Qwen3 30B-A3B
9 months ago
- #AI
- #Language Model
- #Qwen3
- Introduction of Qwen3-30B-A3B-Instruct-2507 with key enhancements in general capabilities, long-tail knowledge, alignment, and long-context understanding.
- Model features include 30.5B total parameters, 262,144 context length, and support for non-thinking mode.
- Performance benchmarks show improvements in knowledge, reasoning, coding, alignment, agent tasks, and multilingualism.
- Quickstart guide provided for using the model with Hugging Face transformers, SGLang, and vLLM.
- Agentic use recommendations with Qwen-Agent for tool calling capabilities.
- Best practices for optimal performance including sampling parameters and output length recommendations.
- Citation details for referencing the Qwen3 Technical Report.