Upload 2 files

Dockerfile

@@ -0,0 +1,150 @@
+# =============================================================================
+# ADUC-SDR Video Suite — High-Perf Diffusers for 8× L40S (SM 8.9)
+# CUDA 12.8 | PyTorch 2.8.0+cu128 | Ubuntu 22.04
+# =============================================================================
+FROM nvidia/cuda:12.8.0-devel-ubuntu22.04
+
+LABEL maintainer="Carlos Rodrigues dos Santos & Development Partner"
+LABEL description="High-performance Diffusers stack with FA2/SDPA, 8×L40S"
+LABEL version="4.4.0"
+LABEL cuda_version="12.8.0"
+LABEL python_version="3.10"
+LABEL pytorch_version="2.8.0+cu128"
+LABEL gpu_optimized_for="8x_NVIDIA_L40S"
+
+# ---------------- Core env & caches ----------------
+ENV DEBIAN_FRONTEND=noninteractive TZ=UTC LANG=C.UTF-8 LC_ALL=C.UTF-8 \
+    PYTHONUNBUFFERED=1 PYTHONDONTWRITEBYTECODE=1 \
+    PIP_NO_CACHE_DIR=1 PIP_DISABLE_PIP_VERSION_CHECK=1
+
+# GPU/Compute
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+ENV TORCH_CUDA_ARCH_LIST="8.9"
+ENV CUDA_DEVICE_ORDER=PCI_BUS_ID
+ENV CUDA_DEVICE_MAX_CONNECTIONS=32
+ENV CUDA_MODULE_LOADING=LAZY
+
+# Threads
+ENV OMP_NUM_THREADS=8 MKL_NUM_THREADS=8 MAX_JOBS=48
+
+# Alloc/caches
+ENV PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512,garbage_collection_threshold:0.8
+ENV CUDA_LAUNCH_BLOCKING=0 CUDA_CACHE_MAXSIZE=2147483648 CUDA_CACHE_DISABLE=0
+
+# NCCL single-node sane defaults (use IB=1/IFNAME em clusters com IB)
+ENV NCCL_DEBUG=INFO NCCL_ASYNC_ERROR_HANDLING=1 NCCL_P2P_DISABLE=0 NCCL_IB_DISABLE=1 \
+    NCCL_MIN_NCHANNELS=8 NCCL_NTHREADS=256 NCCL_SOCKET_IFNAME=lo
+
+# Hugging Face caches
+ENV APP_HOME=/app
+WORKDIR $APP_HOME
+ENV TORCH_HOME=/app/.cache/torch
+ENV HF_HOME=/app/.cache/huggingface
+ENV HF_DATASETS_CACHE=/app/.cache/datasets
+ENV TRANSFORMERS_CACHE=/app/.cache/transformers
+ENV DIFFUSERS_CACHE=/app/.cache/diffusers
+ENV HF_HUB_ENABLE_HF_TRANSFER=1
+ENV TOKENIZERS_PARALLELISM=false
+
+# FlashAttention / SDPA defaults
+ENV FLASH_ATTENTION_DISABLE=0 \
+    FLASH_ATTENTION_FORCE_FP16=1 \
+    ATTN_FORCE_F16=1 \
+    ENABLE_FLASH_SDP=1 \
+    ENABLE_MEMORY_EFFICIENT_SDP=1 \
+    ENABLE_MATH_SDP=0 \
+    XFORMERS_FORCE_DISABLE=1 \
+    TORCH_DTYPE=bfloat16 \
+    NVIDIA_TF32_OVERRIDE=0 \
+    FA_LOG_LEVEL=WARNING
+
+# Link de modelos
+
+
+ENV MODELS_DIR=/app/models
+RUN mkdir -p /home/user/.cache/models && ln -sf /home/user/.cache/models /app/models
+
+# ---------------- Sistema & Python ----------------
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential cmake git git-lfs curl wget ffmpeg ninja-build \
+    python3.10 python3.10-dev python3.10-distutils python3-pip \
+    && apt-get clean && rm -rf /var/lib/apt/lists/*
+
+RUN ln -sf /usr/bin/python3.10 /usr/bin/python3 && \
+    ln -sf /usr/bin/python3.10 /usr/bin/python && \
+    python3 -m pip install --upgrade pip
+
+
+# ---------------- PyTorch cu128 (pinado) ----------------
+RUN pip install --index-url https://download.pytorch.org/whl/cu128 \
+    torch==2.8.0+cu128 torchvision==0.23.0+cu128 torchaudio==2.8.0+cu128
+
+# ---------------- Toolchain, Triton, FA2 (sem bnb) ----------------
+RUN pip install packaging ninja cmake pybind11 scikit-build cython hf_transfer numpy==1.24.4
+
+# Triton 3.x (sem triton.ops)
+RUN pip uninstall -y triton || true && \
+    pip install -v --no-build-isolation triton==3.4.0
+
+# FlashAttention 2.8.x
+#RUN pip install flash-attn==2.8.3 --no-build-isolation || \
+#    pip install flash-attn==2.8.2 --no-build-isolation || \
+#    pip install flash-attn==2.8.1 --no-build-isolation || \
+#    pip install flash-attn==2.8.0.post2 --no-build-isolation
+
+# Diffusers/Transformers estáveis (sem dev)
+RUN pip install --no-cache-dir diffusers==0.31.0 transformers==4.44.2 accelerate==0.34.2 omegaconf==2.3.0
+
+# Opcional: seu fork de otimizações
+# RUN pip install -U git+https://github.com/carlex22/diffusers-aduc-sdr
+
+# ---------------- Repositórios auxiliares ----------------
+RUN git clone https://github.com/bytedance-seed/VINCIE.git  && \
+    cp -r VINCIE/configs/. /app/configs/
+
+# Exemplos adicionais (descomente se precisar)
+RUN git clone https://github.com/bytedance-seed/SeedVR.git && \
+    cp -r SeedVR/configs_3b /app/configs_3b
+
+RUN git clone https://github.com/hkchengrex/MMAudio.git && \
+    cp -r MMAudio/mmaudio /app/mmaudio
+
+RUN git clone https://github.com/Lightricks/LTX-Video.git && \
+    cp -r LTX-Video/ltx_video /app/ltx_video
+
+# ---------------- Dependências da aplicação ----------------
+COPY requirements.txt ./requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+
+RUN pip install --upgrade bitsandbytes
+
+# Scripts e configs
+COPY info.sh ./info.sh
+COPY start.sh ./start.sh
+COPY /configs ./configs
+
+# ---------------- Código e permissões ----------------
+COPY . .
+RUN useradd -m -u 1000 -s /bin/bash appuser && \
+    chown -R appuser:appuser /app && \
+    chmod 0755 /app/start.sh /app/info.sh || true
+
+USER appuser
+
+# Declara volume persistente para HF Spaces
+VOLUME /data
+
+# Env vars para caches em /data
+ENV HF_HOME=/data/.cache/huggingface
+ENV TORCH_HOME=/data/.cache/torch
+ENV HF_DATASETS_CACHE=/data/.cache/datasets
+ENV TRANSFORMERS_CACHE=/data/.cache/transformers
+ENV DIFFUSERS_CACHE=/data/.cache/diffusers
+
+
+
+# ---------------- Entry ----------------
+ENTRYPOINT ["./start.sh"]
+CMD ["gradio"]

start.sh

@@ -0,0 +1,155 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+echo "======================================================="
+echo " ADUC-SDR — Start (VINCIE/SeedVR, 8× L40S)"
+echo "======================================================="
+
+# ---------------------- Env base ----------------------
+export CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0,1,2,3,4,5,6,7}"
+export TORCH_DTYPE="${TORCH_DTYPE:-bfloat16}"
+
+# SDPA / FA toggles
+export ENABLE_FLASH_SDP="${ENABLE_FLASH_SDP:-1}"
+export ENABLE_MEMORY_EFFICIENT_SDP="${ENABLE_MEMORY_EFFICIENT_SDP:-1}"
+export ENABLE_MATH_SDP="${ENABLE_MATH_SDP:-0}"
+export FLASH_ATTENTION_DISABLE="${FLASH_ATTENTION_DISABLE:-0}"
+export XFORMERS_FORCE_DISABLE="${XFORMERS_FORCE_DISABLE:-1}"
+
+# CUDA / NCCL baseline
+export CUDA_MODULE_LOADING="LAZY"
+export CUDA_DEVICE_MAX_CONNECTIONS="${CUDA_DEVICE_MAX_CONNECTIONS:-32}"
+export CUDA_DEVICE_ORDER="PCI_BUS_ID"
+export PYTORCH_CUDA_ALLOC_CONF="max_split_size_mb:512,garbage_collection_threshold:0.8"
+export OMP_NUM_THREADS="${OMP_NUM_THREADS:-8}"
+export MKL_NUM_THREADS="${MKL_NUM_THREADS:-8}"
+export NCCL_DEBUG="INFO"
+export NCCL_ASYNC_ERROR_HANDLING=1
+export NCCL_P2P_DISABLE=0
+export NCCL_IB_DISABLE=1
+export NCCL_SOCKET_IFNAME="lo"
+export NCCL_BLOCKING_WAIT=1
+export TORCH_NCCL_BLOCKING_WAIT=1
+export NCCL_TIMEOUT=600
+
+# ---------------------- Persistência HF/torch ----------------------
+if [ -d /data ]; then
+  export HF_HOME="/data/.cache/huggingface"
+  export TORCH_HOME="/data/.cache/torch"
+else
+  export HF_HOME="/app/.cache/huggingface"
+  export TORCH_HOME="/app/.cache/torch"
+fi
+export HF_HUB_CACHE="${HF_HUB_CACHE:-$HF_HOME/hub}"
+mkdir -p "$HF_HUB_CACHE" "$TORCH_HOME"
+
+mkdir -p /app/.cache
+ln -sf "$HF_HOME" /app/.cache/huggingface
+unset TRANSFORMERS_CACHE
+
+export HF_HUB_ENABLE_HF_TRANSFER=1
+export HF_HUB_DOWNLOAD_TIMEOUT=60
+
+MODEL_REPO="ByteDance-Seed/VINCIE-3B"
+CACHE_MODEL_DIR="$HF_HUB_CACHE/models--ByteDance-Seed--VINCIE-3B"
+CKPT_DIR="/app/ckpt/VINCIE-3B"
+mkdir -p "$CKPT_DIR"
+
+# ---------------------- Download: cache estruturado ou direto ----------------------
+if [ "${DIRECT_TO_CKPT:-0}" -eq 1 ]; then
+  echo "[direct] Baixando ${MODEL_REPO} diretamente para $CKPT_DIR"
+  python - <<'PY'
+import os
+from pathlib import Path
+from huggingface_hub import snapshot_download
+repo_id = "ByteDance-Seed/VINCIE-3B"
+ckpt_dir = Path("/app/ckpt/VINCIE-3B")
+ckpt_dir.mkdir(parents=True, exist_ok=True)
+token = os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_TOKEN")
+snapshot_download(
+  repo_id=repo_id,
+  local_dir=str(ckpt_dir),
+  local_dir_use_symlinks=False,
+  resume_download=True,
+  token=token,
+)
+print("[direct] Snapshot materializado em", ckpt_dir)
+PY
+else
+  echo "Verificando snapshot do ${MODEL_REPO} no cache..."
+  python - <<'PY'
+import os
+from pathlib import Path
+from huggingface_hub import snapshot_download
+hf_home = os.environ.get("HF_HOME","/app/.cache/huggingface")
+cache_dir = os.path.join(hf_home, "hub")
+os.makedirs(cache_dir, exist_ok=True)
+repo_id = "ByteDance-Seed/VINCIE-3B"
+token = os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_TOKEN")
+snapshot_download(
+  repo_id=repo_id,
+  cache_dir=cache_dir,
+  resume_download=True,
+  local_dir_use_symlinks=False,
+  token=token,
+)
+mcache = Path(cache_dir) / ("models--" + repo_id.replace("/", "--"))
+#print("[cache] Estrutura em:", mcache)
+#print("[cache] refs:", list((mcache/"refs").glob("*")))
+#print("[cache] snapshots:", [p.name for p in (mcache/"snapshots").glob("*") if p.is_dir()])
+PY
+fi
+
+
+python3 - <<'PY'
+from huggingface_hub import snapshot_download
+import os
+save_dir = '/app/ckpt/VINCIE-3B'
+os.makedirs(save_dir, exist_ok=True)
+try:
+    print('📥 Baixando VINCIE-3B...')
+    snapshot_download(
+        repo_id='ByteDance-Seed/VINCIE-3B',
+        local_dir=save_dir,
+        cache_dir=cache_dir,
+        #resume_download=True,
+        #local_dir_use_symlinks=False
+    )
+    print('✅ Modelo ok')
+except Exception as e:
+    print(f'⚠️ Download falhou: {e}')
+PY
+
+mkdir -p /app/VINCIE/ckpt
+ln -sfn /app/ckpt/VINCIE-3B /app/VINCIE/ckpt/VINCIE-3B
+
+
+
+#echo "[diag] Cache model dir: $CACHE_MODEL_DIR"
+#ls -la "$CACHE_MODEL_DIR" || true
+#echo "[diag] refs:"; ls -la "$CACHE_MODEL_DIR/refs" || true
+#echo "[diag] snapshots:"; ls -la "$CACHE_MODEL_DIR/snapshots" || true
+#echo "[diag] CKPT_DIR: $CKPT_DIR"; ls -la "$CKPT_DIR" || true
+
+# ---------------------- Builder Apex/Q8 ----------------------
+if nvidia-smi >/dev/null 2>&1; then
+  if [ "${DISABLE_BUILDER:-0}" -eq 0 ]; then
+    echo "Executando builder Apex/Q8..."
+    chmod +x /app/builder.sh || true
+    timeout "${BUILDER_TIMEOUT_SEC:-7200}" bash -lc /app/builder.sh || echo "Builder excedeu tempo/erro, prosseguindo."
+  else
+    echo "Builder desabilitado por DISABLE_BUILDER=1"
+  fi
+else
+  echo "GPU não visível, pulando builder Apex/Q8."
+fi
+
+# ---------------------- Diagnóstico ----------------------
+/app/info.sh || true
+#ls -la /app || true
+#ls -R /app | head -n 2000 || true
+
+# ---------------------- Subindo serviço ----------------------
+echo "🚀 Subindo serviços..."
+# Dica: pode-se exportar VINCIE_DIRECT_TO_CKPT=1 para fallback interno
+python /app/app_vince.py