Qwen3-Coder-30B-A3B-Instruct
9 months ago
- #AI
- #Qwen3
- #Coding
- Qwen3-Coder-30B-A3B-Instruct is introduced with significant performance in Agentic Coding and Browser-Use tasks.
- Features long-context capabilities with native support for 256K tokens, extendable to 1M tokens using Yarn.
- Model specifications include 30.5B total parameters, 3.3B activated, 48 layers, and 32 attention heads.
- Supports non-thinking mode and does not generate <think></think> blocks.
- Quickstart guide provided for using the model with transformers, including a code snippet for content generation.
- Agentic Coding capabilities demonstrated with tool calling examples.
- Recommended sampling parameters for optimal performance: temperature=0.7, top_p=0.8, top_k=20, repetition_penalty=1.05.
- Citation provided for referencing the Qwen3 Technical Report.