File size: 1,407 Bytes
120474f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 | {
"name": "kuclab_v1",
"version": "1.0",
"public": true,
"samples": 28321,
"instruction_rows": 28222,
"messages_rows": 99,
"text_rows": 0,
"identity_name": "KucLab V1.0",
"founder": "Jaroslav Kučera",
"trained_on": "2026-07-26",
"base_model_target": "google/gemma-2-9b-it (preferred) or google/gemma-2-9b",
"ollama_name": "kuclab-v1",
"built_at": "2026-07-26T13:50:59.241361+00:00",
"sources": {
"code:CodeAlpaca-20k": 11000,
"kuclab_v06_reuse": 6000,
"wiki_history_en": 4000,
"wiki_history_cs": 4000,
"kuclab_stack": 1100,
"czech_code_chat": 800,
"kuclab_pro": 1200,
"tools_json_v1": 94,
"personality_v1": 47,
"code_seed_v1": 20,
"public_qa_v1": 8,
"anti_refusal_v1": 15,
"history_seed_v1": 22,
"fluency_v1": 10,
"multiturn_v1": 5
},
"goals": [
"First public KucLab flagship",
"Czech + English fluency",
"Strong software engineering",
"History-focused knowledge",
"Anti-overrefusal / no identity collapse",
"Portable JSON tool calling",
"Cyber direct product voice"
],
"train_tips": [
"Base: google/gemma-2-9b-it + QLoRA on L4 24GB",
"dataset_format: alpaca (auto-detects messages)",
"max_seq_length: 2048",
"epochs: 1.5–2.0, lr: 1e-4 QLoRA, lora_r: 64",
"identity_repeat: 1",
"After train: run eval_prompts.jsonl smoke tests before public announce"
]
} |