File size: 2,477 Bytes
b703376 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 | # Woosh-DFlow text-to-audio BACKEND Space -- CPU build (free HF Space).
#
# For a dedicated-GPU Space, use Dockerfile.gpu instead (rename it to
# `Dockerfile`); the only difference is the torch wheel index (CUDA vs CPU).
#
# The install sequence mirrors Woosh's pyproject.toml, with two upstream quirks:
# * hear21passt is installed with --no-deps (Woosh's uv config overrides its
# requires-dist to empty; its runtime imports are covered by the deps below).
# * The woosh package itself is installed with --no-deps (its deps are pinned
# explicitly here) so pip never re-resolves the torch stack.
FROM python:3.12-slim-bookworm
ENV DEBIAN_FRONTEND=noninteractive \
PYTHONUNBUFFERED=1 \
PIP_NO_CACHE_DIR=1 \
HF_HOME=/tmp/hf \
PORT=7860
RUN apt-get update && apt-get install -y --no-install-recommends \
build-essential git curl unzip ffmpeg libsndfile1 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /app
# 1) Torch stack -- CPU wheels (this is the CPU image).
RUN pip install --upgrade pip \
&& pip install --index-url https://download.pytorch.org/whl/cpu \
torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0
# 2) Woosh's declared runtime deps (from pyproject.toml), then Woosh itself.
RUN pip install \
"einops>=0.8.1" \
"hydra-core>=1.3.2" \
"lightning>=2.5.6" \
"timm>=1.0.22" \
"torchdiffeq>=0.2.5" \
"transformers>=4.57.2" \
"pydantic>=2.12.4" \
"omegaconf>=2.3.0" \
"av>=16.1.0" \
"requests" \
"soundfile>=0.13.1" \
"gradio>=6.9.0" \
&& pip install --no-deps "hear21passt==0.0.26" \
&& pip install --no-deps "git+https://github.com/SonyResearch/Woosh.git@main"
# 3) Model weights (CC-BY-NC, official v1.0.0 release), baked in for fast
# restarts. Woosh-DFlow needs the DFlow LDM + the audio autoencoder (AE) +
# the audio-CLAP text conditioner (TextConditionerA). Each zip extracts to
# the correct checkpoints/<NAME> folder under /app.
RUN set -eux; cd /app; \
for A in Woosh-DFlow Woosh-AE TextConditionerA; do \
echo "Downloading $A.zip ..."; \
curl -fL --retry 3 -o "$A.zip" \
"https://github.com/SonyResearch/Woosh/releases/download/v1.0.0/$A.zip"; \
unzip -q "$A.zip" -d /app; \
rm -f "$A.zip"; \
done; \
ls -R /app/checkpoints | head -n 60
COPY app.py /app/app.py
EXPOSE 7860
CMD ["python", "app.py"]
|