ADFA-6321 | Add Qwen2.5-Coder 1.5B Instruct to Get AI Models

[hal-eisen-adfa]

Adds Qwen2.5-Coder 1.5B Instruct (Q4_K_M) to the Get AI Models catalog. Jira: ADFA-6321

Every model currently in the catalog is a general chat model. This one is trained on code, so it is the one to offer learners who need offline help with Kotlin and Java.

Entry

• Source: Qwen's official GGUF repo, Qwen/Qwen2.5-Coder-1.5B-Instruct-GGUF, with the URL pinned to commit f86cb2c1fa58255f8052cc32aeede1b7482d4361
• File: qwen2.5-coder-1.5b-instruct-q4_k_m.gguf, 1,117,320,768 bytes. SHA-256 cc324af070c2ecbfd324a30884d2f951a7ff756aba85cb811a6ec436933bb046
• Licence: Apache-2.0 on both the base model and the GGUF repo. Neither is access-gated.
• Context: 32,768 tokens (max_position_embeddings)
• minRamBytes: 4 GiB. This is my call: the file is the largest in the catalog, and the ~0.8 GB entries use 3 GiB. The plugin only displays this number; it never checks it.
• behaviouralGatesVerified: false, the same as every other entry

Gates (docs/CURATION.md)

I checked these against the Hugging Face API as the curation doc describes:

• Gate 1: it is an instruct model.
• Gate 2: Apache-2.0 on both repos.
• Gate 5: the chat template is present and includes the tool-call format.
• Gate 6: single file. A HEAD request to the pinned URL returned Content-Length and X-Linked-Etag matching the catalogued size and SHA-256.
• Gate 7: 32k context.

Gates 3 and 4 have not been tested, as for every other entry.

Docs

• CURATION.md: added a table row and a note crediting ADFA-6321
• README.md: now says 7 models, and no longer says every file is under 1 GB
• get-ai-models.html: added a table row; the licence count is now six Apache-2.0 entries

Testing

• ../../gradlew test: 31 tests passed, including the 7 in CatalogLoaderTest, which check the shipped models.json
• Clean assemblePlugin build: the packaged models.json contains the new entry, and the build records revision 6a3127633de8 with no uncommitted changes
• The .cgp is pushed to a Samsung SM-S906U. It has not been installed or exercised in Code On The Go yet, and the 1 GB download has not been run on a device.

Not changed

• The plugin version stays at 1.1.0 / versionCode 2. Bump it if this should ship as a new release.
• The help page's existing "What ships in version 1.0.0" heading is out of date. I left it alone.