntimeAndInputModelValueTests        4/4  (catalog displayName mappings)
#   OnboardingTemplateRecommenderTests   13/13 (tier -> GGUF: nano / base / pro)

swiftlint lint --quiet   # exit 0 (clean)

App-hosted tests run unsigned (the default-signed bundle hits a Team ID mismatch on this machine).

Linked issues

None filed. Follow-up to the base-model migration; addresses recurring confusion from the
five-names-two-families catalog.

Risk / rollout notes

• Relabel, not a re-download. displayName is derived from the filename and the persisted state
  is the filename, so an existing user's installed model keeps working and just shows its new label.
• Qwen3.5-4B-Base retired. Removed from the download list and preferred-load order. An
  already-installed copy still loads and falls back to its raw filename (existing retired-model
  behavior, covered by a test).
• Default + Everyday are now tabby-2-base (gemma-4-E2B, ~4.5 GB). This makes the recommended
  onboarding download larger than before (~1.4 GB Qwen 2B previously). Intentional per the requested
  lineup. Quick stays nano (~0.8 GB); Powerful is pro (gemma-4-E4B, ~5.0 GB).
• Naming caveat carried over: base is both a tier name and the technical term for these
  non-instruct models. Kept per the requested lineup.

Greptile Summary

This PR collapses the five-model Open Source catalog into a clean four-tier ladder (nano / mini / base / pro), retiring the tabby-2-gemma-* sub-brand and the dropped Qwen 4B, and rewires onboarding templates to the new filenames.

• Catalog relabeling: displayName switch in LlamaRuntimeModels.swift updated to the new four-tier names; Qwen 4B removed from downloadableModels; display-name and retired-model fallback tests extended to cover all four active tiers.
• Onboarding rewiring: .everyday now installs gemma-4-E2B (4.5 GB, up from 1.4 GB) and .powerful now installs gemma-4-E4B (5.0 GB, previously the broken dropped-4B path); test assertions updated accordingly.
• preferredModelNames reordered: gemma-4-E2B.i1-Q6_K.gguf moved to position 0, making it the auto-loaded model for any user who has it installed — despite the PR description's explicit note that this change was intentionally deferred.

Confidence Score: 4/5

Safe to merge for new users and single-model installs; users with both Qwen 2B and Gemma E2B installed will silently switch to the larger model on next launch.

The PR description explicitly deferred promoting Gemma E2B to the default load due to the 1.4 GB to 4.5 GB size jump, yet preferredModelNames and the README both implement that promotion. Any user who had the old tabby-2-gemma-mini installed alongside the old tabby-2-base will experience a silent auto-load change.

Cotabby/Models/LlamaRuntimeModels.swift and README.md need to be reconciled — either the preferredModelNames order should revert to keep Qwen 2B first, or the PR description should be updated to acknowledge this as an intentional change.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    subgraph Old["Old catalog (5 models)"]
        O1["tabby-2-mini\nQwen3.5-0.8B (~0.8 GB)"]
        O2["tabby-2-base\nQwen3.5-2B (~1.4 GB)"]
        O3["tabby-2-pro\nQwen3.5-4B (~2.6 GB) - DROPPED"]
        O4["tabby-2-gemma-mini\ngemma-4-E2B (~4.5 GB)"]
        O5["tabby-2-gemma-pro\ngemma-4-E4B (~5.0 GB)"]
    end
    subgraph New["New catalog (4 models)"]
        N1["tabby-2-nano\nQwen3.5-0.8B (~0.8 GB)"]
        N2["tabby-2-mini\nQwen3.5-2B (~1.4 GB)"]
        N3["tabby-2-base\ngemma-4-E2B (~4.5 GB)"]
        N4["tabby-2-pro\ngemma-4-E4B (~5.0 GB)"]
    end
    subgraph Onboarding["Onboarding tiers to GGUF"]
        QK["Quick"] --> N1
        EV["Everyday 1.4GB to 4.5GB"] --> N3
        PW["Powerful"] --> N4
    end
    O1 --> N1
    O2 --> N2
    O3 -. dropped .-> N3
    O4 --> N3
    O5 --> N4

_{Reviews (2): Last reviewed commit: "Make tabby-2-base the default and the Ev..." | Re-trigger Greptile}

Greptile also left 1 inline comment on this PR.