Local AI Training & Chat

Dynamic Configuration

RAM: — MAX_LEN: — CPU: —
Read-only — values loaded from app.py at startup

Hardware (auto-detected)

Training Parameters (from app.py)

BASE_BATCH_SIZE
INITIAL_ACCUMULATION_MIN_STEPS
INITIAL_ACCUMULATION_MAX_STEPS
BASE_LEARNING_RATE
LOGGIN_STEPS
OPTIM
BASE_MAX_LEN
MAX_LEN_INCREMENT
MAX_LEN_CAP

🔧 Convert to GGUF

status: inactive
How to use: After merge, choose the format and click Convert. Output is saved in gguf_output/.
F16 = original size, full precision · Q8_0 = ~50% size, near-lossless quality

📄 Create Dataset

status: inactive
How to use: escolhe um ficheiro (PDF, TXT, DOCX, MD ou JSONL) → selecciona o Schema → clica Processar. The resulting JSONL file is saved in dataset/ and can be used directly for training.
Schema: text = text only (simplest) · lite = text + basic metadata · full = everything incl. tables · prompt/completion = instruction/response format for directed fine-tuning
Size of each excerpt (characters)
Overlap between excerpts
chars = fast · words = no cuts
basic = normal · aggressive = removes pagination
0 = no minimum size filter

Train AI

status: inactive
Training Progress 0%

Training Status / Logs

À espera do início do treino…

💬 Model Chat (Evaluation)

Chat inactive — load model first (Merge or Test Model).

Output Directory

Loading directory contents...
Expected endpoints: GET/POST /api/config, POST /api/train, POST /api/decide_merge, POST /api/load_lora_chat, POST /api/chat, POST /api/gguf/convert, GET /api/gguf/status, GET /api/gguf/dir, GET /api/gguf/download/<file>.