chronos-forecasting
Use Chronos Forecasting for pretrained time-series forecasting, Chronos-2 covariates, data validation, fine-tuning, evaluation, and deployment planning.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- chronos-forecasting
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use Chronos Forecasting for pretrained time-series forecasting, Chronos-2 covariates, data validation, fine-tuning, evaluation, and deployment planning.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/installation-and-environment.md3.3 KB
- 参考文档references/model-overview.md3.3 KB
- 参考文档references/repo-provenance.md2.0 KB
- 参考文档references/repo-routing-metadata.json345 B
- 参考文档references/troubleshooting.md3.7 KB
- 脚本scripts/chronos_api_smoke.py4.7 KB
03 SKILL.md 正文
Chronos Forecasting repo skill
Use this repo skill when a task involves the chronos-forecasting package, Chronos-2, Chronos-Bolt, original Chronos/T5 time-series forecasting models, probabilistic forecast quantiles, pandas long-format time-series data, covariates, predict_df, predict_fev, fine-tuning, benchmark evaluation, or deployment planning.
Install and import check
Base install:
pip install chronos-forecasting
Minimal import check:
import chronos
from chronos import BaseChronosPipeline, Chronos2Pipeline, ChronosBoltPipeline, ChronosPipeline
print(chronos.__version__)
Read references/installation-and-environment.md before choosing optional extras, GPU wheels, cloud dependencies, or benchmark/training dependencies. Run scripts/chronos_api_smoke.py for a safe import/signature/backend smoke that does not download models by default.
Choose the route
| User task | Read |
|---|---|
Chronos-2 zero-shot forecasting, multivariate/covariate prediction, Chronos2Pipeline, predict, predict_quantiles, predict_df, embeddings, long horizons, or model loading |
sub-skills/chronos-2-forecasting/ |
| DataFrame schema repair, timestamp frequency, future covariate alignment, list-of-dicts validation, preprocessing helpers, or target leakage checks | sub-skills/data-formats-and-validation/ |
| Chronos-Bolt or original Chronos/T5 models, direct quantile vs sample forecasts, univariate tensor/list inputs, or family selection among older model IDs | sub-skills/chronos-bolt-and-original/ |
| Chronos-2 fine-tuning/LoRA, original training configs, KernelSynth, fev evaluation, aggregate relative scores, SageMaker/cloud deployment, or side-effecting benchmark/training plans | sub-skills/training-evaluation-deployment/ |
Shared references:
- Model family comparison and public model IDs: references/model-overview.md
- Installation, optional dependencies, CPU/GPU policy, and safe helper list: references/installation-and-environment.md
- Cross-cutting failures for imports, loading, optional extras, backend mismatch, and side-effecting workflow gates: references/troubleshooting.md
- Source snapshot and refresh baseline: references/repo-provenance.md
Operating guardrails
- Prefer
BaseChronosPipeline.from_pretrained(...)when the model anchor may be any Chronos family, then inspecttype(pipeline).__name__before calling family-specific APIs. - Do not trigger Hugging Face downloads, S3 downloads, dataset downloads, SageMaker endpoint creation, model training, or hub pushes unless the user explicitly asks and supplies the needed model/data/credential/budget context.
- Keep
prediction_length,quantile_levels,batch_size,context_length,device_map, and dtype explicit in reproducible snippets. - Validate pandas schemas before disabling
validate_inputsor accepting afuture_dfwith known-future covariates. - Do not claim GPU, cloud, or benchmark verification from a CPU import smoke. Record those as optional/unverified unless actually executed.
- When using this skill for a different checkout, read references/repo-provenance.md; refresh if the commit, package version, public APIs, or evidence paths changed.
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).