model-reference
Explains PyTorch-VAE model constructors, registry lookups, latent-shape conventions, and synthetic forward/loss/sample checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- model-reference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Explains PyTorch-VAE model constructors, registry lookups, latent-shape conventions, and synthetic forward/loss/sample checks.
- disable-model-invocation
- true
02 资源文件 2
- 参考文档references/api-reference.md6.1 KB
- 参考文档references/troubleshooting.md2.4 KB
03 SKILL.md 正文
Model Reference
Use this sub-skill when the user wants to choose a model, inspect constructor arguments, compare variants, or run a tiny synthetic smoke on one architecture. It covers the shared registry, special kwargs, forward/loss outputs, and sample/generate caveats. The bundled command examples assume the generated skill directory is the current working directory.
Read first
references/api-reference.mdfor constructor signatures and model-specific notes.references/troubleshooting.mdfor labels, sample availability, and pretrained-feature pitfalls.scripts/model_smoke.pyfor the bundled synthetic forward/loss/sampling helper.../../references/model-overview.mdwhen you only need a quick family map.
Include here
- Looking up a model name in
models.vae_models. - Understanding constructor kwargs such as
latent_dim,num_samples,categorical_dim,embedding_dim,latent1_dim, andlatent2_dim. - Checking how a model's
forward()result maps toloss_function()inputs. - Knowing which models expect labels, multiple optimizers, or sample/generate flags.
- Running a tiny synthetic smoke against a selected model.
Exclude or route elsewhere
- Experiment orchestration,
data_params, and checkpoint/logging setup ->sub-skills/training/SKILL.md. - Full training jobs -> training.
- Repo provenance and staleness ->
references/repo-provenance.md.
Typical triggers
- "what constructor does BetaTCVAE take"
- "how do I call sample on ConditionalVAE"
- "what kwargs does VQVAE need"
- "why does forward return extra tensors"
- "which models have generate()"
- "run a quick smoke for DFCVAE"
Workflow
- Read the API reference to identify the exact class and kwargs.
- Use the model smoke script with a small batch and the repo root you want to inspect.
- If a model expects labels or a special optimizer index, follow the model-specific notes in the API reference before guessing.
- If the model smoke fails because the model needs CUDA, network downloads, or a missing optional dependency, check the troubleshooting page before retrying.
Hand off to training when needed
If the user switches from "how do I instantiate this model" to "how do I run the experiment", route to the training sub-skill instead.
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.