inference-and-rendering
Load DeTikZify models and adapters, build inference pipelines, generate TikZ, compile and rasterize documents, and troubleshoot programmatic synthesis workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- inference-and-rendering
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Load DeTikZify models and adapters, build inference pipelines, generate TikZ, compile and rasterize documents, and troubleshoot programmatic synthesis workflows.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/api-reference.md2.4 KB
- 参考文档references/troubleshooting.md1.2 KB
- 参考文档references/workflows.md1.6 KB
03 SKILL.md 正文
Inference and Rendering
Use this sub-skill when the task is about DeTikZify's Python inference surface: model loading, processor handling, adapter-backed text conditioning, DetikzifyPipeline, DetikzifyGenerator, TikzDocument, compile/rasterize/save, or MCTS-backed sampling.
Route away from this sub-skill when the main task is the browser UI, training loops, evaluation metrics, or dataset builders. Those workflows have dedicated sub-skills.
Fast Path
- Confirm the package imports and the public API surface is present:
python scripts/api_smoke.py - For a rendering-specific smoke, compile a tiny document through the bundled helper:
python scripts/tikz_smoke.py - If you need search-based synthesis or tree-search behavior, run the MCTS helper:
python scripts/mcts_smoke.py
What This Sub-Skill Owns
detikzify.model.load(...)anddetikzify.model.load_adapter(...)DetikzifyProcessorandAdapterProcessorDetikzifyPipeline.sample(...)andDetikzifyPipeline.simulate(...)DetikzifyGeneratorand the MCTS bridge used inside the pipelineTikzDocument.compile(),rasterize(),save(), and error inspection- image loading / preprocessing helpers that feed inference or rendering
- text-conditioned generation when an adapter is loaded
Common Decisions
- Use
load(model_name_or_path, ...)for image-conditioned generation with the default processor flow. - Use
load_adapter(...)or a model that already carries an adapter when the request needs text prompts. - Use
sample(...)for a single best TikZ program andsimulate(...)when the user wants search / multiple candidates. - Treat
metric="model"as the perceptual path andmetric="fast"as the compiler-diagnostics path. - Set
preprocess=Truewhen you want the image trimmed and padded to a square before generation. - Check
compile_timeoutandmcts_timeoutseparately; they control different stages.
Bundled References
- references/api-reference.md: signatures and object roles for the public inference surface.
- references/workflows.md: image-conditioned, text-conditioned, save/rasterize, and search-based workflows.
- references/troubleshooting.md: adapter, CUDA, image-source, and compile/rasterize failure modes.
Related Helpers
- ../../scripts/api_smoke.py: safe import and signature snapshot.
- ../../scripts/tikz_smoke.py: safe TeX compile/rasterize smoke.
- ../../scripts/mcts_smoke.py: safe dummy-state MCTS sanity check.
Guardrails
- A CPU import does not prove the GPU path works.
- Do not assume text prompts work unless an adapter-capable load path is active.
- Do not treat compile success as proof that the output is non-empty; check the rasterized result too.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.