Dockerfile

# Unsloth Training Hub Dockerfile
# CRITICAL: Installation order matters for Unsloth!

FROM nvidia/cuda:12.1.0-devel-ubuntu22.04

# Prevent interactive prompts
ENV DEBIAN_FRONTEND=noninteractive

# Install system dependencies
RUN apt-get update && apt-get install -y     python3.11     python3.11-venv     python3.11-dev     python3-pip     git     wget     curl     ninja-build     && rm -rf /var/lib/apt/lists/*

# Set Python 3.11 as default
RUN update-alternatives --install /usr/bin/python python /usr/bin/python3.11 1
RUN update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.11 1

# Create virtual environment
RUN python3 -m venv /opt/venv
ENV PATH="/opt/venv/bin:$PATH"

# Upgrade pip and install build dependencies
RUN pip install --upgrade pip wheel setuptools psutil packaging

# ============================================================================
# CRITICAL: UNSLOTH INSTALLATION ORDER
# 1. Install PyTorch first
# 2. Install flash-attn separately with no-build-isolation
# 3. Install unsloth
# 4. Install vllm
# 5. Install diffusers (for GRPO)
# 6. Install TRL from git (latest)
# ============================================================================

# Step 1: PyTorch with CUDA 12.1
RUN pip install torch==2.4.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

# Step 2: Install flash-attn separately (needs torch already installed)
# Using --no-build-isolation so it can see the installed torch
RUN pip install flash-attn --no-build-isolation

# Step 3: Unsloth (main package) - now flash-attn is already installed
RUN pip install "unsloth @ git+https://github.com/unslothai/unsloth.git"

# Step 4: vLLM for fast inference during RL
RUN pip install vllm

# Step 5: diffusers (required for GRPO)
RUN pip install diffusers

# Step 6: TRL from git (latest for GRPO/GSPO support)
RUN pip install git+https://github.com/huggingface/trl.git

# Step 7: Additional dependencies
RUN pip install     gradio>=4.0.0     datasets>=2.18.0     anthropic>=0.39.0     huggingface_hub>=0.24.0     pyyaml     accelerate>=0.30.0     sentencepiece     protobuf

# Set working directory
WORKDIR /app

# Copy application code
COPY . /app/

# Create directories for runs
RUN mkdir -p /app/runs /app/data /app/outputs

# Environment variables
ENV PYTHONUNBUFFERED=1
ENV HF_HUB_ENABLE_HF_TRANSFER=1
ENV UNSLOTH_VLLM_STANDBY=1

# Expose Gradio port
EXPOSE 7860

# Run app
CMD ["python3", "app.py"]