Update Dockerfile

Dockerfile

@@ -1,126 +1,136 @@
 # =============================================================================
-# ADUC-SDR Video Suite — High-Perf Diffusers for 8× L40S (SM 8.9)
+# ADUC-SDR Video Suite — Dockerfile Otimizado
+# Preserva a estrutura de instalação original para alta performance.
 # 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 maintainer="Carlos Rodrigues dos Santos"
+LABEL description="ADUC-SDR: High-performance Diffusers stack for 8x NVIDIA L40S with LTX-Video and SeedVR"
+LABEL version="5.0.0"
 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 ----------------
+# =============================================================================
+# 1. Variáveis de Ambiente e Configuração de Paths
+# =============================================================================
 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
+    PIP_NO_CACHE_DIR=0 PIP_DISABLE_PIP_VERSION_CHECK=1
 
-# GPU/Compute
-ENV NVIDIA_VISIBLE_DEVICES=all
-ENV TORCH_CUDA_ARCH_LIST="8.9"
-ENV CUDA_DEVICE_ORDER=PCI_BUS_ID
-ENV CUDA_DEVICE_MAX_CONNECTIONS=32
+# --- Configurações de GPU e Computação ---
+ENV NVIDIA_VISIBLE_DEVICES=all \
+    TORCH_CUDA_ARCH_LIST="8.9" \
+    CUDA_DEVICE_ORDER=PCI_BUS_ID \
+    CUDA_DEVICE_MAX_CONNECTIONS=32
 
-# Threads
+# --- Configurações de Threads ---
 ENV OMP_NUM_THREADS=8 MKL_NUM_THREADS=8 MAX_JOBS=160
 
-# 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
+# --- Configurações de Alocador de Memória e Caches de GPU ---
+ENV PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512,garbage_collection_threshold:0.8 \
+    CUDA_LAUNCH_BLOCKING=0 CUDA_CACHE_MAXSIZE=2147483648 CUDA_CACHE_DISABLE=0
 
-# App paths
-ENV APP_HOME=/app
-WORKDIR $APP_HOME
+# --- Paths da Aplicação e Dados Persistentes ---
+ENV APP_HOME=/app \
+    HF_HOME=/data/.cache/huggingface \
+    TORCH_HOME=/data/.cache/torch \
+    HF_DATASETS_CACHE=/data/.cache/datasets \
+    TRANSFORMERS_CACHE=/data/.cache/transformers \
+    DIFFUSERS_CACHE=/data/.cache/diffusers \
+    HF_HUB_ENABLE_HF_TRANSFER=1 \
+    TOKENIZERS_PARALLELISM=false
 
-# Persistent data and caches in /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
-ENV HF_HUB_ENABLE_HF_TRANSFER=1
-ENV TOKENIZERS_PARALLELISM=false
+WORKDIR $APP_HOME
 
-# Create non-root user and data dirs early, fix ownership
+# =============================================================================
+# 2. Setup de Usuário e Sistema
+# =============================================================================
+# Cria usuário não-root e diretórios de dados/app.
+# As permissões finais serão aplicadas no final.
 RUN useradd -m -u 1000 -s /bin/bash appuser && \
-    mkdir -p /data /data/models \
-             /data/.cache/huggingface /data/.cache/torch \
-             /data/.cache/datasets /data/.cache/transformers /data/.cache/diffusers && \
-    chown -R appuser:appuser /data
-
-# Models live in /data/models and are visible at /app/models
-ENV MODELS_DIR=/data/models
-RUN ln -sf /data/models /app/models
+    mkdir -p /data $APP_HOME /app/output
 
-# ---------------- System & Python ----------------
+# --- Instalação de Pacotes de Sistema e Python ---
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential gosu tree cmake git git-lfs curl wget ffmpeg ninja-build \
     python3.10 python3.10-dev python3.10-distutils python3-pip \
     ca-certificates libglib2.0-0 libgl1 \
  && 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 && \
+RUN ln -sf /usr/bin/python3.10 /usr/bin/python && \
     python3 -m pip install --upgrade pip
 
-# ---------------- PyTorch cu128 (pinned) ----------------
+# =============================================================================
+# 3. Instalação da Toolchain de Machine Learning (Mantida 100% Original)
+# =============================================================================
+
+# --- PyTorch para CUDA 12.8 ---
 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 (no bnb build) ----------------
+# --- Ferramentas de Compilação, Triton e FlashAttention ---
 RUN pip install packaging ninja cmake pybind11 scikit-build cython hf_transfer "numpy>=1.24.4"
 
-# Triton 3.x (no triton.ops)
+# --- Triton 3.x ---
 RUN pip uninstall -y triton || true && \
     pip install -v --no-build-isolation triton==3.4.0
 
-
-# FlashAttention 2.8.x
+# --- 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
 
-# ---------------- App dependencies ----------------
-COPY requirements.txt ./requirements.txt
+# =============================================================================
+# 4. Instalação das Dependências da Aplicação
+# =============================================================================
+# Copia e instala requirements.txt primeiro para otimizar o cache de camadas do Docker.
+COPY --chown=appuser:appuser requirements.txt ./requirements.txt
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Pin bnb to avoid surprise CUDA/PTX mismatches (adjust as needed)
+# --- Instalação de bitsandbytes e Wheels Customizados (Mantido 100% Original) ---
 RUN pip install --upgrade bitsandbytes
 
-# Custom .whl (Apex + dropout_layer_norm)
+# Instala wheels customizados (Apex, etc.)
 RUN echo "Installing custom wheels..." && \
     pip install --no-cache-dir \
       "https://huggingface.co/euIaxs22/Aduc-sdr/resolve/main/apex-0.1-cp310-cp310-linux_x86_64.whl" \
       "https://huggingface.co/euIaxs22/Aduc-sdr/resolve/main/dropout_layer_norm-0.1-cp310-cp310-linux_x86_64.whl"
 
-# ====================================================================
-# Optional: q8_kernels + LTX-Video (enable if needed; ensure wheel ABI)
+# Instala q8_kernels
 RUN pip install --no-cache-dir \
    "https://huggingface.co/euIaxs22/Aduc-sdr/resolve/main/q8_kernels-0.0.5-cp310-cp310-linux_x86_64.whl"
-# RUN git clone https://github.com/Lightricks/LTX-Video.git /data/LTX-Video && \
-#     cd /data/LTX-Video && python -m pip install -e .[inference]
-# ====================================================================
 
-# Scripts and app
-COPY info.sh ./app/info.sh
-COPY builder.sh ./app/builder.sh
-COPY start.sh ./app/start.sh
-COPY entrypoint.sh ./app/entrypoint.sh
+# NOTA: A clonagem do LTX-Video foi removida daqui.
+# Esta tarefa agora é gerenciada pelo entrypoint para garantir a persistência dos dados.
+# # RUN git clone https://github.com/Lightricks/LTX-Video.git /data/LTX-Video && \
+# #     cd /data/LTX-Video && python -m pip install -e .[inference]
 
-# Copy the rest of the source last for better caching
-COPY . .
+# =============================================================================
+# 5. Cópia do Código-Fonte e Configuração Final
+# =============================================================================
+# Copia o restante do código-fonte da aplicação por último.
+COPY --chown=appuser:appuser . .
 
-# Permissions on app tree
-RUN chown -R appuser:appuser /app /data && \
-    chmod 0755 /app/entrypoint.sh /app/start.sh /app/info.sh /app/builder.sh
+# Garante que todos os scripts de inicialização sejam executáveis
+# e que o usuário 'appuser' seja o dono de todos os arquivos.
+RUN chown -R appuser:appuser $APP_HOME /data && \
+    chmod +x /app/entrypoint.sh /app/start.sh /app/info.sh /app/builder.sh
 
+# =============================================================================
+# 6. Ponto de Entrada
+# =============================================================================
+# Expõe o diretório /data para ser montado como um volume persistente.
 VOLUME /data
 
-ENTRYPOINT ["/app/entrypoint.sh"]
+# Define o usuário padrão para a execução do contêiner.
 USER appuser
 
-# ---------------- Entry ----------------
-CMD ["/app/start.sh"]
+# Define o script que será executado na inicialização do contêiner.
+ENTRYPOINT ["/app/entrypoint.sh"]
+
+# Define o comando padrão a ser executado pelo entrypoint.
+CMD ["/app/start.sh"]