pinned
Running
2
Turkish SLM Leaderboard
🏆
Explore Turkish SLM benchmark rankings
Eren Ata PRO
ErenAta00
AI & ML interests
LLMOps
Recent Activity
liked a model 1 day ago
AlicanKiraz0/Mizan-27B-Turkish-Legal-Decision-Model posted an update 1 day ago
The Maverick XR Agent Dataset is now public.
It is the data behind Maverick 1.0 and 1.1, our offline tool-calling models for Unity XR scenes: the conversations the models were trained on and the test commands they were measured with. Everything is synthetic. A small simulator of Unity-style houses builds each scene and the reference actions, and the results in every conversation come from that simulator.
What is inside:
- 24,761 training conversations for Maverick 1.1 and 20,091 for Maverick 1.0
- The 33 test cells we sealed before training (3,585 commands) and 29 development sets
- Corrections and undo, gaze and pointing, objects in other rooms, short plans, and requests the tools cannot fulfil
- The tool lists, system prompts and the object catalogue
Two things we want to be clear about. The test cells are no longer sealed, so every evaluation row carries a canary string. Please do not train on them. And the card lists the limits plainly: one simulator, mostly templated wording, English only.
License: CC BY-NC 4.0. You can use it for non-commercial work, including training and evaluating models, and you can adapt it. Credit and citation are required. The card has the credit line and the BibTeX entry.
Dataset: https://huggingface.co/datasets/ErenAta00/Maverick-XR-Agent-Dataset
Also on Kaggle, with a starter notebook: https://www.kaggle.com/datasets/erenata/maverick-xr-agent-dataset
Models: https://huggingface.co/ErenAta00/Maverick-4B-Unity-XR-Agent-1.1
Made at the Extended Reality Laboratory (XRLab), Manisa Celal Bayar University: https://huggingface.co/ExtendedRealityLabMCBU updated a dataset 1 day ago
ErenAta00/Maverick-XR-Agent-DatasetOrganizations
