deployment
Guides Zero123Plus demo, serving, and Cog deployment workflows, including Gradio, Streamlit, checkpoints, UI options, and deployment failures.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- deployment
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides Zero123Plus demo, serving, and Cog deployment workflows, including Gradio, Streamlit, checkpoints, UI options, and deployment failures.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/demo-options.md6.8 KB
- 参考文档references/deployment.md8.4 KB
- 参考文档references/troubleshooting.md7.2 KB
- 脚本scripts/cog_predictor.py8.8 KB
- 脚本scripts/launch_gradio_demo.py8.7 KB
03 SKILL.md 正文
Zero123Plus Deployment
Use this sub-skill when a task is about launching, adapting, packaging, or troubleshooting Zero123Plus demos and service surfaces: Streamlit, Gradio, Docker/Gitpod-style environments, Cog/Replicate predictors, checkpoint caches, ports, sharing, queues, and UI options.
Do not use this sub-skill as the main source for image-to-multiview
algorithm details, ControlNets, camera poses, or batch generation code. Route
those questions to the sibling generation sub-skill at
../generation/SKILL.md, then return here for serving
and deployment decisions.
Read first
references/deployment.md: deployment surfaces, dependency/bootstrap patterns, Docker/Gitpod/Cog behavior, and how the surfaces differ.references/demo-options.md: Streamlit and Gradio UI inputs, preprocessing options, examples, and output behavior.references/troubleshooting.md: deployment failure modes for checkpoints, rembg/SAM, ports, queues, Cog weights, pget, CUDA, and source demo side effects.
Bundled deployment templates
scripts/launch_gradio_demo.py: a compact, self-contained Gradio launcher template. It avoids model downloads by default; pass--allow-downloadonly when network/model fetching is approved.scripts/cog_predictor.py: a self-contained Cog predictor template with configurable cache/model paths and no model loading on import.
Routing procedure
- Identify the requested surface: local Streamlit demo, Gradio demo, container image, Gitpod-style workspace, Cog/Replicate predictor, or a custom service.
- Check whether the user has approved network downloads and whether the runtime has a CUDA GPU. Stop or ask before triggering Hugging Face model downloads, SAM checkpoint downloads, rembg/ONNX model downloads, or Cog weight archive fetches.
- For a safer local Gradio launch, prefer the bundled launcher template over importing source UI modules. For Cog, adapt the bundled predictor template and configure its environment variables instead of relying on checkout-relative paths.
- For UI behavior, input preprocessing, background removal, and output layout,
read
references/demo-options.md. For exact pipeline arguments and generated view semantics, route to../generation/SKILL.md. - If a failure involves missing CUDA, unavailable weights, missing SAM/rembg
components, blocked Gradio share tunnels, or Cog cache/download issues, use
references/troubleshooting.mdbefore proposing code changes.
Stop conditions
- No CUDA GPU is available for an actual generation-backed demo or Cog run.
- The required model/checkpoint/cache is absent and downloads are not approved.
- Background removal is required but
rembg,segment_anything, or the SAM checkpoint is missing and the user has not approved installing/downloading it. - The request is really about generation APIs, ControlNet variants, or camera outputs; route to the generation sub-skill instead of duplicating that logic here.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).