Ship pre-built CUDA wheels (TE+TE_torch/VIPE/DA3/gsplat) via Git LFS

.gitattributes

@@ -0,0 +1,3 @@
+# Pre-built CUDA-extension wheels (built on Modal, shipped via Git LFS).
+# They're >10 MB each and can't be compiled on HF's memory-constrained builder.
+wheels/*.whl filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -85,34 +85,26 @@ RUN sed -i 's/^tensorstore==0\.1\.45/tensorstore==0.1.50/' Lyra-2/requirements.t
 RUN uv pip install -r Lyra-2/requirements.txt && \
     uv pip install "git+https://github.com/microsoft/MoGe.git"
 
-# Build backends some extensions expect on PYTHONPATH even with --no-build-isolation.
+# Build backend metadata (needed for wheel install of editables and as a safety net).
 RUN uv pip install setuptools wheel ninja
 
-# --- CUDA extensions (all need --no-build-isolation + CPATH) ---------------
-RUN source /home/user/.buildrc && \
-    uv pip install --no-build-isolation "transformer_engine[pytorch]"
+# --- CUDA extensions: install from pre-built wheels via Git LFS ------------
+# HF Space builders can't compile TE/VIPE/DA3/gsplat (OOMs silently after 10+ min).
+# Wheels in ./wheels/ are compiled once on Modal (1.9 TB RAM) and shipped via Git LFS.
+# They're ABI-pinned to Python 3.12 + torch 2.7.1+cu128 which is what this image has.
+COPY --chown=user wheels/ ./wheels/
+RUN uv pip install ./wheels/*.whl
 
-# flash-attn: prebuilt wheel for torch 2.7 / cu12 / py312.
+# flash-attn: upstream prebuilt wheel for torch 2.7 / cu12 / py312.
 RUN uv pip install \
         "https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.7cxx11abiFALSE-cp312-cp312-linux_x86_64.whl"
 
-RUN source /home/user/.buildrc && \
-    uv pip install --no-build-isolation -e Lyra-2/lyra_2/_src/inference/vipe
-
-RUN source /home/user/.buildrc && \
-    uv pip install --no-build-isolation -e "Lyra-2/lyra_2/_src/inference/depth_anything_3[gs]"
-
-# DA3's undeclared transitive runtime deps (addict is missing from its
-# pyproject.toml; --no-build-isolation skips the rest of the resolver).
+# DA3's undeclared transitive runtime deps (addict missing from its pyproject;
+# these aren't resolved when we install from local wheels with pre-compiled .so).
 RUN uv pip install \
         addict evo e3nn plyfile pillow-heif moviepy open3d typer \
         kornia pycolmap trimesh fastapi uvicorn gradio
 
-# gsplat at DA3's pinned commit (different from Lyra-1's gsplat commit).
-RUN source /home/user/.buildrc && \
-    uv pip install --no-build-isolation \
-        "git+https://github.com/nerfstudio-project/gsplat.git@0b4dddf04cb687367602c01196913cde6a743d70"
-
 # Repin huggingface_hub < 1.0 (chain installs can bump past 1.0 and break
 # transformers) and gdown < 6 (6.0 removed the `fuzzy` kwarg VIPE uses).
 RUN uv pip install "huggingface_hub>=0.36.0,<1.0" "gdown<6"

wheels/depth_anything_3-0.0.0-py3-none-any.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebc7673e06ebca579b1c6aff0020efaaa36af4694dc6a98102879a3a1c83798d
+size 275590

wheels/gsplat-1.5.2-cp312-cp312-linux_x86_64.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9757c825b478a4d3e27d233cf1980544ee141e81784abbc856288c2e2d70964f
+size 9526016

wheels/transformer_engine-2.13.0-py3-none-any.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb1ad42df836dbe704f9bd140089d6358d7ebea2eeb36f90d533af4504fe2005
+size 769798

wheels/transformer_engine_torch-2.13.0-cp312-cp312-linux_x86_64.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:314e500655f11ba6bde091229159de372adbf580e8d93b21c57e50d2329f0831
+size 714891

wheels/vipe-0.1.1+pt27cu128-cp312-cp312-linux_x86_64.whl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a28087896bb743aadb95edc4fb6c5e3f64dac817c902b40e2c815eb62d88759a
+size 1855161