serve — /v1/models returns model ID with precision suffix, breaking OpenAI-compatible clients

[Novatoris]

Describe the bug
/v1/models endpoint returns model ID with precision suffix (e.g. qualcomm/Qwen3-4B-Instruct-2507:W4A16), but using this exact ID in /v1/chat/completions fails with SDKError(Invalid input parameters or handle), quantization 'W4A16' not found for model 'qualcomm/Qwen3-4B-Instruct-2507'. The request only succeeds when the suffix is omitted. This breaks any OpenAI-compatible client that auto-populates the model ID from /v1/models (confirmed with AnythingLLM).

To Reproduce

1. Pull a QAIRT model: geniex pull ai-hub-models/Qwen3-4B-Instruct-2507
2. Start server: geniex serve --host 127.0.0.1:18181 --compute npu
3. Query /v1/models:

Invoke-RestMethod -Uri "http://127.0.0.1:18181/v1/models" -Method Get

Returns qualcomm/Qwen3-4B-Instruct-2507:W4A16

4. Use the returned ID in a chat request:

Invoke-RestMethod -Uri "http://127.0.0.1:18181/v1/chat/completions" -Method Post -ContentType "application/json" -Body '{"model": "qualcomm/Qwen3-4B-Instruct-2507:W4A16", "messages": [{"role": "user", "content": "Say hello"}]}'

5. See error: SDKError(Invalid input parameters or handle), quantization 'W4A16' not found

Expected behavior
/v1/models should return a model ID that works directly with /v1/chat/completions. Either the precision suffix should be stripped from the models list, or the completions endpoint should accept it.

Workaround
Clear the Precision field in geniex.json:

$json = Get-Content "C:\LocalAI\geniex\models\models\qualcomm\Qwen3-4B-Instruct-2507\geniex.json" | ConvertFrom-Json
$json.Precision = ""
$json | ConvertTo-Json -Depth 10 | Set-Content "C:\LocalAI\geniex\models\models\qualcomm\Qwen3-4B-Instruct-2507\geniex.json"

Desktop:

• OS: Windows 11 ARM64
• Device: Snapdragon X Plus X1P42100
• GenieX CLI Version: v0.3.17
• QAIRT Runtime Version: v2.45.0.260326
• LlamaCPP Runtime Hash: ae9291e
• Model: ai-hub-models/Qwen3-4B-Instruct-2507 (W4A16, QAIRT runtime)
• AnythingLLM Version: v1.15.0 (used for OpenAI-compatible client testing)

[mansiverma897993]

Novatoris Root-caused and fixed in PR #1247.

The bug: /v1/models advertises the top-level precision ("name:W4A16"), but the resolver only knew the internal "N/A" key so the server's own ID failed on echo-back. The PR makes the resolver accept the advertised precision as an alias, so the ID now round-trips directly. Your geniex.json workaround is no longer needed.

Verified with 5 new unit tests + full model-manager suite (171 passing). Along with #1246, your AnythingLLM flow should now fully work would appreciate a confirmation once a patched build is out.

Tried my best to fix it the right way hoping this gets reviewed and merged. Thanks !! Zack Li (@zhiyuan8) Alan Zhu (@alanzhuly) Mengsheng Wu (@mengshengwu)