Faz2: scripts/FAZ2_START.sh
Browse files- scripts/FAZ2_START.sh +28 -0
scripts/FAZ2_START.sh
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/bin/bash
|
| 2 |
+
# MeshAI Faz 2 — TRELLIS/Hunyuan LoRA fine-tune (A100 VM)
|
| 3 |
+
# Cloud Shell -> SSH sonrasi tek yapistir.
|
| 4 |
+
set -e
|
| 5 |
+
cd ~/MeshAI_Cloud
|
| 6 |
+
|
| 7 |
+
export HF_TOKEN="${HF_TOKEN:?HF_TOKEN yok — once: export HF_TOKEN='hf_...'}"
|
| 8 |
+
export HF_REPO=HayrettinIscan/MeshAI-Base-Models
|
| 9 |
+
export HF_REPO_TYPE=model
|
| 10 |
+
export HF_PREPROCESSED_REPO=HayrettinIscan/MeshAI-Preprocessed-4K
|
| 11 |
+
export PYTHONUNBUFFERED=1
|
| 12 |
+
|
| 13 |
+
pkill -f 'train_pipeline|stream_train|cloud_orchestrator' 2>/dev/null || true
|
| 14 |
+
sleep 1
|
| 15 |
+
mkdir -p checkpoints data/base_models logs
|
| 16 |
+
|
| 17 |
+
echo "=== Faz2 smoke (16 obje, 1 epoch) ==="
|
| 18 |
+
python3 -u train_pipeline.py --mode faz2 \
|
| 19 |
+
--epochs 1 \
|
| 20 |
+
--limit 16 \
|
| 21 |
+
--lora-rank 8 \
|
| 22 |
+
--validation-every 50 \
|
| 23 |
+
--checkpoint-every 10 \
|
| 24 |
+
--resume_from checkpoints/latest_model.pt
|
| 25 |
+
|
| 26 |
+
echo "=== Smoke OK ise full (limit=0) ==="
|
| 27 |
+
echo "Calistir:"
|
| 28 |
+
echo "python3 -u train_pipeline.py --mode faz2 --epochs 3 --lora-rank 8 --checkpoint-every 50 --resume_from checkpoints/latest_faz2_model.pt"
|