Spaces:
Paused
Paused
Commit History
-
Prepare job build context
Georg58e94fa -
Prepare job build context
Georg09ba3b6 -
Update base image build and deps
Georgc58f0bb -
Optimized Docker build to fix OOM errors
Georg14c6649 -
Optimized Docker build to fix OOM errors
Georg8257936
-
Optimized Docker build to fix OOM errors
Georg39ce43c -
Optimized Docker build to fix OOM errors
Georg23fa3da -
Optimized Docker build to fix OOM errors
Georg4d82988 -
Optimized Docker build to fix OOM errors
Georgb45eb9c -
Optimized Docker build to fix OOM errors
Georg0ad0903 -
Optimized Docker build to fix OOM errors
George8f69fc -
Optimized Docker build to fix OOM errors
Georg711f335 -
Optimized Docker build to fix OOM errors
Georg971c84e -
Optimized Docker build to fix OOM errors
Georga586677 -
Optimized Docker build to fix OOM errors
Georg3d1bba0 -
Optimized Docker build to fix OOM errors
Georgfd7af7f -
Optimized Docker build to fix OOM errors
Georg7d682f3 -
Optimized Docker build to fix OOM errors
Georgb89238c -
Optimized Docker build to fix OOM errors
Georgad260ee -
Optimized Docker build to fix OOM errors
Georg3f8ed1c -
Optimized Docker build to fix OOM errors
Georg96b4daf -
Optimized Docker build to fix OOM errors
Georgc23b196 -
Fix Dockerfile structure for faster rebuilds
Georg +1: Claude Sonnet 4.5Move app file copying from Dockerfile.base to main Dockerfile. This ensures changes to app.py, client.py, or estimator.py trigger rebuilds without requiring base image rebuild. Changes: - Dockerfile: Add COPY for application files before CMD - Dockerfile.base: Remove COPY (base image is cached) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>42ce71e -
Re-add --no-build-isolation flag required by nvdiffrast
Georg1fd398f -
Fix nvdiffrast OOM: set MAX_JOBS=1 before compilation
Georgf895f87 -
Add FoundationPose dependencies, build nvdiffrast on HuggingFace with GPU
Georgc05ac4f -
Add pybind11-dev for mycpp build
Georgd0b92e0 -
Add Boost libraries for mycpp CMake build
Georg65d25c9 -
Fix mycpp build: use CMake instead of setup.py
Georg81abf94 -
Reduce to single CUDA arch (7.5) and limit parallel jobs to reduce memory usage
Georgf054fb1 -
Add python3.10-dev for C++ extension compilation
Georgb2624fb -
Optimized Docker build to fix OOM errors
Georg9f36f2f -
Fix NumPy version and add CUDA compiler for C++ extensions
George317b3c -
Optimized Docker build to fix OOM errors
Georg3968781 -
Two-stage Docker build: base image + GPU compilation
Georgbbc3fdc -
Fix C++ standard and Eigen dependency for CUDA build
Georg +1: Claude Sonnet 4.5Build Errors Fixed: - Add libeigen3-dev for Eigen/Dense header - Add ninja-build for faster compilation - Patch mycuda setup.py: change -std=c++14 to -std=c++17 - PyTorch requires C++17, mycuda was hardcoded to C++14 Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>0d59dc5 -
Fix CUDA architecture detection during build
Georg +1: Claude Sonnet 4.5Add TORCH_CUDA_ARCH_LIST environment variable to specify CUDA architectures for compilation. PyTorch can't auto-detect during Docker build without GPU, causing IndexError in arch_list. Covers GPUs: Turing (7.5), Ampere (8.0, 8.6), Ada (8.9), Hopper (9.0) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>610de19 -
Use regular install instead of editable for mycuda
Georg +1: Claude Sonnet 4.5- Change 'pip install -e .' to 'pip install .' for mycuda - Upgrade setuptools and wheel before building - Editable mode not needed for Docker Space deployment Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>4e59dce
-
Fix torch visibility during C++ extension build
Georg +1: Claude Sonnet 4.5Add --no-build-isolation flag to pip install for mycuda extension. This prevents pip from using an isolated environment where torch is not visible during the build process. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>dbd19a9 -
Fix Dockerfile build errors and always use real model
Georg +1: Claude Sonnet 4.5Build Fixes: - Add cmake and build-essential to system dependencies - Install PyTorch before building C++ extensions (required) - Build only mycuda and mycpp (skip optional kaolin) - Use proper pip install commands for C++ extensions Always Real Mode: - Remove USE_HF_WEIGHTS and USE_REAL_MODEL env var checks - Always download weights from gpue/foundationpose-weights - Set USE_REAL_MODEL=true permanently in app.py - Update UI to show 'Real FoundationPose with GPU' - Update README to mention Docker GPU (not ZeroGPU) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>dd44013 -
Fix Dockerfile Python one-liner syntax error
Georg +1: Claude Sonnet 4.5- Create download_weights.py script instead of inline Python - Update Dockerfile to use script (cleaner and avoids syntax errors) - Copy estimator.py in Dockerfile Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>d5c5d99 -
Convert to Docker Space for FoundationPose C++ support
Georg +1: Claude Sonnet 4.5- Change SDK from gradio to docker in README.md - Add Dockerfile with CUDA 11.8 and FoundationPose build - Create estimator.py wrapper for FoundationPose - Remove ZeroGPU (spaces) dependency - Docker uses persistent GPU - Remove @spaces.GPU decorators from app.py - Clean up unused files (app_simple.py, deploy.sh, test scripts) - Keep Gradio UI for web interface Docker Spaces provide persistent GPU and build environment needed for FoundationPose's C++ extensions. Gradio still works for UI. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>2df2c23 -
Remove Dockerfile - ZeroGPU Spaces use Python directly
Georg +1: Claude Sonnet 4.5ZeroGPU Spaces run with Python environment and requirements.txt, not Docker. Removing Dockerfile to fix deployment. Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>d5c35b5 -
now with real weights
George15abf5 -
initial commit
Georg24857f8