YingLong 110M
Model directoryqcw2333/YingLong_110m110M params | CC-BY-4.0 | Upstream checkpoint
YingLong 110M is a higher-capacity checkpoint in the YingLong family, sitting above the 6M and 50M variants on the quality tier. It is positioned for production use where forecast quality justifies moderate latency rather than for the cheapest possible serving — the point in the ladder where you start trading inference speed for accuracy.
The official card surfaces both the newer YingLong paper and the earlier Timer-XL lineage, while the released code still shows bidirectional forecasting with quantile outputs rather than a causal decoder. In other words it keeps the family's non-causal transformer design: attention over the full input history feeding a multi-quantile output head, so prediction intervals come directly from the model. The official model card states the released YingLong checkpoints were pre-trained on 78B time points, the corpus behind the family's zero-shot reach.
Where to run this model
Compute availability, hardware requirements, and pricing depend on the execution provider.
Get model weights, deployment instructions, and current provider listings.
Hugging Face provider mappings checked 2026-10-06. No Inference Provider mapping was listed for this checkpoint.
Model Classification
Family
YingLong
Type
time series foundation model
Pretrained time-series model for zero-shot or few-shot forecasting workloads.
Resources
Training Data
Official model card states that the released YingLong checkpoints were pre-trained on 78B time points.
Recommended For
- • Dense probabilistic forecasting with fine-grained quantile coverage
- • Workloads that need richer distribution coverage than standard low-count quantile sets
Strengths
- • Quantile-focused output head provides unusually dense probabilistic coverage
- • Clear parameter-size ladder from 6M to 300M for cost-accuracy tradeoffs
Limitations
- • Newer family with less public benchmark coverage than the most established TSFMs
- • Dense quantile output increases per-token cost compared to point-forecast-only models
Capabilities
Tags
Specifications
- Parameters
- 110M
- Architecture
- non-causal transformer forecaster with multi-quantile output head
- Context length
- n/a
- Max context
- n/a
- Minimum history
- n/a
- Recommended history
- n/a
- Input step
- n/a
- Required target series
- 1
- Temperature
- Ignored
- Top P
- Ignored
- Max output
- n/a
- License
- CC-BY-4.0