#!/bin/bash
# BoccaccioAI - Fase 5: Evaluation and Export
# De Lauretis Tech
set -euo pipefail

echo "=== BoccaccioAI - Evaluation ==="

# Chat mode test (same template as training)
python -m src.inference.generate \
    --model-dir checkpoints/finetune \
    --tokenizer-path tokenizer/boccaccio-32k.json \
    --config-path configs/model.yaml \
    --mode chat \
    --prompt "Qual e' la capitale dell'Italia?" \
    --max-new-tokens 128 \
    --temperature 0.3

echo "---"

# Chat mode with system prompt
python -m src.inference.generate \
    --model-dir checkpoints/finetune \
    --tokenizer-path tokenizer/boccaccio-32k.json \
    --config-path configs/model.yaml \
    --mode chat \
    --system-prompt "Sei un assistente utile che risponde in italiano." \
    --prompt "Spiega la fotosintesi in modo semplice." \
    --max-new-tokens 256 \
    --temperature 0.5

echo "---"

# Free generation test (pretrain-style, no template)
python -m src.inference.generate \
    --model-dir checkpoints/finetune \
    --tokenizer-path tokenizer/boccaccio-32k.json \
    --config-path configs/model.yaml \
    --mode generate \
    --prompt "L'intelligenza artificiale in Italia" \
    --max-new-tokens 256 \
    --temperature 0.7

echo "=== Evaluation complete ==="