SuperAPI AI Directory
Partner with us中文AI newsFavorites

AI developer platforms

3,641 results · Page 100 of 102

Tina

AI developer platforms

Tina is a research project that explores how to efficiently inject reasoning abilities into small language models during reinforcement learning using low-rank adaptation (LoRA). The project paper is titled "Tina: Tiny Reasoning Models via LoRA" and has been accepted by ICLR 2026. Its core question is: at how low a cost can reasoning abilities be efficiently imparted to language models through reinforcement learning? Experiments show that, on the same base model, the Tina models achieve performance comparable to or even better than current state-of-the-art baseline models at a cost far lower than full-parameter training. The best Tina model achieved a performance improvement of more than 20% on AIME24, with a Pass@1 accuracy of 43.33%. Reproducing the best checkpoint requires only $9, and reproducing all experiments from scratch requires only $526. The project is developed based on HuggingFace's open-r1 project, uses the GRPO algorithm for LoRA reinforcement learning training, and provides training and evaluation scripts. The code repository includes configuration files, experiment recipes, and dataset constants, among other things, for researchers to reference and reproduce.

Read more

Token中转站

AI developer platforms

DataEyesAI is a unified AI platform that provides data and model cloud services (AI MaaS platform). Its core positioning is "one API platform to access any AI model," offering access to multiple large language models through a unified interface, focusing on running better AI models at lower cost, and providing a full-process product suite that turns websites into data usable by LLMs. The platform emphasizes both price and reliability, claiming low cost and high reliability, can be used with mainstream IDEs and CLIs, provides the DataEyesAI Coding Plan, and through proprietary business data, task adaptation, and inference optimization, transfers advanced foundation model capabilities to lightweight domain models, reducing computing power requirements, inference costs, and deployment complexity while maintaining performance in target scenarios. The official website also mentions application scenarios such as generating mini-dramas through the DataEyesAI API.

Read more