Posts tagged with: Alibaba
Content related to Alibaba
Qwen3-ASR: Alibaba’s Open‑Source 52‑Language ASR Model
Alibaba Cloud’s latest release, Qwen3‑ASR, brings state‑of‑the‑art multilingual speech recognition to the open‑source community. Supporting 52 languages and 22 Chinese dialects, the two 1.7B/0.6B models excel on benchmarks and rival commercial APIs. The repo ships with a full inference toolkit that works with transformers or the high‑performance vLLM backend, automatic timestamping via the Qwen3‑ForcedAligner, and a ready‑to‑run Gradio demo. Whether you’re a researcher, developer, or hobbyist, this guide walks you through downloading, setting up, benchmarking, and deploying Qwen3‑ASR in Docker or directly on GPU, so you can start transcribing speech, music, and songs with ease. Key highlights: multilingual support, streaming inference, forced‑alignment, quick‑start scripts, Docker deployments, and API integration with OpenAI‑compatible endpoints.
Tongyi DeepResearch: Alibaba's Open-Source AI Agent
Explore Tongyi DeepResearch, Alibaba's groundbreaking open-source AI agent. This 30.5 billion parameter model, with an efficient 3.3 billion parameter activation per token, excels in long-horizon, deep information-seeking tasks. Demonstrating state-of-the-art performance across various agentic search benchmarks like Humanity's Last Exam and BrowserComp, Tongyi DeepResearch builds on advancements from the WebAgent project. Discover its features, including automated synthetic data generation, continual pre-training on agentic data, and robust reinforcement learning techniques. Learn how to set up and run the model for your own deep research needs, leveraging its compatibility with ReAct and Heavy inference paradigms.