feat: Variet Engine v1.0 + 5-model tuning complete

Phase 01 (LLM Tuning):
- Gemma4 26B: 74.65 t/s (fast)
- Qwen 35B: 61.62 t/s (balanced)
- Gemma4 31B: 16.0 t/s (deep-coder)
- Qwen 27B: 16.7 t/s (deep-logic)
- Qwen 122B: 8.95 t/s (ultra, GPU 1 only)

Phase 02 (API Engine):
- FastAPI reverse proxy on port 8000
- /engine/switch hot-swap with 503 protection
- config/engine_models.json as single source of truth
- Replaced 4 individual .bat files with unified engine

File cleanup:
- scripts/ 85 files -> 9 + _archive/
- Root .bat files -> _archive/
This commit is contained in:
Variet-Worker
2026-04-07 18:08:58 +09:00
parent 7c7a899fd5
commit c111b3a9b0
414 changed files with 3402 additions and 68598 deletions

View File

@@ -0,0 +1,68 @@
[
{
"name": "Qwen 27B - 256K 극한 (q4_0, ub=512)",
"status": "Success",
"vram": [
"0, 11120 MiB, 12288 MiB",
"1, 11081 MiB, 12288 MiB"
],
"tests": {
"code": {
"time": 17.97,
"tokens": 300,
"tps": 16.7,
"res": "..."
},
"logical": {
"time": 18.01,
"tokens": 300,
"tps": 16.65,
"res": "..."
}
}
},
{
"name": "Gemma 31B - 128K 확장 (q4_0)",
"status": "Success",
"vram": [
"0, 11437 MiB, 12288 MiB",
"1, 11259 MiB, 12288 MiB"
],
"tests": {
"code": {
"time": 18.75,
"tokens": 300,
"tps": 16.0,
"res": "```python\nfrom typing import List, Any\n\ndef merge_sorted_lists(list1: List[Any], list2: List[Any]) -> List[Any]:\n \"\"\"\n Merges two sorted lists i..."
},
"logical": {
"time": 18.79,
"tokens": 300,
"tps": 15.97,
"res": "..."
}
}
},
{
"name": "Gemma 31B - 192K 극한 (q4_0)",
"status": "Success",
"vram": [
"0, 11888 MiB, 12288 MiB",
"1, 11754 MiB, 12288 MiB"
],
"tests": {
"code": {
"time": 18.69,
"tokens": 300,
"tps": 16.05,
"res": "```python\nfrom typing import List, Any\n\ndef merge_sorted_lists(list1: List[Any], list2: List[Any]) -> List[Any]:\n \"\"\"\n Merges two sorted lists i..."
},
"logical": {
"time": 18.77,
"tokens": 300,
"tps": 15.98,
"res": "..."
}
}
}
]