SuperAPI AI Directory
Partner with us中文AI newsFavorites

AI developer platforms

3,661 results · Page 47 of 102

voice-lab

AI developer platforms

Voice Lab is a testing and evaluation framework for voice agents, open-sourced by saharmor on GitHub. It is used for systematic testing and performance evaluation across language models, prompts, and agent personas, helping developers make data-driven decisions when switching models, optimizing prompts, or adjusting persona settings. Although optimized for voice agents, it is also suitable for any LLM-based agent evaluation scenario. Currently, it only supports the text portion of voice agents, that is, testing the underlying language model and prompts.

Read more

voltagent

AI developer platforms

VoltAgent is an end-to-end AI agent engineering platform consisting of two parts: an open-source TypeScript AI agent framework, and the VoltOps Console cloud/self-hosted console. The framework provides capabilities such as memory, RAG, guardrails, tools, MCP, voice, and workflows, and supports multi-agent collaboration and integration with any AI provider; VoltOps provides observability, automation, deployment, evaluation, prompt management, and other features. Developers can build agents with full code control and gain production-grade visibility and operational capabilities.

Read more

Yuan2.0-M32

AI developer platforms

Yuan2.0-M32 is a large language model using a Mixture-of-Experts (MoE) architecture, with a total parameter count of 40B, activating 2 experts at a time and 3.7B activated parameters. It proposes and adopts a new routing network, Attention Router, to improve the efficiency and accuracy of expert selection, achieving a 3.8% accuracy improvement over models using classic routing networks. The model was trained from scratch on 2000B tokens, and its training compute is only 9.25% of that of a dense model of comparable parameter scale. It is competitive in multiple specialized domains such as coding and mathematics, reaching 55.9% and 95.8% accuracy on the MATH and ARC-Challenge benchmarks respectively, surpassing Llama3-70B. Its forward computation per token is 7.4 GFLOPS, only 1/19 of Llama3-70B. The sequence length supports 16K. The related code and model have been open-sourced on GitHub and released on ModelScope and Hugging Face.

Read more

YuLan-Mini

AI developer platforms

YuLan-Mini is an open-source lightweight language model developed and maintained by the AI Box team at Renmin University of China, with 2.4 billion parameters (2.4B, including the embedding layer size). The model was pretrained using only 1.08T tokens. According to official statements, its performance is comparable to industry-leading models trained with significantly more data, especially in the fields of mathematics and code. The project has open-sourced related pretraining resources and a post-training technical report to promote reproducibility. YuLan-Mini belongs to the YuLan model family, which also includes models of larger scale and those using different training strategies. The model provides downloadable resources such as the base version YuLan-Mini, intermediate checkpoints, and the instruction-tuned version YuLan-Mini-Instruct. The pretraining method improves training efficiency through three key innovations: a carefully designed data pipeline (combining data cleaning and data scheduling strategies), a system optimization method that can effectively mitigate training instability, and an effective annealing method that integrates targeted data selection and long-context training. The project has also open-sourced resources such as the data cleaning and synthesis pipeline, classifiers, and W&B logs, which can be used for scenarios such as pretraining your own LLM, learning rate annealing, fine-tuning the instruction version, research on training dynamics, and synthesizing your own data.

Read more

ZenML

AI developer platforms

ZenML is a unified infrastructure layer for production AI, comprising two open-source projects: ZenML and Kitaru. ZenML can orchestrate pipelines and agents on your existing infrastructure; Kitaru replays these pipelines and agents using real production data, helping to evaluate improvements, regressions, and costs before changes go live. ZenML supports writing pipelines and agents in Python and runs them on your existing orchestrators and cloud platforms, allowing migration from notebooks to Kubernetes without rewriting. Both ZenML and Kitaru are licensed under Apache 2.0, supporting self-hosting, source code inspection, and keeping data in place. The platform is enterprise-grade, has SOC 2 Type II and ISO 27001 certifications, and runs on your own infrastructure.

Read more