Upload autosetup.sh with huggingface_hub

autosetup.sh

@@ -0,0 +1,355 @@
+#!/bin/bash
+###############################################################################
+# AUTO SETUP: 4K Image Model Training Pipeline
+# Chạy trên VPS mới (Azure, 2x/8x H100) để restore và resume toàn bộ pipeline.
+#
+# Usage:
+#   bash autosetup.sh <hf_token>
+#
+# Ví dụ:
+#   bash autosetup.sh hf_xxxxxxxxxxxxxxxxxxxx
+###############################################################################
+
+set -e
+
+HF_TOKEN=${1:-""}
+PROJECT_DIR="/home/adminuser/chungcat"
+DATA0="/data0"
+DATA1="/data1"
+
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+NC='\033[0m'
+
+log() { echo -e "${GREEN}[$(date '+%H:%M:%S')]${NC} $1"; }
+warn() { echo -e "${YELLOW}[$(date '+%H:%M:%S')] WARNING:${NC} $1"; }
+err() { echo -e "${RED}[$(date '+%H:%M:%S')] ERROR:${NC} $1"; exit 1; }
+
+###############################################################################
+# 1. CHECK ARGS
+###############################################################################
+if [ -z "$HF_TOKEN" ]; then
+    err "Usage: bash autosetup.sh <hf_token>"
+fi
+
+log "=== 4K Image Model - Full Auto Setup ==="
+log "Project dir: $PROJECT_DIR"
+
+###############################################################################
+# 2. FORMAT & MOUNT NVMe DISKS
+###############################################################################
+log "[1/9] Setting up NVMe storage..."
+
+setup_nvme() {
+    local dev=$1
+    local mount=$2
+    local label=$3
+
+    if mountpoint -q "$mount" 2>/dev/null; then
+        log "  $mount already mounted"
+        return
+    fi
+
+    if [ -b "$dev" ]; then
+        # Check if already has filesystem
+        if ! blkid "$dev" | grep -q ext4; then
+            log "  Formatting $dev..."
+            sudo mkfs.ext4 -L "$label" "$dev"
+        fi
+        sudo mkdir -p "$mount"
+        sudo mount "$dev" "$mount"
+        sudo chown adminuser:adminuser "$mount"
+        log "  Mounted $dev → $mount"
+    else
+        warn "  $dev not found, skipping"
+    fi
+}
+
+setup_nvme /dev/nvme0n1 "$DATA0" "data0"
+setup_nvme /dev/nvme1n1 "$DATA1" "data1"
+
+# Also check /dev/sdb
+if [ -b /dev/sdb1 ] && ! mountpoint -q /mnt 2>/dev/null; then
+    sudo mount /dev/sdb1 /mnt 2>/dev/null || true
+fi
+
+df -h "$DATA0" "$DATA1" 2>/dev/null || true
+
+###############################################################################
+# 3. CREATE PROJECT STRUCTURE
+###############################################################################
+log "[2/9] Creating project structure..."
+
+mkdir -p "$PROJECT_DIR"/{scripts/{data_collection,training,serving},configs,logs}
+mkdir -p "$DATA0"/{datasets/{raw,processed},checkpoints,models}
+mkdir -p "$DATA1"/{outputs,logs}
+
+# Symlinks
+ln -sf "$DATA0/datasets/processed" "$PROJECT_DIR/data/processed" 2>/dev/null || true
+ln -sf "$DATA0/datasets/raw" "$PROJECT_DIR/data/raw" 2>/dev/null || true
+ln -sf "$DATA0/checkpoints" "$PROJECT_DIR/checkpoints" 2>/dev/null || true
+ln -sf "$DATA1/outputs" "$PROJECT_DIR/outputs" 2>/dev/null || true
+
+mkdir -p "$PROJECT_DIR/data" 2>/dev/null || true
+ln -sf "$DATA0/datasets/processed" "$PROJECT_DIR/data/processed"
+ln -sf "$DATA0/datasets/raw" "$PROJECT_DIR/data/raw"
+ln -sf "$DATA0/checkpoints" "$PROJECT_DIR/checkpoints"
+ln -sf "$DATA1/outputs" "$PROJECT_DIR/outputs"
+
+###############################################################################
+# 4. INSTALL DEPENDENCIES
+###############################################################################
+log "[3/9] Installing Python dependencies..."
+
+pip3 install --upgrade pip -q 2>/dev/null
+
+pip3 install -q \
+    torch torchvision torchaudio \
+    diffusers transformers accelerate deepspeed \
+    bitsandbytes peft datasets webdataset img2dataset \
+    wandb safetensors xformers \
+    opencv-python-headless tqdm anthropic \
+    huggingface_hub fastapi uvicorn \
+    pandas pyarrow 2>&1 | tail -3
+
+log "  Dependencies installed"
+
+###############################################################################
+# 5. VERIFY GPU
+###############################################################################
+log "[4/9] Verifying GPU..."
+
+python3 -c "
+import torch
+gpus = torch.cuda.device_count()
+print(f'  PyTorch {torch.__version__} | CUDA {torch.version.cuda} | {gpus} GPUs')
+for i in range(gpus):
+    name = torch.cuda.get_device_name(i)
+    mem = torch.cuda.get_device_properties(i).total_memory / 1024**3
+    print(f'    GPU {i}: {name} ({mem:.0f}GB)')
+if gpus == 0:
+    print('  WARNING: No GPU detected!')
+"
+
+###############################################################################
+# 6. LOGIN HUGGINGFACE
+###############################################################################
+log "[5/9] Logging into HuggingFace..."
+
+hf auth login --token "$HF_TOKEN" 2>/dev/null || \
+    python3 -c "from huggingface_hub import login; login(token='$HF_TOKEN')"
+
+HF_USER=$(python3 -c "from huggingface_hub import HfApi; print(HfApi().whoami()['name'])")
+log "  Logged in as: $HF_USER"
+
+###############################################################################
+# 7. RESTORE FROM HUGGINGFACE
+###############################################################################
+log "[6/9] Restoring from HuggingFace backup..."
+
+python3 -c "
+from huggingface_hub import snapshot_download, HfApi
+import os
+
+user = '$HF_USER'
+project = '$PROJECT_DIR'
+
+repos = {
+    'scripts': f'{user}/4k-image-model-scripts',
+    'data': f'{user}/4k-image-model-data',
+    'checkpoints': f'{user}/4k-image-model-checkpoints',
+}
+
+api = HfApi()
+
+# Restore scripts
+try:
+    print('  Restoring scripts...')
+    snapshot_download(repo_id=repos['scripts'], repo_type='dataset', local_dir=project)
+    print('    OK')
+except Exception as e:
+    print(f'    Skip: {e}')
+
+# Restore data metadata
+try:
+    print('  Restoring data metadata...')
+    snapshot_download(repo_id=repos['data'], repo_type='dataset', local_dir=project)
+    print('    OK')
+except Exception as e:
+    print(f'    Skip: {e}')
+
+# Restore checkpoints
+try:
+    print('  Restoring checkpoints...')
+    snapshot_download(repo_id=repos['checkpoints'], repo_type='model', local_dir='$DATA0/checkpoints')
+    print('    OK')
+except Exception as e:
+    print(f'    Skip: {e}')
+
+print('  Restore complete!')
+"
+
+###############################################################################
+# 8. DOWNLOAD DATA (if not already present)
+###############################################################################
+log "[7/9] Checking/downloading training data..."
+
+# Download COYO metadata if not present
+if [ ! -d "$DATA0/datasets/raw/coyo/data" ]; then
+    log "  Downloading COYO metadata..."
+    python3 -c "
+from huggingface_hub import snapshot_download
+snapshot_download(
+    repo_id='kakaobrain/coyo-700m',
+    repo_type='dataset',
+    local_dir='$DATA0/datasets/raw/coyo',
+    allow_patterns=['data/part-00000-*.parquet', 'data/part-00001-*.parquet', 'data/part-00002-*.parquet'],
+    max_workers=8,
+)
+print('    COYO metadata downloaded')
+" 2>&1 | tail -3
+else
+    log "  COYO metadata already present"
+fi
+
+# Filter COYO if not done
+if [ ! -f "$DATA0/datasets/raw/coyo_filtered/coyo_aesthetic.parquet" ]; then
+    log "  Filtering COYO dataset..."
+    mkdir -p "$DATA0/datasets/raw/coyo_filtered"
+    # Copy filter script output location
+    python3 "$PROJECT_DIR/scripts/data_collection/filter_coyo.py" \
+        --input-dir "$DATA0/datasets/raw/coyo/data" \
+        --output-dir "$DATA0/datasets/raw/coyo_filtered" \
+        --min-aesthetic 5.0 --min-clip 0.2 --max-watermark 0.8 --max-records 1000000 2>&1 | tail -3
+else
+    log "  Filtered data already present"
+fi
+
+# Download images if shards don't exist
+SHARD_DIR="$DATA0/datasets/processed/flux_train/shards"
+if [ ! -d "$SHARD_DIR" ] || [ $(ls "$SHARD_DIR"/*.tar 2>/dev/null | wc -l) -lt 10 ]; then
+    log "  Downloading training images (background)..."
+    mkdir -p "$SHARD_DIR"
+    python3 -c "
+import pandas as pd
+df = pd.read_parquet('$DATA0/datasets/raw/coyo_filtered/coyo_aesthetic.parquet')
+df_out = df[['url', 'text']].rename(columns={'url': 'URL', 'text': 'TEXT'})
+df_out.to_parquet('$SHARD_DIR/_urls.parquet', index=False)
+print(f'    Prepared {len(df_out)} URLs')
+"
+    nohup img2dataset \
+        --url_list "$SHARD_DIR/_urls.parquet" \
+        --input_format parquet \
+        --url_col URL \
+        --caption_col TEXT \
+        --output_format webdataset \
+        --output_folder "$SHARD_DIR" \
+        --processes_count 1 \
+        --thread_count 128 \
+        --image_size 1024 \
+        --resize_mode center_crop \
+        --resize_only_if_bigger True \
+        --number_sample_per_shard 1000 \
+        --retries 2 \
+        > "$DATA1/logs/download_aesthetic.log" 2>&1 &
+    log "  Image download started (PID: $!)"
+else
+    log "  Training images already present ($(ls $SHARD_DIR/*.tar 2>/dev/null | wc -l) shards)"
+fi
+
+# Download 4K images
+SR_DIR="$DATA0/datasets/processed/sr_4k/shards"
+if [ ! -d "$SR_DIR" ] || [ $(ls "$SR_DIR"/*.tar 2>/dev/null | wc -l) -lt 5 ]; then
+    log "  Downloading 4K images (background)..."
+    mkdir -p "$SR_DIR"
+    if [ -f "$DATA0/datasets/raw/coyo_filtered/coyo_4k.parquet" ]; then
+        python3 -c "
+import pandas as pd
+df = pd.read_parquet('$DATA0/datasets/raw/coyo_filtered/coyo_4k.parquet')
+df_out = df[['url', 'text']].rename(columns={'url': 'URL', 'text': 'TEXT'})
+df_out.to_parquet('$SR_DIR/_urls.parquet', index=False)
+print(f'    Prepared {len(df_out)} 4K URLs')
+"
+        nohup img2dataset \
+            --url_list "$SR_DIR/_urls.parquet" \
+            --input_format parquet \
+            --url_col URL \
+            --caption_col TEXT \
+            --output_format webdataset \
+            --output_folder "$SR_DIR" \
+            --processes_count 1 \
+            --thread_count 128 \
+            --image_size 4096 \
+            --resize_mode keep_ratio \
+            --resize_only_if_bigger True \
+            --number_sample_per_shard 100 \
+            --retries 2 \
+            > "$DATA1/logs/download_4k.log" 2>&1 &
+        log "  4K download started (PID: $!)"
+    fi
+else
+    log "  4K images already present ($(ls $SR_DIR/*.tar 2>/dev/null | wc -l) shards)"
+fi
+
+###############################################################################
+# 9. DOWNLOAD FLUX MODEL (if not cached)
+###############################################################################
+log "[8/9] Checking Flux model..."
+
+python3 -c "
+from huggingface_hub import snapshot_download
+import os
+cache_dir = '$DATA0/models'
+model_dir = os.path.join(cache_dir, 'models--black-forest-labs--FLUX.1-schnell')
+if os.path.exists(model_dir):
+    print('  Flux Schnell already cached')
+else:
+    print('  Downloading Flux Schnell...')
+    snapshot_download(
+        repo_id='black-forest-labs/FLUX.1-schnell',
+        cache_dir=cache_dir,
+    )
+    print('  Flux Schnell downloaded')
+" 2>&1 | tail -3
+
+###############################################################################
+# 10. START AUTO-BACKUP
+###############################################################################
+log "[9/9] Starting auto-backup..."
+
+# Kill existing backup process if any
+pkill -f "backup.py --auto" 2>/dev/null || true
+
+nohup python3 "$PROJECT_DIR/scripts/backup.py" --auto --interval 30 --user "$HF_USER" \
+    > "$PROJECT_DIR/logs/backup.log" 2>&1 &
+log "  Auto-backup started (PID: $!, every 30 min)"
+
+###############################################################################
+# DONE
+###############################################################################
+echo ""
+log "=========================================="
+log "  SETUP COMPLETE!"
+log "=========================================="
+echo ""
+echo "  Storage:"
+df -h "$DATA0" "$DATA1" 2>/dev/null | grep -v Filesystem
+echo ""
+echo "  Background jobs:"
+echo "    - Image download: check $DATA1/logs/download_aesthetic.log"
+echo "    - 4K download: check $DATA1/logs/download_4k.log"
+echo "    - Auto-backup: check $PROJECT_DIR/logs/backup.log"
+echo ""
+echo "  Next steps:"
+echo "    1. Wait for data download to finish"
+echo "    2. Start training:"
+echo "       bash $PROJECT_DIR/scripts/training/run_train_flux.sh"
+echo "    3. Manual backup anytime:"
+echo "       python3 $PROJECT_DIR/scripts/backup.py --backup"
+echo ""
+echo "  Monitor:"
+echo "    - Download progress: du -sh $DATA0/datasets/processed/flux_train/shards/"
+echo "    - GPU usage: nvidia-smi"
+echo "    - Training logs: tail -f $PROJECT_DIR/logs/"
+echo ""