Videos and official demos
The home page does not embed a player. Community videos are outbound links. This page does not state titles beyond the labels below, and it does not state view counts. Use the written docs when you need a command you can copy.
Why these links are outbound
An embedded player would load a third-party script on the home page and would make this site look like it hosts the video. It does not. The clips below were published by other people. They are not affiliated with Convai Innovations, with NandhaKishorM, with TypeSafe, or with laya-model.com. We do not count views, we do not rank the clips, and we do not treat a label as a claim about what the picture shows beyond the words in the label.
If you want to see the model answer a ticket, use this site instead of a video. The support-triage workbench posts to the hosted API when the gateway is configured. Offline demo is labeled as such and does not load weights. The status page tells you which of those two modes this deploy is in.
Community videos
- CoderOne
A community clip. This page does not restate a title beyond the label, and it does not state a view count. The creator is not affiliated with Convai Innovations or with this site.
- RUNTIME install
A community clip filed here under an install label. The written install path for this site is the get-started page: Python 3.10 or newer, a virtual environment, and pip install laya. The video is not a substitute for that page, and it is not an official Convai tutorial.
- TerraNet fine-tune
A community clip filed here under a fine-tune label. The written fine-tune path is the upstream notebook linked below. 0.766 is the fine-tuned laya-typed-decisions checkpoint on the model card, not the score of a fresh pip install.
Open the links in a new context if you want the picture. Nothing on this page plays inside the page. If a link stops resolving, the written pages below are the source this site actually maintains.
Official written demos
The project's own demos are documents and notebooks, not a channel on this domain. Start with the one that matches the job.
- Hugging Face model card, convaiinnovations/laya. Weights, the checkpoint names, and the published scores. This site does not re-host the files. The card is also where the 0.766 fine-tune is distinguished from the 0.362 and 0.352 bases.
- Colab notebook linked from the README. A browser notebook for people who do not want to install locally first. It is upstream's notebook, not a notebook executed by laya-model.com.
- Dev.to write-up by the maintainer. A narrative about the non-autoregressive decision model. It is not the API reference for this website.
- README for install,
Router().predict, and the hardware notes this site repeats on get started. BENCHMARKS.md for the numbers. Comparison tables that this site copies with sources stay on benchmarks. - Fine-tuning notebook for the 0.766 checkpoint path. The fine-tune guide is the short version: what the notebook trains, and what a plain install does not include.
- Upstream docs and Docker doc. Serving with Docker on your own machine is summarized in serve with Docker. That path is local. It is not the prepaid API on this domain.
A practical order
If you are deciding whether the hosted API fits, run one ticket in the workbench and read the decide API guide. That guide covers state, the three question types, the model aliases this gateway actually resolves, and the status codes. Pricing and keys are in the credits and keys guide. A full support-desk example, including the free confidence gate, is in the support-triage guide.
If you are deciding whether to run weights yourself, read get started, then the README, then the Docker doc if you want a container. The package and the weights stay under the Apache-2.0 license. This website's terms cover the hosted API only. The two are easy to mix up because both answer choice, score, and noul. The hosted API bills input tokens. A local Router() does not bill this site.
Community clips can be a fast look at someone else's screen. They are not the contract. When a clip and a written page disagree, follow the written page and the upstream file it cites. Questions about a broken link on this page go to the contact page. Questions about the upstream notebooks go to the upstream repository.