foundation-models-and-customization
Navigate Time-Series-Library model files, optional Mamba and large time-series model dependencies, augmentation flags, and custom model or script additions.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- foundation-models-and-customization
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Navigate Time-Series-Library model files, optional Mamba and large time-series model dependencies, augmentation flags, and custom model or script additions.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/customization.md3.0 KB
- 参考文档references/model-catalog-and-dependencies.md3.2 KB
- 参考文档references/troubleshooting.md2.8 KB
03 SKILL.md 正文
TSLib Models and Customization
Use this sub-skill when the task is about choosing a TSLib model, diagnosing optional model dependencies, using zero-shot foundation models or Mamba variants, adding a new models/<Name>.py, adapting benchmark scripts, or using the built-in augmentation flags.
Route Here
- Explain how
Exp_Basicdiscovers model files and how--model <Name>maps tomodels/<Name>.py. - Choose a core smoke-test model versus an optional dependency-heavy model.
- Diagnose missing imports for
mamba_ssm,chronos,timesfm,uni2ts,tirex, ortransformers. - Add or adapt a custom TSLib model file and create matching benchmark/smoke commands.
- Use augmentation flags such as
--jitter,--scaling,--permutation,--timewarp,--wdba, or DTW-guided variants. - Follow contribution expectations for new published-paper model additions.
Reroute
- Forecast command construction, TimeXer recipes, or zero-shot task commands: use
../forecasting/SKILL.md. - Data layouts,
run.pyflag basics, GPU flags, and output folders: use../data-and-cli/SKILL.md. - Imputation/anomaly/classification task-specific recipes: use
../imputation-anomaly-classification/SKILL.md.
Start Fast
Inventory safe core and optional model imports before recommending a model:
python scripts/inspect_tslib_models.py --repo-root . --models DLinear TimesNet TimeXer PatchTST --optional-models
For smoke tests, prefer DLinear, TimesNet, TimeXer, PatchTST, or Transformer before Mamba/LTSM files. Optional model import failures should be treated as routing information, not as a broken core TSLib install.
Custom Model Contract
A model file normally lives at models/<Name>.py and exposes either Model or a class named <Name>. Most task models implement:
class Model(nn.Module):
def __init__(self, configs): ...
def forward(self, x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None): ...
Exp_Basic scans models/ dynamically at runtime, so adding a Python file is enough for discovery if imports and the class contract are valid.
References and Helpers
references/model-catalog-and-dependencies.mdmaps model families to dependency and backend surfaces.references/customization.mdcovers custom model/script addition and augmentation usage.references/troubleshooting.mdcovers optional dependency, model import, shape, and source-tree problems.scripts/inspect_tslib_models.pyprobes model imports and class availability without downloading data or training.../../references/model-catalog.mdis the shared catalog used by the root skill.
Avoid
- Do not install all optional model stacks just because one import failed. First confirm the requested model family.
- Do not claim zero-shot, Mamba, or remote-code model execution is verified unless that exact package, model cache, and device path has been run.
- Do not add a model file without a small CPU-compatible shape/import smoke when possible.
- Do not copy upstream benchmark shell scripts without adjusting GPU ids, data paths, and epochs for the user's environment.
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.