update

Dockerfile

@@ -1,13 +1,15 @@
-# Python 3.11 + PyTorch 2.7.0 + CUDA 12.8 + CuDNN 9.5
+# Python 3.11 + PyTorch 2.7.1 + CUDA 12.8 + CuDNN 9.5
 FROM pytorch/pytorch:2.7.1-cuda12.8-cudnn9-devel
 
 # 기본 작업 경로 설정
 WORKDIR /app
 
 # 필수 패키지 설치
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    git wget curl libgl1 libglib2.0-0 libcudnn9-cuda-12 libcudnn9-dev-cuda-12 \
-    && rm -rf /var/lib/apt/lists/*
+RUN sed -i 's|archive.ubuntu.com|mirror.kakao.com|g' /etc/apt/sources.list && \
+    apt-get update && \
+    apt-get install -y apt-utils && \
+    apt-get install -y --no-install-recommends git wget curl && \
+    rm -rf /var/lib/apt/lists/*
 
 # Python 패키지 캐싱 방지
 ENV PIP_NO_CACHE_DIR=1
@@ -20,8 +22,12 @@ COPY . /app/sdxl_train_captioner
 # requirements.txt 설치
 WORKDIR /app/sdxl_train_captioner
 
+# 2. xformers
+RUN pip install xformers==0.0.31
+
 RUN pip install --no-cache-dir -r requirements.txt
-RUN pip install flash-attn --no-build-isolation
+# 문제 발생 시 버전 고정: ==2.7.4.post1
+RUN pip install flash-attn --no-build-isolation 
 
 RUN mkdir -p /app/sdxl_train_captioner/dataset
 RUN mkdir -p /app/sdxl_train_captioner/models

doc/pytorch-cuda-ver-check.txt

@@ -0,0 +1,2 @@
+python -c "import torch; print(torch.__version__)"
+python -c "import transformers; print(transformers.__version__)"

docker-build.cmd

@@ -1,3 +1,3 @@
-docker build --no-cache -t aicompanion/sdxl_train_captioner:0.9.7 .
+docker build --no-cache -t aicompanion/sdxl_train_captioner:0.9.8 .
 
-docker tag aicompanion/sdxl_train_captioner:0.9.7 aicompanion/sdxl_train_captioner:latest
+docker tag aicompanion/sdxl_train_captioner:0.9.8 aicompanion/sdxl_train_captioner:latest

requirements.txt

@@ -1,52 +1,52 @@
 # Core packages
-accelerate==0.33.0
-transformers==4.44.0
-diffusers[torch]==0.25.0
+torch==2.7.1+cu128 --index-url https://download.pytorch.org/whl/cu128
+torchvision==0.22.1+cu128 --index-url https://download.pytorch.org/whl/cu128
+torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128
+accelerate==0.34.2
+transformers==4.45.2  # ✅ 업데이트
+diffusers[torch]==0.34.0  # ✅ 업데이트
 ftfy==6.1.1
 einops==0.7.0
-pytorch-lightning==1.9.0
+# pytorch-lightning==1.9.0  # ❌ 제거! (범인)
+lightning==2.5.5  # ✅ 최신 버전
 lion-pytorch==0.0.6
 schedulefree==1.4
 pytorch-optimizer==3.5.0
 prodigy-plus-schedule-free==1.9.0
 prodigyopt==1.1.2
 tensorboard
-safetensors==0.4.4
+safetensors==0.4.5  # ✅ 업데이트
 altair==4.2.2
 easygui==0.98.3
 toml==0.10.2
 voluptuous==0.13.1
-huggingface-hub==0.24.5
+huggingface-hub>=0.27.0  # ✅ 업데이트
 imagesize==1.4.1
-numpy<=2.0
-requests==2.28.2
-timm==0.4.12
-fairscale==0.4.4
-opencv-python==4.5.5.64
-opencv-python-headless==4.5.5.64
+numpy==1.26.4  # ✅ 구체적 버전
+requests==2.31.0  # ✅ 업데이트
+timm==1.0.21  # ✅ 업데이트 (2번째 범인)
+fairscale==0.4.13  # ✅ 업데이트 (3번째 범인)
+opencv-python==4.7.0.72
+opencv-python-headless==4.7.0.72
+numpy==1.26.4
 
-# WD14 captioning (optional)
-# tensorflow==2.10.1
-onnx==1.15.0
-# onnxruntime-gpu==1.17.1
-# onnxruntime==1.17.1
+# WD14 captioning
+onnx==1.16.0  # ✅ 업데이트
 onnxruntime-gpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/simple/
 
 # BLIP captioning
-blip==0.1.0
-# salesforce-lavis==1.0.2  # 필요 시 주석 해제
+# blip==0.1.0  # ❌ 주석처리 (구버전, 문제 가능)
+Pillow>=10.0.0  # ✅ BLIP 대신 필요하면
 
 # NLP utils
 nltk==3.9.2
 sentencepiece==0.2.0
 
 # OpenCLIP for SDXL
-open-clip-torch==2.20.0
+open-clip-torch==2.26.1  # ✅ 업데이트
 
 # Logging
-rich==13.7.0
+rich==13.9.4  # ✅ 업데이트
 
 # Kohya_ss library
-xformers --index-url https://download.pytorch.org/whl/cu128
-triton==3.0.0
-
+triton==3.3.1