Update Dockerfile

Dockerfile

@@ -1,12 +1,55 @@
-# Base image otimizada para NVIDIA L40S com CUDA 12.2 (compatível com HF Spaces e multi-GPU)
-FROM nvidia/cuda:12.2.0-runtime-ubuntu22.04
+# =============================================================================
+# 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
 
-# Labels para Hugging Face Spaces
-LABEL maintainer="Seu Nome <[email protected]>"
-LABEL description="VINCIE Service com persistência e multi-GPU"
+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"
 
-# Argumentos de build (ajuste se necessário)
-ARG DEBIAN_FRONTEND=noninteractive
+# ---------------- 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
+
+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
 
 # Diretórios persistentes (usados em HF Spaces)
 ENV REPOS_DIR=/data/repos
@@ -14,18 +57,42 @@ ENV MODELS_DIR=/data/models
 
 # ---------------- Sistema & Python ----------------
 RUN apt-get update && apt-get install -y --no-install-recommends \
-    python3.10 python3-pip python3-dev git wget curl build-essential libgl1-mesa-glx libglib2.0-0 \
+    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/*
 
-# Atualiza pip e instala wheel para pacotes
-RUN pip3 install --upgrade pip wheel setuptools
+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
 
 # Cria usuário não-root para segurança (HF recomenda)
 RUN useradd -m appuser && mkdir -p /app
 WORKDIR /app
 USER appuser
 
-# ---------------- Dependências da aplicação ----------------
 
 # Copia arquivos necessários
 COPY . .
@@ -33,11 +100,6 @@ COPY . .
 COPY requirements.txt ./requirements.txt
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Instala dependências do requirements.txt
-#RUN pip3 install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu122/ \
-#    && pip3 install --no-cache-dir huggingface-hub transformers gradio  # Adicione se não estiver em requirements.txt
-
-
 # Permissões para scripts
 RUN chmod +x /app/start.sh /app/builder.sh /app/info.sh